2020.findings.xml

<?xml version='1.0' encoding='UTF-8'?>
<collection id="2020.findings">
  <volume id="emnlp" ingest-date="2020-11-12">
    <meta>
      <booktitle>Findings of the Association for Computational Linguistics: EMNLP 2020</booktitle>
      <editor><first>Trevor</first><last>Cohn</last></editor>
      <editor><first>Yulan</first><last>He</last></editor>
      <editor id="yang-liu-icsi"><first>Yang</first><last>Liu</last></editor>
      <publisher>Association for Computational Linguistics</publisher>
      <address>Online</address>
      <month>November</month>
      <year>2020</year>
    </meta>
    <frontmatter>
      <url hash="3704669c">2020.findings-emnlp.0</url>
    </frontmatter>
    <paper id="1">
      <title>Fully Quantized Transformer for Machine Translation</title>
      <author><first>Gabriele</first><last>Prato</last></author>
      <author><first>Ella</first><last>Charlaix</last></author>
      <author><first>Mehdi</first><last>Rezagholizadeh</last></author>
      <pages>1–14</pages>
      <abstract>State-of-the-art neural machine translation methods employ massive amounts of parameters. Drastically reducing computational costs of such methods without affecting performance has been up to this point unsuccessful. To this end, we propose FullyQT: an all-inclusive quantization strategy for the Transformer. To the best of our knowledge, we are the first to show that it is possible to avoid any loss in translation quality with a fully quantized Transformer. Indeed, compared to full-precision, our 8-bit models score greater or equal BLEU on most tasks. Comparing ourselves to all previously proposed methods, we achieve state-of-the-art quantization results.</abstract>
      <url hash="71254eb4">2020.findings-emnlp.1</url>
      <doi>10.18653/v1/2020.findings-emnlp.1</doi>
    </paper>
    <paper id="2">
      <title>Summarizing <fixed-case>C</fixed-case>hinese Medical Answer with Graph Convolution Networks and Question-focused Dual Attention</title>
      <author><first>Ningyu</first><last>Zhang</last></author>
      <author><first>Shumin</first><last>Deng</last></author>
      <author><first>Juan</first><last>Li</last></author>
      <author><first>Xi</first><last>Chen</last></author>
      <author><first>Wei</first><last>Zhang</last></author>
      <author><first>Huajun</first><last>Chen</last></author>
      <pages>15–24</pages>
      <abstract>Online search engines are a popular source of medical information for users, where users can enter questions and obtain relevant answers. It is desirable to generate answer summaries for online search engines, particularly summaries that can reveal direct answers to questions. Moreover, answer summaries are expected to reveal the most relevant information in response to questions; hence, the summaries should be generated with a focus on the question, which is a challenging topic-focused summarization task. In this paper, we propose an approach that utilizes graph convolution networks and question-focused dual attention for Chinese medical answer summarization. We first organize the original long answer text into a medical concept graph with graph convolution networks to better understand the internal structure of the text and the correlation between medical concepts. Then, we introduce a question-focused dual attention mechanism to generate summaries relevant to questions. Experimental results demonstrate that the proposed model can generate more coherent and informative summaries compared with baseline models.</abstract>
      <url hash="78e38e4d">2020.findings-emnlp.2</url>
      <doi>10.18653/v1/2020.findings-emnlp.2</doi>
    </paper>
    <paper id="3">
      <title>Stay Hungry, Stay Focused: Generating Informative and Specific Questions in Information-Seeking Conversations</title>
      <author><first>Peng</first><last>Qi</last></author>
      <author><first>Yuhao</first><last>Zhang</last></author>
      <author><first>Christopher D.</first><last>Manning</last></author>
      <pages>25–40</pages>
      <abstract>We investigate the problem of generating informative questions in information-asymmetric conversations. Unlike previous work on question generation which largely assumes knowledge of what the answer might be, we are interested in the scenario where the questioner is not given the context from which answers are drawn, but must reason pragmatically about how to acquire new information, given the shared conversation history. We identify two core challenges: (1) formally defining the informativeness of potential questions, and (2) exploring the prohibitively large space of potential questions to find the good candidates. To generate pragmatic questions, we use reinforcement learning to optimize an informativeness metric we propose, combined with a reward function designed to promote more specific questions. We demonstrate that the resulting pragmatic questioner substantially improves the informativeness and specificity of questions generated over a baseline model, as evaluated by our metrics as well as humans.</abstract>
      <url hash="ce6352f3">2020.findings-emnlp.3</url>
      <doi>10.18653/v1/2020.findings-emnlp.3</doi>
    </paper>
    <paper id="4">
      <title>Adapting <fixed-case>BERT</fixed-case> for Word Sense Disambiguation with Gloss Selection Objective and Example Sentences</title>
      <author><first>Boon Peng</first><last>Yap</last></author>
      <author><first>Andrew</first><last>Koh</last></author>
      <author><first>Eng Siong</first><last>Chng</last></author>
      <pages>41–46</pages>
      <abstract>Domain adaptation or transfer learning using pre-trained language models such as BERT has proven to be an effective approach for many natural language processing tasks. In this work, we propose to formulate word sense disambiguation as a relevance ranking task, and fine-tune BERT on sequence-pair ranking task to select the most probable sense definition given a context sentence and a list of candidate sense definitions. We also introduce a data augmentation technique for WSD using existing example sentences from WordNet. Using the proposed training objective and data augmentation technique, our models are able to achieve state-of-the-art results on the English all-words benchmark datasets.</abstract>
      <url hash="7b542509">2020.findings-emnlp.4</url>
      <doi>10.18653/v1/2020.findings-emnlp.4</doi>
    </paper>
    <paper id="5">
      <title>Adversarial Text Generation via Sequence Contrast Discrimination</title>
      <author><first>Ke</first><last>Wang</last></author>
      <author><first>Xiaojun</first><last>Wan</last></author>
      <pages>47–53</pages>
      <abstract>In this paper, we propose a sequence contrast loss driven text generation framework, which learns the difference between real texts and generated texts and uses that difference. Specifically, our discriminator contains a discriminative sequence generator instead of a binary classifier, and measures the ‘relative realism’ of generated texts against real texts by making use of them simultaneously. Moreover, our generator uses discriminative sequences to directly improve itself, which not only replaces the gradient propagation process from the discriminator to the generator, but also avoids the time-consuming sampling process of estimating rewards in some previous methods. We conduct extensive experiments with various metrics, substantiating that our framework brings improvements in terms of training stability and the quality of generated texts.</abstract>
      <url hash="1b9604ce">2020.findings-emnlp.5</url>
      <attachment type="OptionalSupplementaryMaterial" hash="a550a176">2020.findings-emnlp.5.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.5</doi>
    </paper>
    <paper id="6">
      <title><fixed-case>GRACE</fixed-case>: Gradient Harmonized and Cascaded Labeling for Aspect-based Sentiment Analysis</title>
      <author><first>Huaishao</first><last>Luo</last></author>
      <author><first>Lei</first><last>Ji</last></author>
      <author><first>Tianrui</first><last>Li</last></author>
      <author><first>Daxin</first><last>Jiang</last></author>
      <author><first>Nan</first><last>Duan</last></author>
      <pages>54–64</pages>
      <abstract>In this paper, we focus on the imbalance issue, which is rarely studied in aspect term extraction and aspect sentiment classification when regarding them as sequence labeling tasks. Besides, previous works usually ignore the interaction between aspect terms when labeling polarities. We propose a GRadient hArmonized and CascadEd labeling model (GRACE) to solve these problems. Specifically, a cascaded labeling module is developed to enhance the interchange between aspect terms and improve the attention of sentiment tokens when labeling sentiment polarities. The polarities sequence is designed to depend on the generated aspect terms labels. To alleviate the imbalance issue, we extend the gradient harmonized mechanism used in object detection to the aspect-based sentiment analysis by adjusting the weight of each label dynamically. The proposed GRACE adopts a post-pretraining BERT as its backbone. Experimental results demonstrate that the proposed model achieves consistency improvement on multiple benchmark datasets and generates state-of-the-art results.</abstract>
      <url hash="cd3e668d">2020.findings-emnlp.6</url>
      <doi>10.18653/v1/2020.findings-emnlp.6</doi>
    </paper>
    <paper id="7">
      <title>Reducing Sentiment Bias in Language Models via Counterfactual Evaluation</title>
      <author><first>Po-Sen</first><last>Huang</last></author>
      <author><first>Huan</first><last>Zhang</last></author>
      <author><first>Ray</first><last>Jiang</last></author>
      <author><first>Robert</first><last>Stanforth</last></author>
      <author><first>Johannes</first><last>Welbl</last></author>
      <author><first>Jack</first><last>Rae</last></author>
      <author><first>Vishal</first><last>Maini</last></author>
      <author><first>Dani</first><last>Yogatama</last></author>
      <author><first>Pushmeet</first><last>Kohli</last></author>
      <pages>65–83</pages>
      <abstract>Advances in language modeling architectures and the availability of large text corpora have driven progress in automatic text generation. While this results in models capable of generating coherent texts, it also prompts models to internalize social biases present in the training corpus. This paper aims to quantify and reduce a particular type of bias exhibited by language models: bias in the sentiment of generated text. Given a conditioning context (e.g., a writing prompt) and a language model, we analyze if (and how) the sentiment of the generated text is affected by changes in values of sensitive attributes (e.g., country names, occupations, genders) in the conditioning context using a form of counterfactual evaluation. We quantify sentiment bias by adopting individual and group fairness metrics from the fair machine learning literature, and demonstrate that large-scale models trained on two different corpora (news articles, and Wikipedia) exhibit considerable levels of bias. We then propose embedding and sentiment prediction-derived regularization on the language model’s latent representations. The regularizations improve fairness metrics while retaining comparable levels of perplexity and semantic similarity.</abstract>
      <url hash="b7d1741c">2020.findings-emnlp.7</url>
      <doi>10.18653/v1/2020.findings-emnlp.7</doi>
    </paper>
    <paper id="8">
      <title>Improving Text Understanding via Deep Syntax-Semantics Communication</title>
      <author><first>Hao</first><last>Fei</last></author>
      <author><first>Yafeng</first><last>Ren</last></author>
      <author><first>Donghong</first><last>Ji</last></author>
      <pages>84–93</pages>
      <abstract>Recent studies show that integrating syntactic tree models with sequential semantic models can bring improved task performance, while these methods mostly employ shallow integration of syntax and semantics. In this paper, we propose a deep neural communication model between syntax and semantics to improve the performance of text understanding. Local communication is performed between syntactic tree encoder and sequential semantic encoder for mutual learning of information exchange. Global communication can further ensure comprehensive information propagation. Results on multiple syntax-dependent tasks show that our model outperforms strong baselines by a large margin. In-depth analysis indicates that our method is highly effective in composing sentence semantics.</abstract>
      <url hash="5589c3bb">2020.findings-emnlp.8</url>
      <doi>10.18653/v1/2020.findings-emnlp.8</doi>
    </paper>
    <paper id="9">
      <title><fixed-case>GRUEN</fixed-case> for Evaluating Linguistic Quality of Generated Text</title>
      <author><first>Wanzheng</first><last>Zhu</last></author>
      <author><first>Suma</first><last>Bhat</last></author>
      <pages>94–108</pages>
      <abstract>Automatic evaluation metrics are indispensable for evaluating generated text. To date, these metrics have focused almost exclusively on the content selection aspect of the system output, ignoring the linguistic quality aspect altogether. We bridge this gap by proposing GRUEN for evaluating Grammaticality, non-Redundancy, focUs, structure and coherENce of generated text. GRUEN utilizes a BERT-based model and a class of syntactic, semantic, and contextual features to examine the system output. Unlike most existing evaluation metrics which require human references as an input, GRUEN is reference-less and requires only the system output. Besides, it has the advantage of being unsupervised, deterministic, and adaptable to various tasks. Experiments on seven datasets over four language generation tasks show that the proposed metric correlates highly with human judgments.</abstract>
      <url hash="8a0d8f9a">2020.findings-emnlp.9</url>
      <doi>10.18653/v1/2020.findings-emnlp.9</doi>
    </paper>
    <paper id="10">
      <title>A Greedy Bit-flip Training Algorithm for Binarized Knowledge Graph Embeddings</title>
      <author><first>Katsuhiko</first><last>Hayashi</last></author>
      <author><first>Koki</first><last>Kishimoto</last></author>
      <author><first>Masashi</first><last>Shimbo</last></author>
      <pages>109–114</pages>
      <abstract>This paper presents a simple and effective discrete optimization method for training binarized knowledge graph embedding model B-CP. Unlike the prior work using a SGD-based method and quantization of real-valued vectors, the proposed method directly optimizes binary embedding vectors by a series of bit flipping operations. On the standard knowledge graph completion tasks, the B-CP model trained with the proposed method achieved comparable performance with that trained with SGD as well as state-of-the-art real-valued models with similar embedding dimensions.</abstract>
      <url hash="cfd35379">2020.findings-emnlp.10</url>
      <doi>10.18653/v1/2020.findings-emnlp.10</doi>
    </paper>
    <paper id="11">
      <title>Difference-aware Knowledge Selection for Knowledge-grounded Conversation Generation</title>
      <author><first>Chujie</first><last>Zheng</last></author>
      <author><first>Yunbo</first><last>Cao</last></author>
      <author><first>Daxin</first><last>Jiang</last></author>
      <author><first>Minlie</first><last>Huang</last></author>
      <pages>115–125</pages>
      <abstract>In a multi-turn knowledge-grounded dialog, the difference between the knowledge selected at different turns usually provides potential clues to knowledge selection, which has been largely neglected in previous research. In this paper, we propose a difference-aware knowledge selection method. It first computes the difference between the candidate knowledge sentences provided at the current turn and those chosen in the previous turns. Then, the differential information is fused with or disentangled from the contextual information to facilitate final knowledge selection. Automatic, human observational, and interactive evaluation shows that our method is able to select knowledge more accurately and generate more informative responses, significantly outperforming the state-of-the-art baselines.</abstract>
      <url hash="46b80b12">2020.findings-emnlp.11</url>
      <doi>10.18653/v1/2020.findings-emnlp.11</doi>
    </paper>
    <paper id="12">
      <title>An Attentive Recurrent Model for Incremental Prediction of Sentence-final Verbs</title>
      <author><first>Wenyan</first><last>Li</last></author>
      <author><first>Alvin</first><last>Grissom II</last></author>
      <author><first>Jordan</first><last>Boyd-Graber</last></author>
      <pages>126–136</pages>
      <abstract>Verb prediction is important for understanding human processing of verb-final languages, with practical applications to real-time simultaneous interpretation from verb-final to verb-medial languages. While previous approaches use classical statistical models, we introduce an attention-based neural model to incrementally predict final verbs on incomplete sentences in Japanese and German SOV sentences. To offer flexibility to the model, we further incorporate synonym awareness. Our approach both better predicts the final verbs in Japanese and German and provides more interpretable explanations of why those verbs are selected.</abstract>
      <url hash="9d2fdd69">2020.findings-emnlp.12</url>
      <attachment type="OptionalSupplementaryMaterial" hash="dda52e50">2020.findings-emnlp.12.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.12</doi>
    </paper>
    <paper id="13">
      <title>Transformer-<fixed-case>GCRF</fixed-case>: Recovering <fixed-case>C</fixed-case>hinese Dropped Pronouns with General Conditional Random Fields</title>
      <author><first>Jingxuan</first><last>Yang</last></author>
      <author><first>Kerui</first><last>Xu</last></author>
      <author><first>Jun</first><last>Xu</last></author>
      <author><first>Si</first><last>Li</last></author>
      <author><first>Sheng</first><last>Gao</last></author>
      <author><first>Jun</first><last>Guo</last></author>
      <author><first>Ji-Rong</first><last>Wen</last></author>
      <author><first>Nianwen</first><last>Xue</last></author>
      <pages>137–147</pages>
      <abstract>Pronouns are often dropped in Chinese conversations and recovering the dropped pronouns is important for NLP applications such as Machine Translation. Existing approaches usually formulate this as a sequence labeling task of predicting whether there is a dropped pronoun before each token and its type. Each utterance is considered to be a sequence and labeled independently. Although these approaches have shown promise, labeling each utterance independently ignores the dependencies between pronouns in neighboring utterances. Modeling these dependencies is critical to improving the performance of dropped pronoun recovery. In this paper, we present a novel framework that combines the strength of Transformer network with General Conditional Random Fields (GCRF) to model the dependencies between pronouns in neighboring utterances. Results on three Chinese conversation datasets show that the Transformer-GCRF model outperforms the state-of-the-art dropped pronoun recovery models. Exploratory analysis also demonstrates that the GCRF did help to capture the dependencies between pronouns in neighboring utterances, thus contributes to the performance improvements.</abstract>
      <url hash="15bfde53">2020.findings-emnlp.13</url>
      <doi>10.18653/v1/2020.findings-emnlp.13</doi>
    </paper>
    <paper id="14">
      <title>Neural Speed Reading Audited</title>
      <author><first>Anders</first><last>Søgaard</last></author>
      <pages>148–153</pages>
      <abstract>Several approaches to neural speed reading have been presented at major NLP and machine learning conferences in 2017–20; i.e., “human-inspired” recurrent network architectures that learn to “read” text faster by skipping irrelevant words, typically optimizing the joint objective of minimizing classification error rate and FLOPs used at inference time. This paper reflects on the meaningfulness of the speed reading task, showing that (a) better and faster approaches to, say, document classification, already exist, which also learn to ignore part of the input (I give an example with 7% error reduction and a 136x speed-up over the state of the art in neural speed reading); and that (b) any claims that neural speed reading is “human-inspired”, are ill-founded.</abstract>
      <url hash="29e00fa1">2020.findings-emnlp.14</url>
      <doi>10.18653/v1/2020.findings-emnlp.14</doi>
    </paper>
    <paper id="15">
      <title>Converting the Point of View of Messages Spoken to Virtual Assistants</title>
      <author><first>Gunhee</first><last>Lee</last></author>
      <author><first>Vera</first><last>Zu</last></author>
      <author><first>Sai Srujana</first><last>Buddi</last></author>
      <author><first>Dennis</first><last>Liang</last></author>
      <author><first>Purva</first><last>Kulkarni</last></author>
      <author><first>Jack</first><last>FitzGerald</last></author>
      <pages>154–163</pages>
      <abstract>Virtual Assistants can be quite literal at times. If the user says “tell Bob I love him,” most virtual assistants will extract the message “I love him” and send it to the user’s contact named Bob, rather than properly converting the message to “I love you.” We designed a system to allow virtual assistants to take a voice message from one user, convert the point of view of the message, and then deliver the result to its target user. We developed a rule-based model, which integrates a linear text classification model, part-of-speech tagging, and constituency parsing with rule-based transformation methods. We also investigated Neural Machine Translation (NMT) approaches, including LSTMs, CopyNet, and T5. We explored 5 metrics to gauge both naturalness and faithfulness automatically, and we chose to use BLEU plus METEOR for faithfulness and relative perplexity using a separately trained language model (GPT) for naturalness. Transformer-Copynet and T5 performed similarly on faithfulness metrics, with T5 achieving slight edge, a BLEU score of 63.8 and a METEOR score of 83.0. CopyNet was the most natural, with a relative perplexity of 1.59. CopyNet also has 37 times fewer parameters than T5. We have publicly released our dataset, which is composed of 46,565 crowd-sourced samples.</abstract>
      <url hash="4c526144">2020.findings-emnlp.15</url>
      <doi>10.18653/v1/2020.findings-emnlp.15</doi>
    </paper>
    <paper id="16">
      <title>Robustness to Modification with Shared Words in Paraphrase Identification</title>
      <author><first>Zhouxing</first><last>Shi</last></author>
      <author><first>Minlie</first><last>Huang</last></author>
      <pages>164–171</pages>
      <abstract>Revealing the robustness issues of natural language processing models and improving their robustness is important to their performance under difficult situations. In this paper, we study the robustness of paraphrase identification models from a new perspective – via modification with shared words, and we show that the models have significant robustness issues when facing such modifications. To modify an example consisting of a sentence pair, we either replace some words shared by both sentences or introduce new shared words. We aim to construct a valid new example such that a target model makes a wrong prediction. To find a modification solution, we use beam search constrained by heuristic rules, and we leverage a BERT masked language model for generating substitution words compatible with the context. Experiments show that the performance of the target models has a dramatic drop on the modified examples, thereby revealing the robustness issue. We also show that adversarial training can mitigate this issue.</abstract>
      <url hash="675a123c">2020.findings-emnlp.16</url>
      <doi>10.18653/v1/2020.findings-emnlp.16</doi>
    </paper>
    <paper id="17">
      <title>Few-shot Natural Language Generation for Task-Oriented Dialog</title>
      <author><first>Baolin</first><last>Peng</last></author>
      <author><first>Chenguang</first><last>Zhu</last></author>
      <author><first>Chunyuan</first><last>Li</last></author>
      <author><first>Xiujun</first><last>Li</last></author>
      <author><first>Jinchao</first><last>Li</last></author>
      <author><first>Michael</first><last>Zeng</last></author>
      <author><first>Jianfeng</first><last>Gao</last></author>
      <pages>172–182</pages>
      <abstract>As a crucial component in task-oriented dialog systems, the Natural Language Generation (NLG) module converts a dialog act represented in a semantic form into a response in natural language. The success of traditional template-based or statistical models typically relies on heavily annotated data, which is infeasible for new domains. Therefore, it is pivotal for an NLG system to generalize well with limited labelled data in real applications. To this end, we present FewshotWOZ, the first NLG benchmark to simulate the few-shot learning setting in task-oriented dialog systems. Further, we develop the SC-GPT model. It is pre-trained on a large set of annotated NLG corpus to acquire the controllable generation ability, and fine-tuned with only a few domain-specific labels to adapt to new domains. Experiments on FewshotWOZ and the large Multi-Domain-WOZ datasets show that the proposed SC-GPT significantly outperforms existing methods, measured by various automatic metrics and human evaluations.</abstract>
      <url hash="bed257d1">2020.findings-emnlp.17</url>
      <doi>10.18653/v1/2020.findings-emnlp.17</doi>
    </paper>
    <paper id="18">
      <title>Mimic and Conquer: Heterogeneous Tree Structure Distillation for Syntactic <fixed-case>NLP</fixed-case></title>
      <author><first>Hao</first><last>Fei</last></author>
      <author><first>Yafeng</first><last>Ren</last></author>
      <author><first>Donghong</first><last>Ji</last></author>
      <pages>183–193</pages>
      <abstract>Syntax has been shown useful for various NLP tasks, while existing work mostly encodes singleton syntactic tree using one hierarchical neural network. In this paper, we investigate a simple and effective method, Knowledge Distillation, to integrate heterogeneous structure knowledge into a unified sequential LSTM encoder. Experimental results on four typical syntax-dependent tasks show that our method outperforms tree encoders by effectively integrating rich heterogeneous structure syntax, meanwhile reducing error propagation, and also outperforms ensemble methods, in terms of both the efficiency and accuracy.</abstract>
      <url hash="a398c8b9">2020.findings-emnlp.18</url>
      <doi>10.18653/v1/2020.findings-emnlp.18</doi>
    </paper>
    <paper id="19">
      <title>A Hierarchical Network for Abstractive Meeting Summarization with Cross-Domain Pretraining</title>
      <author><first>Chenguang</first><last>Zhu</last></author>
      <author><first>Ruochen</first><last>Xu</last></author>
      <author><first>Michael</first><last>Zeng</last></author>
      <author><first>Xuedong</first><last>Huang</last></author>
      <pages>194–203</pages>
      <abstract>With the abundance of automatic meeting transcripts, meeting summarization is of great interest to both participants and other parties. Traditional methods of summarizing meetings depend on complex multi-step pipelines that make joint optimization intractable. Meanwhile, there are a handful of deep neural models for text summarization and dialogue systems. However, the semantic structure and styles of meeting transcripts are quite different from articles and conversations. In this paper, we propose a novel abstractive summary network that adapts to the meeting scenario. We design a hierarchical structure to accommodate long meeting transcripts and a role vector to depict the difference among speakers. Furthermore, due to the inadequacy of meeting summary data, we pretrain the model on large-scale news summary data. Empirical results show that our model outperforms previous approaches in both automatic metrics and human evaluation. For example, on ICSI dataset, the ROUGE-1 score increases from 34.66% to 46.28%.</abstract>
      <url hash="a7a97807">2020.findings-emnlp.19</url>
      <doi>10.18653/v1/2020.findings-emnlp.19</doi>
    </paper>
    <paper id="20">
      <title>Active Testing: An Unbiased Evaluation Method for Distantly Supervised Relation Extraction</title>
      <author><first>Pengshuai</first><last>Li</last></author>
      <author><first>Xinsong</first><last>Zhang</last></author>
      <author><first>Weijia</first><last>Jia</last></author>
      <author><first>Wei</first><last>Zhao</last></author>
      <pages>204–211</pages>
      <abstract>Distant supervision has been a widely used method for neural relation extraction for its convenience of automatically labeling datasets. However, existing works on distantly supervised relation extraction suffer from the low quality of test set, which leads to considerable biased performance evaluation. These biases not only result in unfair evaluations but also mislead the optimization of neural relation extraction. To mitigate this problem, we propose a novel evaluation method named active testing through utilizing both the noisy test set and a few manual annotations. Experiments on a widely used benchmark show that our proposed approach can yield approximately unbiased evaluations for distantly supervised relation extractors.</abstract>
      <url hash="1f1c39fc">2020.findings-emnlp.20</url>
      <attachment type="OptionalSupplementaryMaterial" hash="b7f14ee4">2020.findings-emnlp.20.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.20</doi>
    </paper>
    <paper id="21">
      <title>Semantic Matching for Sequence-to-Sequence Learning</title>
      <author><first>Ruiyi</first><last>Zhang</last></author>
      <author><first>Changyou</first><last>Chen</last></author>
      <author><first>Xinyuan</first><last>Zhang</last></author>
      <author><first>Ke</first><last>Bai</last></author>
      <author><first>Lawrence</first><last>Carin</last></author>
      <pages>212–222</pages>
      <abstract>In sequence-to-sequence models, classical optimal transport (OT) can be applied to semantically match generated sentences with target sentences. However, in non-parallel settings, target sentences are usually unavailable. To tackle this issue without losing the benefits of classical OT, we present a semantic matching scheme based on the Optimal Partial Transport (OPT). Specifically, our approach partially matches semantically meaningful words between source and partial target sequences. To overcome the difficulty of detecting active regions in OPT (corresponding to the words needed to be matched), we further exploit prior knowledge to perform partial matching. Extensive experiments are conducted to evaluate the proposed approach, showing consistent improvements over sequence-to-sequence tasks.</abstract>
      <url hash="c7deecb8">2020.findings-emnlp.21</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f5bcb0ab">2020.findings-emnlp.21.OptionalSupplementaryMaterial.bbl</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.21</doi>
    </paper>
    <paper id="22">
      <title>How Decoding Strategies Affect the Verifiability of Generated Text</title>
      <author><first>Luca</first><last>Massarelli</last></author>
      <author><first>Fabio</first><last>Petroni</last></author>
      <author><first>Aleksandra</first><last>Piktus</last></author>
      <author><first>Myle</first><last>Ott</last></author>
      <author><first>Tim</first><last>Rocktäschel</last></author>
      <author><first>Vassilis</first><last>Plachouras</last></author>
      <author><first>Fabrizio</first><last>Silvestri</last></author>
      <author><first>Sebastian</first><last>Riedel</last></author>
      <pages>223–235</pages>
      <abstract>Recent progress in pre-trained language models led to systems that are able to generate text of an increasingly high quality. While several works have investigated the fluency and grammatical correctness of such models, it is still unclear to which extent the generated text is consistent with factual world knowledge. Here, we go beyond fluency and also investigate the verifiability of text generated by state-of-the-art pre-trained language models. A generated sentence is verifiable if it can be corroborated or disproved by Wikipedia, and we find that the verifiability of generated text strongly depends on the decoding strategy. In particular, we discover a tradeoff between factuality (i.e., the ability of generating Wikipedia corroborated text) and repetitiveness. While decoding strategies such as top-k and nucleus sampling lead to less repetitive generations, they also produce less verifiable text. Based on these finding, we introduce a simple and effective decoding strategy which, in comparison to previously used decoding strategies, produces less repetitive and more verifiable text.</abstract>
      <url hash="6288dd54">2020.findings-emnlp.22</url>
      <doi>10.18653/v1/2020.findings-emnlp.22</doi>
    </paper>
    <paper id="23">
      <title>Minimize Exposure Bias of <fixed-case>S</fixed-case>eq2<fixed-case>S</fixed-case>eq Models in Joint Entity and Relation Extraction</title>
      <author><first>Ranran Haoran</first><last>Zhang</last></author>
      <author><first>Qianying</first><last>Liu</last></author>
      <author><first>Aysa Xuemo</first><last>Fan</last></author>
      <author><first>Heng</first><last>Ji</last></author>
      <author><first>Daojian</first><last>Zeng</last></author>
      <author><first>Fei</first><last>Cheng</last></author>
      <author><first>Daisuke</first><last>Kawahara</last></author>
      <author><first>Sadao</first><last>Kurohashi</last></author>
      <pages>236–246</pages>
      <abstract>Joint entity and relation extraction aims to extract relation triplets from plain text directly. Prior work leverages Sequence-to-Sequence (Seq2Seq) models for triplet sequence generation. However, Seq2Seq enforces an unnecessary order on the unordered triplets and involves a large decoding length associated with error accumulation. These methods introduce exposure bias, which may cause the models overfit to the frequent label combination, thus limiting the generalization ability. We propose a novel Sequence-to-Unordered-Multi-Tree (Seq2UMTree) model to minimize the effects of exposure bias by limiting the decoding length to three within a triplet and removing the order among triplets. We evaluate our model on two datasets, DuIE and NYT, and systematically study how exposure bias alters the performance of Seq2Seq models. Experiments show that the state-of-the-art Seq2Seq model overfits to both datasets while Seq2UMTree shows significantly better generalization. Our code is available at <url>https://github.com/WindChimeRan/OpenJERE</url>.</abstract>
      <url hash="4e78d163">2020.findings-emnlp.23</url>
      <doi>10.18653/v1/2020.findings-emnlp.23</doi>
    </paper>
    <paper id="24">
      <title>Gradient-based Analysis of <fixed-case>NLP</fixed-case> Models is Manipulable</title>
      <author><first>Junlin</first><last>Wang</last></author>
      <author><first>Jens</first><last>Tuyls</last></author>
      <author><first>Eric</first><last>Wallace</last></author>
      <author><first>Sameer</first><last>Singh</last></author>
      <pages>247–258</pages>
      <abstract>Gradient-based analysis methods, such as saliency map visualizations and adversarial input perturbations, have found widespread use in interpreting neural NLP models due to their simplicity, flexibility, and most importantly, the fact that they directly reflect the model internals. In this paper, however, we demonstrate that the gradients of a model are easily manipulable, and thus bring into question the reliability of gradient-based analyses. In particular, we merge the layers of a target model with a Facade Model that overwhelms the gradients without affecting the predictions. This Facade Model can be trained to have gradients that are misleading and irrelevant to the task, such as focusing only on the stop words in the input. On a variety of NLP tasks (sentiment analysis, NLI, and QA), we show that the merged model effectively fools different analysis tools: saliency maps differ significantly from the original model’s, input reduction keeps more irrelevant input tokens, and adversarial perturbations identify unimportant tokens as being highly important.</abstract>
      <url hash="0ed856ab">2020.findings-emnlp.24</url>
      <attachment type="OptionalSupplementaryMaterial" hash="590940a8">2020.findings-emnlp.24.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.24</doi>
    </paper>
    <paper id="25">
      <title>Pretrain-<fixed-case>KGE</fixed-case>: Learning Knowledge Representation from Pretrained Language Models</title>
      <author><first>Zhiyuan</first><last>Zhang</last></author>
      <author><first>Xiaoqian</first><last>Liu</last></author>
      <author><first>Yi</first><last>Zhang</last></author>
      <author><first>Qi</first><last>Su</last></author>
      <author><first>Xu</first><last>Sun</last></author>
      <author><first>Bin</first><last>He</last></author>
      <pages>259–266</pages>
      <abstract>Conventional knowledge graph embedding (KGE) often suffers from limited knowledge representation, leading to performance degradation especially on the low-resource problem. To remedy this, we propose to enrich knowledge representation via pretrained language models by leveraging world knowledge from pretrained models. Specifically, we present a universal training framework named <i>Pretrain-KGE</i> consisting of three phases: semantic-based fine-tuning phase, knowledge extracting phase and KGE training phase. Extensive experiments show that our proposed Pretrain-KGE can improve results over KGE models, especially on solving the low-resource problem.</abstract>
      <url hash="72c0a869">2020.findings-emnlp.25</url>
      <doi>10.18653/v1/2020.findings-emnlp.25</doi>
    </paper>
    <paper id="26">
      <title>A Self-Refinement Strategy for Noise Reduction in Grammatical Error Correction</title>
      <author><first>Masato</first><last>Mita</last></author>
      <author><first>Shun</first><last>Kiyono</last></author>
      <author><first>Masahiro</first><last>Kaneko</last></author>
      <author><first>Jun</first><last>Suzuki</last></author>
      <author><first>Kentaro</first><last>Inui</last></author>
      <pages>267–280</pages>
      <abstract>Existing approaches for grammatical error correction (GEC) largely rely on supervised learning with manually created GEC datasets. However, there has been little focus on verifying and ensuring the quality of the datasets, and on how lower-quality data might affect GEC performance. We indeed found that there is a non-negligible amount of “noise” where errors were inappropriately edited or left uncorrected. To address this, we designed a self-refinement method where the key idea is to denoise these datasets by leveraging the prediction consistency of existing models, and outperformed strong denoising baseline methods. We further applied task-specific techniques and achieved state-of-the-art performance on the CoNLL-2014, JFLEG, and BEA-2019 benchmarks. We then analyzed the effect of the proposed denoising method, and found that our approach leads to improved coverage of corrections and facilitated fluency edits which are reflected in higher recall and overall performance.</abstract>
      <url hash="4f6adf45">2020.findings-emnlp.26</url>
      <doi>10.18653/v1/2020.findings-emnlp.26</doi>
    </paper>
    <paper id="27">
      <title>Understanding tables with intermediate pre-training</title>
      <author><first>Julian</first><last>Eisenschlos</last></author>
      <author><first>Syrine</first><last>Krichene</last></author>
      <author><first>Thomas</first><last>Müller</last></author>
      <pages>281–296</pages>
      <abstract>Table entailment, the binary classification task of finding if a sentence is supported or refuted by the content of a table, requires parsing language and table structure as well as numerical and discrete reasoning. While there is extensive work on textual entailment, table entailment is less well studied. We adapt TAPAS (Herzig et al., 2020), a table-based BERT model, to recognize entailment. Motivated by the benefits of data augmentation, we create a balanced dataset of millions of automatically created training examples which are learned in an intermediate step prior to fine-tuning. This new data is not only useful for table entailment, but also for SQA (Iyyer et al., 2017), a sequential table QA task. To be able to use long examples as input of BERT models, we evaluate table pruning techniques as a pre-processing step to drastically improve the training and prediction efficiency at a moderate drop in accuracy. The different methods set the new state-of-the-art on the TabFact (Chen et al., 2020) and SQA datasets.</abstract>
      <url hash="7e9b03bf">2020.findings-emnlp.27</url>
      <attachment type="OptionalSupplementaryMaterial" hash="3ab55da8">2020.findings-emnlp.27.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.27</doi>
    </paper>
    <paper id="28">
      <title>Enhance Robustness of Sequence Labelling with Masked Adversarial Training</title>
      <author><first>Luoxin</first><last>Chen</last></author>
      <author><first>Xinyue</first><last>Liu</last></author>
      <author><first>Weitong</first><last>Ruan</last></author>
      <author><first>Jianhua</first><last>Lu</last></author>
      <pages>297–302</pages>
      <abstract>Adversarial training (AT) has shown strong regularization effects on deep learning algorithms by introducing small input perturbations to improve model robustness. In language tasks, adversarial training brings word-level robustness by adding input noise, which is beneficial for text classification. However, it lacks sufficient contextual information enhancement and thus is less useful for sequence labelling tasks such as chunking and named entity recognition (NER). To address this limitation, we propose masked adversarial training (MAT) to improve robustness from contextual information in sequence labelling. MAT masks or replaces some words in the sentence when computing adversarial loss from perturbed inputs and consequently enhances model robustness using more context-level information. In our experiments, our method shows significant improvements on accuracy and robustness of sequence labelling. By further incorporating with ELMo embeddings, our model achieves better or comparable results to state-of-the-art on CoNLL 2000 and 2003 benchmarks using much less parameters.</abstract>
      <url hash="c90e8180">2020.findings-emnlp.28</url>
      <doi>10.18653/v1/2020.findings-emnlp.28</doi>
    </paper>
    <paper id="29">
      <title>Multilingual Argument Mining: Datasets and Analysis</title>
      <author><first>Orith</first><last>Toledo-Ronen</last></author>
      <author><first>Matan</first><last>Orbach</last></author>
      <author><first>Yonatan</first><last>Bilu</last></author>
      <author><first>Artem</first><last>Spector</last></author>
      <author><first>Noam</first><last>Slonim</last></author>
      <pages>303–317</pages>
      <abstract>The growing interest in argument mining and computational argumentation brings with it a plethora of Natural Language Understanding (NLU) tasks and corresponding datasets. However, as with many other NLU tasks, the dominant language is English, with resources in other languages being few and far between. In this work, we explore the potential of transfer learning using the multilingual BERT model to address argument mining tasks in non-English languages, based on English datasets and the use of machine translation. We show that such methods are well suited for classifying the stance of arguments and detecting evidence, but less so for assessing the quality of arguments, presumably because quality is harder to preserve under translation. In addition, focusing on the translate-train approach, we show how the choice of languages for translation, and the relations among them, affect the accuracy of the resultant model. Finally, to facilitate evaluation of transfer learning on argument mining tasks, we provide a human-generated dataset with more than 10k arguments in multiple languages, as well as machine translation of the English datasets.</abstract>
      <url hash="2984952a">2020.findings-emnlp.29</url>
      <doi>10.18653/v1/2020.findings-emnlp.29</doi>
    </paper>
    <paper id="30">
      <title>Improving Grammatical Error Correction with Machine Translation Pairs</title>
      <author><first>Wangchunshu</first><last>Zhou</last></author>
      <author><first>Tao</first><last>Ge</last></author>
      <author><first>Chang</first><last>Mu</last></author>
      <author><first>Ke</first><last>Xu</last></author>
      <author><first>Furu</first><last>Wei</last></author>
      <author><first>Ming</first><last>Zhou</last></author>
      <pages>318–328</pages>
      <abstract>We propose a novel data synthesis method to generate diverse error-corrected sentence pairs for improving grammatical error correction, which is based on a pair of machine translation models (e.g., Chinese to English) of different qualities (i.e., poor and good). The poor translation model can resemble the ESL (English as a second language) learner and tends to generate translations of low quality in terms of fluency and grammaticality, while the good translation model generally generates fluent and grammatically correct translations. With the pair of translation models, we can generate unlimited numbers of poor to good English sentence pairs from text in the source language (e.g., Chinese) of the translators. Our approach can generate various error-corrected patterns and nicely complement the other data synthesis approaches for GEC. Experimental results demonstrate the data generated by our approach can effectively help a GEC model to improve the performance and achieve the state-of-the-art single-model performance in BEA-19 and CoNLL-14 benchmark datasets.</abstract>
      <url hash="1995e577">2020.findings-emnlp.30</url>
      <doi>10.18653/v1/2020.findings-emnlp.30</doi>
    </paper>
    <paper id="31">
      <title>Machines Getting with the Program: Understanding Intent Arguments of Non-Canonical Directives</title>
      <author><first>Won Ik</first><last>Cho</last></author>
      <author><first>Youngki</first><last>Moon</last></author>
      <author><first>Sangwhan</first><last>Moon</last></author>
      <author><first>Seok Min</first><last>Kim</last></author>
      <author><first>Nam Soo</first><last>Kim</last></author>
      <pages>329–339</pages>
      <abstract>Modern dialog managers face the challenge of having to fulfill human-level conversational skills as part of common user expectations, including but not limited to discourse with no clear objective. Along with these requirements, agents are expected to extrapolate intent from the user’s dialogue even when subjected to non-canonical forms of speech. This depends on the agent’s comprehension of paraphrased forms of such utterances. Especially in low-resource languages, the lack of data is a bottleneck that prevents advancements of the comprehension performance for these types of agents. In this regard, here we demonstrate the necessity of extracting the intent argument of non-canonical directives in a natural language format, which may yield more accurate parsing, and suggest guidelines for building a parallel corpus for this purpose. Following the guidelines, we construct a Korean corpus of 50K instances of question/command-intent pairs, including the labels for classification of the utterance type. We also propose a method for mitigating class imbalance, demonstrating the potential applications of the corpus generation method and its multilingual extensibility.</abstract>
      <url hash="e52c1ad5">2020.findings-emnlp.31</url>
      <doi>10.18653/v1/2020.findings-emnlp.31</doi>
    </paper>
    <paper id="32">
      <title>The <fixed-case>RELX</fixed-case> Dataset and Matching the Multilingual Blanks for Cross-Lingual Relation Classification</title>
      <author><first>Abdullatif</first><last>Köksal</last></author>
      <author><first>Arzucan</first><last>Özgür</last></author>
      <pages>340–350</pages>
      <abstract>Relation classification is one of the key topics in information extraction, which can be used to construct knowledge bases or to provide useful information for question answering. Current approaches for relation classification are mainly focused on the English language and require lots of training data with human annotations. Creating and annotating a large amount of training data for low-resource languages is impractical and expensive. To overcome this issue, we propose two cross-lingual relation classification models: a baseline model based on Multilingual BERT and a new multilingual pretraining setup, which significantly improves the baseline with distant supervision. For evaluation, we introduce a new public benchmark dataset for cross-lingual relation classification in English, French, German, Spanish, and Turkish, called RELX. We also provide the RELX-Distant dataset, which includes hundreds of thousands of sentences with relations from Wikipedia and Wikidata collected by distant supervision for these languages. Our code and data are available at: https://github.com/boun-tabi/RELX</abstract>
      <url hash="493bd041">2020.findings-emnlp.32</url>
      <doi>10.18653/v1/2020.findings-emnlp.32</doi>
    </paper>
    <paper id="33">
      <title>Control, Generate, Augment: A Scalable Framework for Multi-Attribute Text Generation</title>
      <author><first>Giuseppe</first><last>Russo</last></author>
      <author><first>Nora</first><last>Hollenstein</last></author>
      <author><first>Claudiu Cristian</first><last>Musat</last></author>
      <author><first>Ce</first><last>Zhang</last></author>
      <pages>351–366</pages>
      <abstract>We introduce CGA, a conditional VAE architecture, to control, generate, and augment text. CGA is able to generate natural English sentences controlling multiple semantic and syntactic attributes by combining adversarial learning with a context-aware loss and a cyclical word dropout routine. We demonstrate the value of the individual model components in an ablation study. The scalability of our approach is ensured through a single discriminator, independently of the number of attributes. We show high quality, diversity and attribute control in the generated sentences through a series of automatic and human assessments. As the main application of our work, we test the potential of this new NLG model in a data augmentation scenario. In a downstream NLP task, the sentences generated by our CGA model show significant improvements over a strong baseline, and a classification performance often comparable to adding same amount of additional real data.</abstract>
      <url hash="a0ff56f1">2020.findings-emnlp.33</url>
      <doi>10.18653/v1/2020.findings-emnlp.33</doi>
    </paper>
    <paper id="34">
      <title>Open-Ended Visual Question Answering by Multi-Modal Domain Adaptation</title>
      <author><first>Yiming</first><last>Xu</last></author>
      <author><first>Lin</first><last>Chen</last></author>
      <author><first>Zhongwei</first><last>Cheng</last></author>
      <author><first>Lixin</first><last>Duan</last></author>
      <author><first>Jiebo</first><last>Luo</last></author>
      <pages>367–376</pages>
      <abstract>We study the problem of visual question answering (VQA) in images by exploiting supervised domain adaptation, where there is a large amount of labeled data in the source domain but only limited labeled data in the target domain, with the goal to train a good target model. A straightforward solution is to fine-tune a pre-trained source model by using those limited labeled target data, but it usually cannot work well due to the considerable difference between the data distributions of the source and target domains. Moreover, the availability of multiple modalities (i.e., images, questions and answers) in VQA poses further challenges in modeling the transferability between various modalities. In this paper, we address the above issues by proposing a novel supervised multi-modal domain adaptation method for VQA to learn joint feature embeddings across different domains and modalities. Specifically, we align the data distributions of the source and target domains by considering those modalities both jointly and separately. Extensive experiments on the benchmark VQA 2.0 and VizWiz datasets demonstrate that our proposed method outperforms the existing state-of-the-art baselines for open-ended VQA in this challenging domain adaptation setting.</abstract>
      <url hash="e84f2b61">2020.findings-emnlp.34</url>
      <doi>10.18653/v1/2020.findings-emnlp.34</doi>
    </paper>
    <paper id="35">
      <title>Dual Low-Rank Multimodal Fusion</title>
      <author><first>Tao</first><last>Jin</last></author>
      <author><first>Siyu</first><last>Huang</last></author>
      <author><first>Yingming</first><last>Li</last></author>
      <author><first>Zhongfei</first><last>Zhang</last></author>
      <pages>377–387</pages>
      <abstract>Tensor-based fusion methods have been proven effective in multimodal fusion tasks. However, existing tensor-based methods make a poor use of the fine-grained temporal dynamics of multimodal sequential features. Motivated by this observation, this paper proposes a novel multimodal fusion method called Fine-Grained Temporal Low-Rank Multimodal Fusion (FT-LMF). FT-LMF correlates the features of individual time steps between multiple modalities, while it involves multiplications of high-order tensors in its calculation. This paper further proposes Dual Low-Rank Multimodal Fusion (Dual-LMF) to reduce the computational complexity of FT-LMF through low-rank tensor approximation along dual dimensions of input features. Dual-LMF is conceptually simple and practically effective and efficient. Empirical studies on benchmark multimodal analysis tasks show that our proposed methods outperform the state-of-the-art tensor-based fusion methods with a similar computational complexity.</abstract>
      <url hash="02353d35">2020.findings-emnlp.35</url>
      <attachment type="OptionalSupplementaryMaterial" hash="7b2ce962">2020.findings-emnlp.35.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.35</doi>
    </paper>
    <paper id="36">
      <title>Contextual Modulation for Relation-Level Metaphor Identification</title>
      <author><first>Omnia</first><last>Zayed</last></author>
      <author><first>John P.</first><last>McCrae</last></author>
      <author><first>Paul</first><last>Buitelaar</last></author>
      <pages>388–406</pages>
      <abstract>Identifying metaphors in text is very challenging and requires comprehending the underlying comparison. The automation of this cognitive process has gained wide attention lately. However, the majority of existing approaches concentrate on word-level identification by treating the task as either single-word classification or sequential labelling without explicitly modelling the interaction between the metaphor components. On the other hand, while existing relation-level approaches implicitly model this interaction, they ignore the context where the metaphor occurs. In this work, we address these limitations by introducing a novel architecture for identifying relation-level metaphoric expressions of certain grammatical relations based on contextual modulation. In a methodology inspired by works in visual reasoning, our approach is based on conditioning the neural network computation on the deep contextualised features of the candidate expressions using feature-wise linear modulation. We demonstrate that the proposed architecture achieves state-of-the-art results on benchmark datasets. The proposed methodology is generic and could be applied to other textual classification problems that benefit from contextual interaction.</abstract>
      <url hash="87a488fb">2020.findings-emnlp.36</url>
      <doi>10.18653/v1/2020.findings-emnlp.36</doi>
    </paper>
    <paper id="37">
      <title>Context-aware Stand-alone Neural Spelling Correction</title>
      <author><first>Xiangci</first><last>Li</last></author>
      <author><first>Hairong</first><last>Liu</last></author>
      <author><first>Liang</first><last>Huang</last></author>
      <pages>407–414</pages>
      <abstract>Existing natural language processing systems are vulnerable to noisy inputs resulting from misspellings. On the contrary, humans can easily infer the corresponding correct words from their misspellings and surrounding context. Inspired by this, we address the stand-alone spelling correction problem, which only corrects the spelling of each token without additional token insertion or deletion, by utilizing both spelling information and global context representations. We present a simple yet powerful solution that jointly detects and corrects misspellings as a sequence labeling task by fine-turning a pre-trained language model. Our solution outperform the previous state-of-the-art result by 12.8% absolute F0.5 score.</abstract>
      <url hash="d4f9903a">2020.findings-emnlp.37</url>
      <doi>10.18653/v1/2020.findings-emnlp.37</doi>
    </paper>
    <paper id="38">
      <title>A Novel Workflow for Accurately and Efficiently Crowdsourcing Predicate Senses and Argument Labels</title>
      <author><first>Youxuan</first><last>Jiang</last></author>
      <author><first>Huaiyu</first><last>Zhu</last></author>
      <author><first>Jonathan K.</first><last>Kummerfeld</last></author>
      <author><first>Yunyao</first><last>Li</last></author>
      <author><first>Walter</first><last>Lasecki</last></author>
      <pages>415–421</pages>
      <abstract>Resources for Semantic Role Labeling (SRL) are typically annotated by experts at great expense. Prior attempts to develop crowdsourcing methods have either had low accuracy or required substantial expert annotation. We propose a new multi-stage crowd workflow that substantially reduces expert involvement without sacrificing accuracy. In particular, we introduce a unique filter stage based on the key observation that crowd workers are able to almost perfectly filter out incorrect options for labels. Our three-stage workflow produces annotations with 95% accuracy for predicate labels and 93% for argument labels, which is comparable to expert agreement. Compared to prior work on crowdsourcing for SRL, we decrease expert effort by 4x, from 56% to 14% of cases. Our approach enables more scalable annotation of SRL, and could enable annotation of NLP tasks that have previously been considered too complex to effectively crowdsource.</abstract>
      <url hash="ba5398bc">2020.findings-emnlp.38</url>
      <attachment type="OptionalSupplementaryMaterial" hash="ffe48dc2">2020.findings-emnlp.38.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.38</doi>
    </paper>
    <paper id="39">
      <title><fixed-case>K</fixed-case>or<fixed-case>NLI</fixed-case> and <fixed-case>K</fixed-case>or<fixed-case>STS</fixed-case>: New Benchmark Datasets for <fixed-case>K</fixed-case>orean Natural Language Understanding</title>
      <author><first>Jiyeon</first><last>Ham</last></author>
      <author><first>Yo Joong</first><last>Choe</last></author>
      <author><first>Kyubyong</first><last>Park</last></author>
      <author><first>Ilji</first><last>Choi</last></author>
      <author><first>Hyungjoon</first><last>Soh</last></author>
      <pages>422–430</pages>
      <abstract>Natural language inference (NLI) and semantic textual similarity (STS) are key tasks in natural language understanding (NLU). Although several benchmark datasets for those tasks have been released in English and a few other languages, there are no publicly available NLI or STS datasets in the Korean language. Motivated by this, we construct and release new datasets for Korean NLI and STS, dubbed KorNLI and KorSTS, respectively. Following previous approaches, we machine-translate existing English training sets and manually translate development and test sets into Korean. To accelerate research on Korean NLU, we also establish baselines on KorNLI and KorSTS. Our datasets are publicly available at https://github.com/kakaobrain/KorNLUDatasets.</abstract>
      <url hash="9b37635d">2020.findings-emnlp.39</url>
      <doi>10.18653/v1/2020.findings-emnlp.39</doi>
    </paper>
    <paper id="40">
      <title>Dialogue Generation on Infrequent Sentence Functions via Structured Meta-Learning</title>
      <author><first>Yifan</first><last>Gao</last></author>
      <author><first>Piji</first><last>Li</last></author>
      <author><first>Wei</first><last>Bi</last></author>
      <author><first>Xiaojiang</first><last>Liu</last></author>
      <author><first>Michael</first><last>Lyu</last></author>
      <author><first>Irwin</first><last>King</last></author>
      <pages>431–440</pages>
      <abstract>Sentence function is an important linguistic feature indicating the communicative purpose in uttering a sentence. Incorporating sentence functions into conversations has shown improvements in the quality of generated responses. However, the number of utterances for different types of fine-grained sentence functions is extremely imbalanced. Besides a small number of high-resource sentence functions, a large portion of sentence functions is infrequent. Consequently, dialogue generation conditioned on these infrequent sentence functions suffers from data deficiency. In this paper, we investigate a structured meta-learning (SML) approach for dialogue generation on infrequent sentence functions. We treat dialogue generation conditioned on different sentence functions as separate tasks, and apply model-agnostic meta-learning to high-resource sentence functions data. Furthermore, SML enhances meta-learning effectiveness by promoting knowledge customization among different sentence functions but simultaneously preserving knowledge generalization for similar sentence functions. Experimental results demonstrate that SML not only improves the informativeness and relevance of generated responses, but also can generate responses consistent with the target sentence functions. Code will be public to facilitate the research along this line.</abstract>
      <url hash="aeed4d22">2020.findings-emnlp.40</url>
      <doi>10.18653/v1/2020.findings-emnlp.40</doi>
    </paper>
    <paper id="41">
      <title>Exploring Versatile Generative Language Model Via Parameter-Efficient Transfer Learning</title>
      <author><first>Zhaojiang</first><last>Lin</last></author>
      <author><first>Andrea</first><last>Madotto</last></author>
      <author><first>Pascale</first><last>Fung</last></author>
      <pages>441–459</pages>
      <abstract>Fine-tuning pre-trained generative language models to down-stream language generation tasks has shown promising results. However, this comes with the cost of having a single, large model for each task, which is not ideal in low-memory/power scenarios (e.g., mobile). In this paper, we propose an effective way to fine-tune multiple down-stream generation tasks simultaneously using a single, large pretrained model. The experiments on five diverse language generation tasks show that by just using an additional 2-3% parameters for each task, our model can maintain or even improve the performance of fine-tuning the whole model.</abstract>
      <url hash="94c828b9">2020.findings-emnlp.41</url>
      <attachment type="OptionalSupplementaryMaterial" hash="2a014db2">2020.findings-emnlp.41.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.41</doi>
    </paper>
    <paper id="42">
      <title>A Fully Hyperbolic Neural Model for Hierarchical Multi-Class Classification</title>
      <author><first>Federico</first><last>López</last></author>
      <author><first>Michael</first><last>Strube</last></author>
      <pages>460–475</pages>
      <abstract>Label inventories for fine-grained entity typing have grown in size and complexity. Nonetheless, they exhibit a hierarchical structure. Hyperbolic spaces offer a mathematically appealing approach for learning hierarchical representations of symbolic data. However, it is not clear how to integrate hyperbolic components into downstream tasks. This is the first work that proposes a fully hyperbolic model for multi-class multi-label classification, which performs all operations in hyperbolic space. We evaluate the proposed model on two challenging datasets and compare to different baselines that operate under Euclidean assumptions. Our hyperbolic model infers the latent hierarchy from the class distribution, captures implicit hyponymic relations in the inventory, and shows performance on par with state-of-the-art methods on fine-grained classification with remarkable reduction of the parameter size. A thorough analysis sheds light on the impact of each component in the final prediction and showcases its ease of integration with Euclidean layers.</abstract>
      <url hash="91ec97ee">2020.findings-emnlp.42</url>
      <doi>10.18653/v1/2020.findings-emnlp.42</doi>
    </paper>
    <paper id="43">
      <title>Claim Check-Worthiness Detection as Positive Unlabelled Learning</title>
      <author><first>Dustin</first><last>Wright</last></author>
      <author><first>Isabelle</first><last>Augenstein</last></author>
      <pages>476–488</pages>
      <abstract>As the first step of automatic fact checking, claim check-worthiness detection is a critical component of fact checking systems. There are multiple lines of research which study this problem: check-worthiness ranking from political speeches and debates, rumour detection on Twitter, and citation needed detection from Wikipedia. To date, there has been no structured comparison of these various tasks to understand their relatedness, and no investigation into whether or not a unified approach to all of them is achievable. In this work, we illuminate a central challenge in claim check-worthiness detection underlying all of these tasks, being that they hinge upon detecting both how factual a sentence is, as well as how likely a sentence is to be believed without verification. As such, annotators only mark those instances they judge to be clear-cut check-worthy. Our best performing method is a unified approach which automatically corrects for this using a variant of positive unlabelled learning that finds instances which were incorrectly labelled as not check-worthy. In applying this, we out-perform the state of the art in two of the three tasks studied for claim check-worthiness detection in English.</abstract>
      <url hash="720def73">2020.findings-emnlp.43</url>
      <doi>10.18653/v1/2020.findings-emnlp.43</doi>
    </paper>
    <paper id="44">
      <title><fixed-case>C</fixed-case>oncept<fixed-case>B</fixed-case>ert: Concept-Aware Representation for Visual Question Answering</title>
      <author><first>François</first><last>Gardères</last></author>
      <author><first>Maryam</first><last>Ziaeefard</last></author>
      <author><first>Baptiste</first><last>Abeloos</last></author>
      <author><first>Freddy</first><last>Lecue</last></author>
      <pages>489–498</pages>
      <abstract>Visual Question Answering (VQA) is a challenging task that has received increasing attention from both the computer vision and the natural language processing communities. A VQA model combines visual and textual features in order to answer questions grounded in an image. Current works in VQA focus on questions which are answerable by direct analysis of the question and image alone. We present a concept-aware algorithm, ConceptBert, for questions which require common sense, or basic factual knowledge from external structured content. Given an image and a question in natural language, ConceptBert requires visual elements of the image and a Knowledge Graph (KG) to infer the correct answer. We introduce a multi-modal representation which learns a joint Concept-Vision-Language embedding inspired by the popular BERT architecture. We exploit ConceptNet KG for encoding the common sense knowledge and evaluate our methodology on the Outside Knowledge-VQA (OK-VQA) and VQA datasets.</abstract>
      <url hash="180c5663">2020.findings-emnlp.44</url>
      <doi>10.18653/v1/2020.findings-emnlp.44</doi>
    </paper>
    <paper id="45">
      <title>Bootstrapping a Crosslingual Semantic Parser</title>
      <author><first>Tom</first><last>Sherborne</last></author>
      <author><first>Yumo</first><last>Xu</last></author>
      <author><first>Mirella</first><last>Lapata</last></author>
      <pages>499–517</pages>
      <abstract>Recent progress in semantic parsing scarcely considers languages other than English but professional translation can be prohibitively expensive. We adapt a semantic parser trained on a single language, such as English, to new languages and multiple domains with minimal annotation. We query if machine translation is an adequate substitute for training data, and extend this to investigate bootstrapping using joint training with English, paraphrasing, and multilingual pre-trained models. We develop a Transformer-based parser combining paraphrases by ensembling attention over multiple encoders and present new versions of ATIS and Overnight in German and Chinese for evaluation. Experimental results indicate that MT can approximate training data in a new language for accurate parsing when augmented with paraphrasing through multiple MT engines. Considering when MT is inadequate, we also find that using our approach achieves parsing accuracy within 2% of complete translation using only 50% of training data.</abstract>
      <url hash="33984c37">2020.findings-emnlp.45</url>
      <doi>10.18653/v1/2020.findings-emnlp.45</doi>
    </paper>
    <paper id="46">
      <title>Revisiting Representation Degeneration Problem in Language Modeling</title>
      <author><first>Zhong</first><last>Zhang</last></author>
      <author><first>Chongming</first><last>Gao</last></author>
      <author><first>Cong</first><last>Xu</last></author>
      <author><first>Rui</first><last>Miao</last></author>
      <author><first>Qinli</first><last>Yang</last></author>
      <author><first>Junming</first><last>Shao</last></author>
      <pages>518–527</pages>
      <abstract>Weight tying is now a common setting in many language generation tasks such as language modeling and machine translation. However, a recent study reveals that there is a potential flaw in weight tying. They find that the learned word embeddings are likely to degenerate and lie in a narrow cone when training a language model. They call it the representation degeneration problem and propose a cosine regularization to solve it. Nevertheless, we prove that the cosine regularization is insufficient to solve the problem, as the degeneration is still likely to happen under certain conditions. In this paper, we revisit the representation degeneration problem and theoretically analyze the limitations of the previously proposed solution. Afterward, we propose an alternative regularization method called Laplacian regularization to tackle the problem. Experiments on language modeling demonstrate the effectiveness of the proposed Laplacian regularization.</abstract>
      <url hash="caeb7d96">2020.findings-emnlp.46</url>
      <doi>10.18653/v1/2020.findings-emnlp.46</doi>
    </paper>
    <paper id="47">
      <title>The workweek is the best time to start a family – A Study of <fixed-case>GPT</fixed-case>-2 Based Claim Generation</title>
      <author><first>Shai</first><last>Gretz</last></author>
      <author><first>Yonatan</first><last>Bilu</last></author>
      <author><first>Edo</first><last>Cohen-Karlik</last></author>
      <author><first>Noam</first><last>Slonim</last></author>
      <pages>528–544</pages>
      <abstract>Argument generation is a challenging task whose research is timely considering its potential impact on social media and the dissemination of information. Here we suggest a pipeline based on GPT-2 for generating coherent claims, and explore the types of claims that it produces, and their veracity, using an array of manual and automatic assessments. In addition, we explore the interplay between this task and the task of Claim Retrieval, showing how they can complement one another.</abstract>
      <url hash="c9e7447c">2020.findings-emnlp.47</url>
      <doi>10.18653/v1/2020.findings-emnlp.47</doi>
    </paper>
    <paper id="48">
      <title>Dynamic Data Selection for Curriculum Learning via Ability Estimation</title>
      <author><first>John P.</first><last>Lalor</last></author>
      <author><first>Hong</first><last>Yu</last></author>
      <pages>545–555</pages>
      <abstract>Curriculum learning methods typically rely on heuristics to estimate the difficulty of training examples or the ability of the model. In this work, we propose replacing difficulty heuristics with learned difficulty parameters. We also propose Dynamic Data selection for Curriculum Learning via Ability Estimation (DDaCLAE), a strategy that probes model ability at each training epoch to select the best training examples at that point. We show that models using learned difficulty and/or ability outperform heuristic-based curriculum learning models on the GLUE classification tasks.</abstract>
      <url hash="15608826">2020.findings-emnlp.48</url>
      <doi>10.18653/v1/2020.findings-emnlp.48</doi>
    </paper>
    <paper id="49">
      <title>Fixed Encoder Self-Attention Patterns in Transformer-Based Machine Translation</title>
      <author><first>Alessandro</first><last>Raganato</last></author>
      <author><first>Yves</first><last>Scherrer</last></author>
      <author><first>Jörg</first><last>Tiedemann</last></author>
      <pages>556–568</pages>
      <abstract>Transformer-based models have brought a radical change to neural machine translation. A key feature of the Transformer architecture is the so-called multi-head attention mechanism, which allows the model to focus simultaneously on different parts of the input. However, recent works have shown that most attention heads learn simple, and often redundant, positional patterns. In this paper, we propose to replace all but one attention head of each encoder layer with simple fixed – non-learnable – attentive patterns that are solely based on position and do not require any external knowledge. Our experiments with different data sizes and multiple language pairs show that fixing the attention heads on the encoder side of the Transformer at training time does not impact the translation quality and even increases BLEU scores by up to 3 points in low-resource scenarios.</abstract>
      <url hash="1c0718e0">2020.findings-emnlp.49</url>
      <doi>10.18653/v1/2020.findings-emnlp.49</doi>
    </paper>
    <paper id="50">
      <title><fixed-case>ZEST</fixed-case>: Zero-shot Learning from Text Descriptions using Textual Similarity and Visual Summarization</title>
      <author><first>Tzuf</first><last>Paz-Argaman</last></author>
      <author><first>Reut</first><last>Tsarfaty</last></author>
      <author><first>Gal</first><last>Chechik</last></author>
      <author><first>Yuval</first><last>Atzmon</last></author>
      <pages>569–579</pages>
      <abstract>We study the problem of recognizing visual entities from the textual descriptions of their classes. Specifically, given birds’ images with free-text descriptions of their species, we learn to classify images of previously-unseen species based on specie descriptions. This setup has been studied in the vision community under the name zero-shot learning from text, focusing on learning to transfer knowledge about visual aspects of birds from seen classes to previously-unseen ones. Here, we suggest focusing on the textual description and distilling from the description the most relevant information to effectively match visual features to the parts of the text that discuss them. Specifically, (1) we propose to leverage the similarity between species, reflected in the similarity between text descriptions of the species. (2) we derive visual summaries of the texts, i.e., extractive summaries that focus on the visual features that tend to be reflected in images. We propose a simple attention-based model augmented with the similarity and visual summaries components. Our empirical results consistently and significantly outperform the state-of-the-art on the largest benchmarks for text-based zero-shot learning, illustrating the critical importance of texts for zero-shot image-recognition.</abstract>
      <url hash="307870e2">2020.findings-emnlp.50</url>
      <doi>10.18653/v1/2020.findings-emnlp.50</doi>
    </paper>
    <paper id="51">
      <title>Few-Shot Multi-Hop Relation Reasoning over Knowledge Bases</title>
      <author><first>Chuxu</first><last>Zhang</last></author>
      <author><first>Lu</first><last>Yu</last></author>
      <author><first>Mandana</first><last>Saebi</last></author>
      <author><first>Meng</first><last>Jiang</last></author>
      <author><first>Nitesh</first><last>Chawla</last></author>
      <pages>580–585</pages>
      <abstract>Multi-hop relation reasoning over knowledge base is to generate effective and interpretable relation prediction through reasoning paths. The current methods usually require sufficient training data (fact triples) for each query relation, impairing their performances over few-shot relations (with limited triples) which are common in knowledge base. To this end, we propose FIRE, a novel few-shot multi-hop relation learning model. FIRE applies reinforcement learning to model the sequential steps of multi-hop reasoning, besides performs heterogeneous structure encoding and knowledge-aware search space pruning. The meta-learning technique is employed to optimize model parameters that could quickly adapt to few-shot relations. Empirical study on two datasets demonstrate that FIRE outperforms state-of-the-art methods.</abstract>
      <url hash="37de63ad">2020.findings-emnlp.51</url>
      <doi>10.18653/v1/2020.findings-emnlp.51</doi>
    </paper>
    <paper id="52">
      <title>A structure-enhanced graph convolutional network for sentiment analysis</title>
      <author><first>Fanyu</first><last>Meng</last></author>
      <author><first>Junlan</first><last>Feng</last></author>
      <author><first>Danping</first><last>Yin</last></author>
      <author><first>Si</first><last>Chen</last></author>
      <author><first>Min</first><last>Hu</last></author>
      <pages>586–595</pages>
      <abstract>Syntactic information is essential for both sentiment analysis(SA) and aspect-based sentiment analysis(ABSA). Previous work has already achieved great progress utilizing Graph Convolutional Network(GCN) over dependency tree of a sentence. However, these models do not fully exploit the syntactic information obtained from dependency parsing such as the diversified types of dependency relations. The message passing process of GCN should be distinguished based on these syntactic information.To tackle this problem, we design a novel weighted graph convolutional network(WGCN) which can exploit rich syntactic information based on the feature combination. Furthermore, we utilize BERT instead of Bi-LSTM to generate contextualized representations as inputs for GCN and present an alignment method to keep word-level dependencies consistent with wordpiece unit of BERT. With our proposal, we are able to improve the state-of-the-art on four ABSA tasks out of six and two SA tasks out of three.</abstract>
      <url hash="3175d0c8">2020.findings-emnlp.52</url>
      <doi>10.18653/v1/2020.findings-emnlp.52</doi>
    </paper>
    <paper id="53">
      <title><fixed-case>PB</fixed-case>o<fixed-case>S</fixed-case>: Probabilistic Bag-of-Subwords for Generalizing Word Embedding</title>
      <author><first>Zhao</first><last>Jinman</last></author>
      <author><first>Shawn</first><last>Zhong</last></author>
      <author><first>Xiaomin</first><last>Zhang</last></author>
      <author><first>Yingyu</first><last>Liang</last></author>
      <pages>596–611</pages>
      <abstract>We look into the task of generalizing word embeddings: given a set of pre-trained word vectors over a finite vocabulary, the goal is to predict embedding vectors for out-of-vocabulary words, without extra contextual information. We rely solely on the spellings of words and propose a model, along with an efficient algorithm, that simultaneously models subword segmentation and computes subword-based compositional word embedding. We call the model probabilistic bag-of-subwords (PBoS), as it applies bag-of-subwords for all possible segmentations based on their likelihood. Inspections and affix prediction experiment show that PBoS is able to produce meaningful subword segmentations and subword rankings without any source of explicit morphological knowledge. Word similarity and POS tagging experiments show clear advantages of PBoS over previous subword-level models in the quality of generated word embeddings across languages.</abstract>
      <url hash="667f7f62">2020.findings-emnlp.53</url>
      <doi>10.18653/v1/2020.findings-emnlp.53</doi>
    </paper>
    <paper id="54">
      <title>Interpretable Entity Representations through Large-Scale Typing</title>
      <author><first>Yasumasa</first><last>Onoe</last></author>
      <author><first>Greg</first><last>Durrett</last></author>
      <pages>612–624</pages>
      <abstract>In standard methodology for natural language processing, entities in text are typically embedded in dense vector spaces with pre-trained models. The embeddings produced this way are effective when fed into downstream models, but they require end-task fine-tuning and are fundamentally difficult to interpret. In this paper, we present an approach to creating entity representations that are human readable and achieve high performance on entity-related tasks out of the box. Our representations are vectors whose values correspond to posterior probabilities over fine-grained entity types, indicating the confidence of a typing model’s decision that the entity belongs to the corresponding type. We obtain these representations using a fine-grained entity typing model, trained either on supervised ultra-fine entity typing data (Choi et al. 2018) or distantly-supervised examples from Wikipedia. On entity probing tasks involving recognizing entity identity, our embeddings used in parameter-free downstream models achieve competitive performance with ELMo- and BERT-based embeddings in trained models. We also show that it is possible to reduce the size of our type set in a learning-based way for particular domains. Finally, we show that these embeddings can be post-hoc modified through a small number of rules to incorporate domain knowledge and improve performance.</abstract>
      <url hash="09691a65">2020.findings-emnlp.54</url>
      <doi>10.18653/v1/2020.findings-emnlp.54</doi>
    </paper>
    <paper id="55">
      <title>Empirical Studies of Institutional Federated Learning For Natural Language Processing</title>
      <author><first>Xinghua</first><last>Zhu</last></author>
      <author><first>Jianzong</first><last>Wang</last></author>
      <author><first>Zhenhou</first><last>Hong</last></author>
      <author><first>Jing</first><last>Xiao</last></author>
      <pages>625–634</pages>
      <abstract>Federated learning has sparkled new interests in the deep learning society to make use of isolated data sources from independent institutes. With the development of novel training tools, we have successfully deployed federated natural language processing networks on GPU-enabled server clusters. This paper demonstrates federated training of a popular NLP model, TextCNN, with applications in sentence intent classification. Furthermore, differential privacy is introduced to protect participants in the training process, in a manageable manner. Distinguished from previous client-level privacy protection schemes, the proposed differentially private federated learning procedure is defined in the dataset sample level, inherent with the applications among institutions instead of individual users. Optimal settings of hyper-parameters for the federated TextCNN model are studied through comprehensive experiments. We also evaluated the performance of federated TextCNN model under imbalanced data load configuration. Experiments show that, the sampling ratio has a large impact on the performance of the FL models, causing up to 38.4% decrease in the test accuracy, while they are robust to different noise multiplier levels, with less than 3% variance in the test accuracy. It is also found that the FL models are sensitive to data load balancedness among client datasets. When the data load is imbalanced, model performance dropped by up to 10%.</abstract>
      <url hash="cdb93921">2020.findings-emnlp.55</url>
      <doi>10.18653/v1/2020.findings-emnlp.55</doi>
    </paper>
    <paper id="56">
      <title><fixed-case>N</fixed-case>eu<fixed-case>R</fixed-case>educe: Reducing Mixed <fixed-case>B</fixed-case>oolean-Arithmetic Expressions by Recurrent Neural Network</title>
      <author><first>Weijie</first><last>Feng</last></author>
      <author><first>Binbin</first><last>Liu</last></author>
      <author><first>Dongpeng</first><last>Xu</last></author>
      <author><first>Qilong</first><last>Zheng</last></author>
      <author><first>Yun</first><last>Xu</last></author>
      <pages>635–644</pages>
      <abstract>Mixed Boolean-Arithmetic (MBA) expressions involve both arithmetic calculation (e.g.,plus, minus, multiply) and bitwise computation (e.g., and, or, negate, xor). MBA expressions have been widely applied in software obfuscation, transforming programs from a simple form to a complex form. MBA expressions are challenging to be simplified, because the interleaving bitwise and arithmetic operations causing mathematical reduction laws to be ineffective. Our goal is to recover the original, simple form from an obfuscated MBA expression. In this paper, we first propose NeuReduce, a string to string method based on neural networks to automatically learn and reduce complex MBA expressions. We develop a comprehensive MBA dataset, including one million diversified MBA expression samples and corresponding simplified forms. After training on the dataset, NeuReduce can reduce MBA rules to homelier but mathematically equivalent forms. By comparing with three state-of-the-art MBA reduction methods, our evaluation result shows that NeuReduce outperforms all other tools in terms of accuracy, solving time, and performance overhead.</abstract>
      <url hash="7563f08c">2020.findings-emnlp.56</url>
      <doi>10.18653/v1/2020.findings-emnlp.56</doi>
    </paper>
    <paper id="57">
      <title>From Language to Language-ish: How Brain-Like is an <fixed-case>LSTM</fixed-case>’s Representation of Nonsensical Language Stimuli?</title>
      <author><first>Maryam</first><last>Hashemzadeh</last></author>
      <author><first>Greta</first><last>Kaufeld</last></author>
      <author><first>Martha</first><last>White</last></author>
      <author><first>Andrea E.</first><last>Martin</last></author>
      <author><first>Alona</first><last>Fyshe</last></author>
      <pages>645–656</pages>
      <abstract>The representations generated by many models of language (word embeddings, recurrent neural networks and transformers) correlate to brain activity recorded while people read. However, these decoding results are usually based on the brain’s reaction to syntactically and semantically sound language stimuli. In this study, we asked: how does an LSTM (long short term memory) language model, trained (by and large) on semantically and syntactically intact language, represent a language sample with degraded semantic or syntactic information? Does the LSTM representation still resemble the brain’s reaction? We found that, even for some kinds of nonsensical language, there is a statistically significant relationship between the brain’s activity and the representations of an LSTM. This indicates that, at least in some instances, LSTMs and the human brain handle nonsensical data similarly.</abstract>
      <url hash="a64b646f">2020.findings-emnlp.57</url>
      <doi>10.18653/v1/2020.findings-emnlp.57</doi>
    </paper>
    <paper id="58">
      <title>Revisiting Pre-Trained Models for <fixed-case>C</fixed-case>hinese Natural Language Processing</title>
      <author><first>Yiming</first><last>Cui</last></author>
      <author><first>Wanxiang</first><last>Che</last></author>
      <author><first>Ting</first><last>Liu</last></author>
      <author><first>Bing</first><last>Qin</last></author>
      <author><first>Shijin</first><last>Wang</last></author>
      <author><first>Guoping</first><last>Hu</last></author>
      <pages>657–668</pages>
      <abstract>Bidirectional Encoder Representations from Transformers (BERT) has shown marvelous improvements across various NLP tasks, and consecutive variants have been proposed to further improve the performance of the pre-trained language models. In this paper, we target on revisiting Chinese pre-trained language models to examine their effectiveness in a non-English language and release the Chinese pre-trained language model series to the community. We also propose a simple but effective model called MacBERT, which improves upon RoBERTa in several ways, especially the masking strategy that adopts MLM as correction (Mac). We carried out extensive experiments on eight Chinese NLP tasks to revisit the existing pre-trained language models as well as the proposed MacBERT. Experimental results show that MacBERT could achieve state-of-the-art performances on many NLP tasks, and we also ablate details with several findings that may help future research. https://github.com/ymcui/MacBERT</abstract>
      <url hash="380bb6a3">2020.findings-emnlp.58</url>
      <doi>10.18653/v1/2020.findings-emnlp.58</doi>
    </paper>
    <paper id="59">
      <title>Cascaded Semantic and Positional Self-Attention Network for Document Classification</title>
      <author><first>Juyong</first><last>Jiang</last></author>
      <author><first>Jie</first><last>Zhang</last></author>
      <author><first>Kai</first><last>Zhang</last></author>
      <pages>669–677</pages>
      <abstract>Transformers have shown great success in learning representations for language modelling. However, an open challenge still remains on how to systematically aggregate semantic information (word embedding) with positional (or temporal) information (word orders). In this work, we propose a new architecture to aggregate the two sources of information using cascaded semantic and positional self-attention network (CSPAN) in the context of document classification. The CSPAN uses a semantic self-attention layer cascaded with Bi-LSTM to process the semantic and positional information in a sequential manner, and then adaptively combine them together through a residue connection. Compared with commonly used positional encoding schemes, CSPAN can exploit the interaction between semantics and word positions in a more interpretable and adaptive manner, and the classification performance can be notably improved while simultaneously preserving a compact model size and high convergence rate. We evaluate the CSPAN model on several benchmark data sets for document classification with careful ablation studies, and demonstrate the encouraging results compared with state of the art.</abstract>
      <url hash="f38ec081">2020.findings-emnlp.59</url>
      <doi>10.18653/v1/2020.findings-emnlp.59</doi>
    </paper>
    <paper id="60">
      <title>Toward Recognizing More Entity Types in <fixed-case>NER</fixed-case>: An Efficient Implementation using Only Entity Lexicons</title>
      <author><first>Minlong</first><last>Peng</last></author>
      <author><first>Ruotian</first><last>Ma</last></author>
      <author><first>Qi</first><last>Zhang</last></author>
      <author><first>Lujun</first><last>Zhao</last></author>
      <author><first>Mengxi</first><last>Wei</last></author>
      <author><first>Changlong</first><last>Sun</last></author>
      <author><first>Xuanjing</first><last>Huang</last></author>
      <pages>678–688</pages>
      <abstract>In this work, we explore the way to quickly adjust an existing named entity recognition (NER) system to make it capable of recognizing entity types not defined in the system. As an illustrative example, consider the case that a NER system has been built to recognize person and organization names, and now it requires to additionally recognize job titles. Such a situation is common in the industrial areas, where the entity types required to recognize vary a lot in different products and keep changing. To avoid laborious data labeling and achieve fast adaptation, we propose to adjust the existing NER system using the previously labeled data and entity lexicons of the newly introduced entity types. We formulate such a task as a partially supervised learning problem and accordingly propose an effective algorithm to solve the problem. Comprehensive experimental studies on several public NER datasets validate the effectiveness of our method.</abstract>
      <url hash="fb5fd953">2020.findings-emnlp.60</url>
      <doi>10.18653/v1/2020.findings-emnlp.60</doi>
    </paper>
    <paper id="61">
      <title>From Disjoint Sets to Parallel Data to Train <fixed-case>S</fixed-case>eq2<fixed-case>S</fixed-case>eq Models for Sentiment Transfer</title>
      <author><first>Paulo</first><last>Cavalin</last></author>
      <author><first>Marisa</first><last>Vasconcelos</last></author>
      <author><first>Marcelo</first><last>Grave</last></author>
      <author><first>Claudio</first><last>Pinhanez</last></author>
      <author><first>Victor Henrique</first><last>Alves Ribeiro</last></author>
      <pages>689–698</pages>
      <abstract>We present a method for creating parallel data to train Seq2Seq neural networks for sentiment transfer. Most systems for this task, which can be viewed as monolingual machine translation (MT), have relied on unsupervised methods, such as Generative Adversarial Networks (GANs)-inspired approaches, for coping with the lack of parallel corpora. Given that the literature shows that Seq2Seq methods have been consistently outperforming unsupervised methods in MT-related tasks, in this work we exploit the use of semantic similarity computation for converting non-parallel data onto a parallel corpus. That allows us to train a transformer neural network for the sentiment transfer task, and compare its performance against unsupervised approaches. With experiments conducted on two well-known public datasets, i.e. Yelp and Amazon, we demonstrate that the proposed methodology outperforms existing unsupervised methods very consistently in fluency, and presents competitive results in terms of sentiment conversion and content preservation. We believe that this works opens up an opportunity for seq2seq neural networks to be better exploited in problems for which they have not been applied owing to the lack of parallel training data.</abstract>
      <url hash="0fbbe87f">2020.findings-emnlp.61</url>
      <doi>10.18653/v1/2020.findings-emnlp.61</doi>
    </paper>
    <paper id="62">
      <title>Learning to Stop: A Simple yet Effective Approach to Urban Vision-Language Navigation</title>
      <author><first>Jiannan</first><last>Xiang</last></author>
      <author><first>Xin</first><last>Wang</last></author>
      <author><first>William Yang</first><last>Wang</last></author>
      <pages>699–707</pages>
      <abstract>Vision-and-Language Navigation (VLN) is a natural language grounding task where an agent learns to follow language instructions and navigate to specified destinations in real-world environments. A key challenge is to recognize and stop at the correct location, especially for complicated outdoor environments. Existing methods treat the STOP action equally as other actions, which results in undesirable behaviors that the agent often fails to stop at the destination even though it might be on the right path. Therefore, we propose Learning to Stop (L2Stop), a simple yet effective policy module that differentiates STOP and other actions. Our approach achieves the new state of the art on a challenging urban VLN dataset Touchdown, outperforming the baseline by 6.89% (absolute improvement) on Success weighted by Edit Distance (SED).</abstract>
      <url hash="8ad0458e">2020.findings-emnlp.62</url>
      <attachment type="OptionalSupplementaryMaterial" hash="59da7918">2020.findings-emnlp.62.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.62</doi>
    </paper>
    <paper id="63">
      <title>Document Ranking with a Pretrained Sequence-to-Sequence Model</title>
      <author><first>Rodrigo</first><last>Nogueira</last></author>
      <author><first>Zhiying</first><last>Jiang</last></author>
      <author><first>Ronak</first><last>Pradeep</last></author>
      <author><first>Jimmy</first><last>Lin</last></author>
      <pages>708–718</pages>
      <abstract>This work proposes the use of a pretrained sequence-to-sequence model for document ranking. Our approach is fundamentally different from a commonly adopted classification-based formulation based on encoder-only pretrained transformer architectures such as BERT. We show how a sequence-to-sequence model can be trained to generate relevance labels as “target tokens”, and how the underlying logits of these target tokens can be interpreted as relevance probabilities for ranking. Experimental results on the MS MARCO passage ranking task show that our ranking approach is superior to strong encoder-only models. On three other document retrieval test collections, we demonstrate a zero-shot transfer-based approach that outperforms previous state-of-the-art models requiring in-domain cross-validation. Furthermore, we find that our approach significantly outperforms an encoder-only architecture in a data-poor setting. We investigate this observation in more detail by varying target tokens to probe the model’s use of latent knowledge. Surprisingly, we find that the choice of target tokens impacts effectiveness, even for words that are closely related semantically. This finding sheds some light on why our sequence-to-sequence formulation for document ranking is effective. Code and models are available at pygaggle.ai.</abstract>
      <url hash="cb03f975">2020.findings-emnlp.63</url>
      <doi>10.18653/v1/2020.findings-emnlp.63</doi>
    </paper>
    <paper id="64">
      <title>Pruning Redundant Mappings in Transformer Models via Spectral-Normalized Identity Prior</title>
      <author><first>Zi</first><last>Lin</last></author>
      <author><first>Jeremiah</first><last>Liu</last></author>
      <author><first>Zi</first><last>Yang</last></author>
      <author><first>Nan</first><last>Hua</last></author>
      <author><first>Dan</first><last>Roth</last></author>
      <pages>719–730</pages>
      <abstract>Traditional (unstructured) pruning methods for a Transformer model focus on regularizing the individual weights by penalizing them toward zero. In this work, we explore spectral-normalized identity priors (SNIP), a structured pruning approach which penalizes an entire residual module in a Transformer model toward an identity mapping. Our method identifies and discards unimportant non-linear mappings in the residual connections by applying a thresholding operator on the function norm, and is applicable to any structured module including a single attention head, an entire attention blocks, or a feed-forward subnetwork. Furthermore, we introduce spectral normalization to stabilize the distribution of the post-activation values of the Transformer layers, further improving the pruning effectiveness of the proposed methodology. We conduct experiments with BERT on 5 GLUE benchmark tasks to demonstrate that SNIP achieves effective pruning results while maintaining comparable performance. Specifically, we improve the performance over the state-of-the-art by 0.5 to 1.0% on average at 50% compression ratio.</abstract>
      <url hash="3a01d707">2020.findings-emnlp.64</url>
      <doi>10.18653/v1/2020.findings-emnlp.64</doi>
    </paper>
    <paper id="65">
      <title>Rethinking Self-Attention: Towards Interpretability in Neural Parsing</title>
      <author><first>Khalil</first><last>Mrini</last></author>
      <author><first>Franck</first><last>Dernoncourt</last></author>
      <author><first>Quan Hung</first><last>Tran</last></author>
      <author><first>Trung</first><last>Bui</last></author>
      <author><first>Walter</first><last>Chang</last></author>
      <author><first>Ndapa</first><last>Nakashole</last></author>
      <pages>731–742</pages>
      <abstract>Attention mechanisms have improved the performance of NLP tasks while allowing models to remain explainable. Self-attention is currently widely used, however interpretability is difficult due to the numerous attention distributions. Recent work has shown that model representations can benefit from label-specific information, while facilitating interpretation of predictions. We introduce the Label Attention Layer: a new form of self-attention where attention heads represent labels. We test our novel layer by running constituency and dependency parsing experiments and show our new model obtains new state-of-the-art results for both tasks on both the Penn Treebank (PTB) and Chinese Treebank. Additionally, our model requires fewer self-attention layers compared to existing work. Finally, we find that the Label Attention heads learn relations between syntactic categories and show pathways to analyze errors.</abstract>
      <url hash="04abd82c">2020.findings-emnlp.65</url>
      <doi>10.18653/v1/2020.findings-emnlp.65</doi>
    </paper>
    <paper id="66">
      <title><fixed-case>P</fixed-case>olicy<fixed-case>QA</fixed-case>: A Reading Comprehension Dataset for Privacy Policies</title>
      <author><first>Wasi</first><last>Ahmad</last></author>
      <author><first>Jianfeng</first><last>Chi</last></author>
      <author><first>Yuan</first><last>Tian</last></author>
      <author><first>Kai-Wei</first><last>Chang</last></author>
      <pages>743–749</pages>
      <abstract>Privacy policy documents are long and verbose. A question answering (QA) system can assist users in finding the information that is relevant and important to them. Prior studies in this domain frame the QA task as retrieving the most relevant text segment or a list of sentences from the policy document given a question. On the contrary, we argue that providing users with a short text span from policy documents reduces the burden of searching the target information from a lengthy text segment. In this paper, we present PolicyQA, a dataset that contains 25,017 reading comprehension style examples curated from an existing corpus of 115 website privacy policies. PolicyQA provides 714 human-annotated questions written for a wide range of privacy practices. We evaluate two existing neural QA models and perform rigorous analysis to reveal the advantages and challenges offered by PolicyQA.</abstract>
      <url hash="1c42141d">2020.findings-emnlp.66</url>
      <doi>10.18653/v1/2020.findings-emnlp.66</doi>
    </paper>
    <paper id="67">
      <title>A Linguistic Analysis of Visually Grounded Dialogues Based on Spatial Expressions</title>
      <author><first>Takuma</first><last>Udagawa</last></author>
      <author><first>Takato</first><last>Yamazaki</last></author>
      <author><first>Akiko</first><last>Aizawa</last></author>
      <pages>750–765</pages>
      <abstract>Recent models achieve promising results in visually grounded dialogues. However, existing datasets often contain undesirable biases and lack sophisticated linguistic analyses, which make it difficult to understand how well current models recognize their precise linguistic structures. To address this problem, we make two design choices: first, we focus on OneCommon Corpus (CITATION), a simple yet challenging common grounding dataset which contains minimal bias by design. Second, we analyze their linguistic structures based on spatial expressions and provide comprehensive and reliable annotation for 600 dialogues. We show that our annotation captures important linguistic structures including predicate-argument structure, modification and ellipsis. In our experiments, we assess the model’s understanding of these structures through reference resolution. We demonstrate that our annotation can reveal both the strengths and weaknesses of baseline models in essential levels of detail. Overall, we propose a novel framework and resource for investigating fine-grained language understanding in visually grounded dialogues.</abstract>
      <url hash="62e260d2">2020.findings-emnlp.67</url>
      <doi>10.18653/v1/2020.findings-emnlp.67</doi>
    </paper>
    <paper id="68">
      <title>Efficient Context and Schema Fusion Networks for Multi-Domain Dialogue State Tracking</title>
      <author><first>Su</first><last>Zhu</last></author>
      <author><first>Jieyu</first><last>Li</last></author>
      <author><first>Lu</first><last>Chen</last></author>
      <author><first>Kai</first><last>Yu</last></author>
      <pages>766–781</pages>
      <abstract>Dialogue state tracking (DST) aims at estimating the current dialogue state given all the preceding conversation. For multi-domain DST, the data sparsity problem is a major obstacle due to increased numbers of state candidates and dialogue lengths. To encode the dialogue context efficiently, we utilize the previous dialogue state (predicted) and the current dialogue utterance as the input for DST. To consider relations among different domain-slots, the schema graph involving prior knowledge is exploited. In this paper, a novel context and schema fusion network is proposed to encode the dialogue context and schema graph by using internal and external attention mechanisms. Experiment results show that our approach can outperform strong baselines, and the previous state-of-the-art method (SOM-DST) can also be improved by our proposed schema graph.</abstract>
      <url hash="b1bfb31c">2020.findings-emnlp.68</url>
      <doi>10.18653/v1/2020.findings-emnlp.68</doi>
    </paper>
    <paper id="69">
      <title>Syntactic and Semantic-driven Learning for Open Information Extraction</title>
      <author><first>Jialong</first><last>Tang</last></author>
      <author><first>Yaojie</first><last>Lu</last></author>
      <author><first>Hongyu</first><last>Lin</last></author>
      <author><first>Xianpei</first><last>Han</last></author>
      <author><first>Le</first><last>Sun</last></author>
      <author><first>Xinyan</first><last>Xiao</last></author>
      <author><first>Hua</first><last>Wu</last></author>
      <pages>782–792</pages>
      <abstract>One of the biggest bottlenecks in building accurate, high coverage neural open IE systems is the need for large labelled corpora. The diversity of open domain corpora and the variety of natural language expressions further exacerbate this problem. In this paper, we propose a syntactic and semantic-driven learning approach, which can learn neural open IE models without any human-labelled data by leveraging syntactic and semantic knowledge as noisier, higher-level supervision. Specifically, we first employ syntactic patterns as data labelling functions and pretrain a base model using the generated labels. Then we propose a syntactic and semantic-driven reinforcement learning algorithm, which can effectively generalize the base model to open situations with high accuracy. Experimental results show that our approach significantly outperforms the supervised counterparts, and can even achieve competitive performance to supervised state-of-the-art (SoA) model.</abstract>
      <url hash="46f1ab30">2020.findings-emnlp.69</url>
      <attachment type="OptionalSupplementaryMaterial" hash="8f579c68">2020.findings-emnlp.69.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.69</doi>
    </paper>
    <paper id="70">
      <title>Group-wise Contrastive Learning for Neural Dialogue Generation</title>
      <author><first>Hengyi</first><last>Cai</last></author>
      <author><first>Hongshen</first><last>Chen</last></author>
      <author><first>Yonghao</first><last>Song</last></author>
      <author><first>Zhuoye</first><last>Ding</last></author>
      <author><first>Yongjun</first><last>Bao</last></author>
      <author><first>Weipeng</first><last>Yan</last></author>
      <author><first>Xiaofang</first><last>Zhao</last></author>
      <pages>793–802</pages>
      <abstract>Neural dialogue response generation has gained much popularity in recent years. Maximum Likelihood Estimation (MLE) objective is widely adopted in existing dialogue model learning. However, models trained with MLE objective function are plagued by the low-diversity issue when it comes to the open-domain conversational setting. Inspired by the observation that humans not only learn from the positive signals but also benefit from correcting behaviors of undesirable actions, in this work, we introduce contrastive learning into dialogue generation, where the model explicitly perceives the difference between the well-chosen positive and negative utterances. Specifically, we employ a pretrained baseline model as a reference. During contrastive learning, the target dialogue model is trained to give higher conditional probabilities for the positive samples, and lower conditional probabilities for those negative samples, compared to the reference model. To manage the multi-mapping relations prevalent in human conversation, we augment contrastive dialogue learning with group-wise dual sampling. Extensive experimental results show that the proposed group-wise contrastive learning framework is suited for training a wide range of neural dialogue generation models with very favorable performance over the baseline training approaches.</abstract>
      <url hash="83a839a7">2020.findings-emnlp.70</url>
      <doi>10.18653/v1/2020.findings-emnlp.70</doi>
    </paper>
    <paper id="71">
      <title><fixed-case>E</fixed-case>-<fixed-case>BERT</fixed-case>: Efficient-Yet-Effective Entity Embeddings for <fixed-case>BERT</fixed-case></title>
      <author><first>Nina</first><last>Poerner</last></author>
      <author><first>Ulli</first><last>Waltinger</last></author>
      <author><first>Hinrich</first><last>Schütze</last></author>
      <pages>803–818</pages>
      <abstract>We present a novel way of injecting factual knowledge about entities into the pretrained BERT model (Devlin et al., 2019): We align Wikipedia2Vec entity vectors (Yamada et al., 2016) with BERT’s native wordpiece vector space and use the aligned entity vectors as if they were wordpiece vectors. The resulting entity-enhanced version of BERT (called E-BERT) is similar in spirit to ERNIE (Zhang et al., 2019) and KnowBert (Peters et al., 2019), but it requires no expensive further pre-training of the BERT encoder. We evaluate E-BERT on unsupervised question answering (QA), supervised relation classification (RC) and entity linking (EL). On all three tasks, E-BERT outperforms BERT and other baselines. We also show quantitatively that the original BERT model is overly reliant on the surface form of entity names (e.g., guessing that someone with an Italian-sounding name speaks Italian), and that E-BERT mitigates this problem.</abstract>
      <url hash="7a39c8e7">2020.findings-emnlp.71</url>
      <attachment type="OptionalSupplementaryMaterial" hash="7801a05a">2020.findings-emnlp.71.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.71</doi>
    </paper>
    <paper id="72">
      <title>A Multi-task Learning Framework for Opinion Triplet Extraction</title>
      <author><first>Chen</first><last>Zhang</last></author>
      <author><first>Qiuchi</first><last>Li</last></author>
      <author><first>Dawei</first><last>Song</last></author>
      <author><first>Benyou</first><last>Wang</last></author>
      <pages>819–828</pages>
      <abstract>The state-of-the-art Aspect-based Sentiment Analysis (ABSA) approaches are mainly based on either detecting aspect terms and their corresponding sentiment polarities, or co-extracting aspect and opinion terms. However, the extraction of aspect-sentiment pairs lacks opinion terms as a reference, while co-extraction of aspect and opinion terms would not lead to meaningful pairs without determining their sentiment dependencies. To address the issue, we present a novel view of ABSA as an opinion triplet extraction task, and propose a multi-task learning framework to jointly extract aspect terms and opinion terms, and simultaneously parses sentiment dependencies between them with a biaffine scorer. At inference phase, the extraction of triplets is facilitated by a triplet decoding method based on the above outputs. We evaluate the proposed framework on four SemEval benchmarks for ASBA. The results demonstrate that our approach significantly outperforms a range of strong baselines and state-of-the-art approaches.</abstract>
      <url hash="5fe04480">2020.findings-emnlp.72</url>
      <doi>10.18653/v1/2020.findings-emnlp.72</doi>
    </paper>
    <paper id="73">
      <title>Event Extraction as Multi-turn Question Answering</title>
      <author><first>Fayuan</first><last>Li</last></author>
      <author><first>Weihua</first><last>Peng</last></author>
      <author><first>Yuguang</first><last>Chen</last></author>
      <author><first>Quan</first><last>Wang</last></author>
      <author><first>Lu</first><last>Pan</last></author>
      <author><first>Yajuan</first><last>Lyu</last></author>
      <author><first>Yong</first><last>Zhu</last></author>
      <pages>829–838</pages>
      <abstract>Event extraction, which aims to identify event triggers of pre-defined event types and their arguments of specific roles, is a challenging task in NLP. Most traditional approaches formulate this task as classification problems, with event types or argument roles taken as golden labels. Such approaches fail to model rich interactions among event types and arguments of different roles, and cannot generalize to new types or roles. This work proposes a new paradigm that formulates event extraction as multi-turn question answering. Our approach, MQAEE, casts the extraction task into a series of reading comprehension problems, by which it extracts triggers and arguments successively from a given sentence. A history answer embedding strategy is further adopted to model question answering history in the multi-turn process. By this new formulation, MQAEE makes full use of dependency among arguments and event types, and generalizes well to new types with new argument roles. Empirical results on ACE 2005 shows that MQAEE outperforms current state-of-the-art, pushing the final F1 of argument extraction to 53.4% (+2.0%). And it also has a good generalization ability, achieving competitive performance on 13 new event types even if trained only with a few samples of them.</abstract>
      <url hash="41e66a9a">2020.findings-emnlp.73</url>
      <doi>10.18653/v1/2020.findings-emnlp.73</doi>
    </paper>
    <paper id="74">
      <title>Improving <fixed-case>QA</fixed-case> Generalization by Concurrent Modeling of Multiple Biases</title>
      <author><first>Mingzhu</first><last>Wu</last></author>
      <author><first>Nafise Sadat</first><last>Moosavi</last></author>
      <author><first>Andreas</first><last>Rücklé</last></author>
      <author><first>Iryna</first><last>Gurevych</last></author>
      <pages>839–853</pages>
      <abstract>Existing NLP datasets contain various biases that models can easily exploit to achieve high performances on the corresponding evaluation sets. However, focusing on dataset-specific biases limits their ability to learn more generalizable knowledge about the task from more general data patterns. In this paper, we investigate the impact of debiasing methods for improving generalization and propose a general framework for improving the performance on both in-domain and out-of-domain datasets by concurrent modeling of multiple biases in the training data. Our framework weights each example based on the biases it contains and the strength of those biases in the training data. It then uses these weights in the training objective so that the model relies less on examples with high bias weights. We extensively evaluate our framework on extractive question answering with training data from various domains with multiple biases of different strengths. We perform the evaluations in two different settings, in which the model is trained on a single domain or multiple domains simultaneously, and show its effectiveness in both settings compared to state-of-the-art debiasing methods.</abstract>
      <url hash="c2757a73">2020.findings-emnlp.74</url>
      <doi>10.18653/v1/2020.findings-emnlp.74</doi>
    </paper>
    <paper id="75">
      <title>Actor-Double-Critic: Incorporating Model-Based Critic for Task-Oriented Dialogue Systems</title>
      <author><first>Yen-chen</first><last>Wu</last></author>
      <author><first>Bo-Hsiang</first><last>Tseng</last></author>
      <author><first>Milica</first><last>Gasic</last></author>
      <pages>854–863</pages>
      <abstract>In order to improve the sample-efficiency of deep reinforcement learning (DRL), we implemented imagination augmented agent (I2A) in spoken dialogue systems (SDS). Although I2A achieves a higher success rate than baselines by augmenting predicted future into a policy network, its complicated architecture introduces unwanted instability. In this work, we propose actor-double-critic (ADC) to improve the stability and overall performance of I2A. ADC simplifies the architecture of I2A to reduce excessive parameters and hyper-parameters. More importantly, a separate model-based critic shares parameters between actions and makes back-propagation explicit. In our experiments on Cambridge Restaurant Booking task, ADC enhances success rates considerably and shows robustness to imperfect environment models. In addition, ADC exhibits the stability and sample-efficiency as significantly reducing the baseline standard deviation of success rates and reaching the 80% success rate with half training data.</abstract>
      <url hash="0ea27564">2020.findings-emnlp.75</url>
      <doi>10.18653/v1/2020.findings-emnlp.75</doi>
    </paper>
    <paper id="76">
      <title>Controlled Hallucinations: Learning to Generate Faithfully from Noisy Data</title>
      <author><first>Katja</first><last>Filippova</last></author>
      <pages>864–870</pages>
      <abstract>Neural text generation (data- or text-to-text) demonstrates remarkable performance when training data is abundant which for many applications is not the case. To collect a large corpus of parallel data, heuristic rules are often used but they inevitably let noise into the data, such as phrases in the output which cannot be explained by the input. Consequently, models pick up on the noise and may hallucinate–generate fluent but unsupported text. Our contribution is a simple but powerful technique to treat such hallucinations as a controllable aspect of the generated text, without dismissing any input and without modifying the model architecture. On the WikiBio corpus (Lebret et al., 2016), a particularly noisy dataset, we demonstrate the efficacy of the technique both in an automatic and in a human evaluation.</abstract>
      <url hash="1c6cbc62">2020.findings-emnlp.76</url>
      <doi>10.18653/v1/2020.findings-emnlp.76</doi>
    </paper>
    <paper id="77">
      <title>Sequential Span Classification with Neural Semi-<fixed-case>M</fixed-case>arkov <fixed-case>CRF</fixed-case>s for Biomedical Abstracts</title>
      <author><first>Kosuke</first><last>Yamada</last></author>
      <author><first>Tsutomu</first><last>Hirao</last></author>
      <author><first>Ryohei</first><last>Sasano</last></author>
      <author><first>Koichi</first><last>Takeda</last></author>
      <author><first>Masaaki</first><last>Nagata</last></author>
      <pages>871–877</pages>
      <abstract>Dividing biomedical abstracts into several segments with rhetorical roles is essential for supporting researchers’ information access in the biomedical domain. Conventional methods have regarded the task as a sequence labeling task based on sequential sentence classification, i.e., they assign a rhetorical label to each sentence by considering the context in the abstract. However, these methods have a critical problem: they are prone to mislabel longer continuous sentences with the same rhetorical label. To tackle the problem, we propose sequential span classification that assigns a rhetorical label, not to a single sentence but to a span that consists of continuous sentences. Accordingly, we introduce Neural Semi-Markov Conditional Random Fields to assign the labels to such spans by considering all possible spans of various lengths. Experimental results obtained from PubMed 20k RCT and NICTA-PIBOSO datasets demonstrate that our proposed method achieved the best micro sentence-F1 score as well as the best micro span-F1 score.</abstract>
      <url hash="6f785f1b">2020.findings-emnlp.77</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f0b89ff6">2020.findings-emnlp.77.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.77</doi>
    </paper>
    <paper id="78">
      <title>Where to Submit? Helping Researchers to Choose the Right Venue</title>
      <author><first>Konstantin</first><last>Kobs</last></author>
      <author><first>Tobias</first><last>Koopmann</last></author>
      <author><first>Albin</first><last>Zehe</last></author>
      <author><first>David</first><last>Fernes</last></author>
      <author><first>Philipp</first><last>Krop</last></author>
      <author><first>Andreas</first><last>Hotho</last></author>
      <pages>878–883</pages>
      <abstract>Whenever researchers write a paper, the same question occurs: “Where to submit?” In this work, we introduce WTS, an open and interpretable NLP system that recommends conferences and journals to researchers based on the title, abstract, and/or keywords of a given paper. We adapt the TextCNN architecture and automatically analyze its predictions using the Integrated Gradients method to highlight words and phrases that led to the recommendation of a scientific venue. We train and test our method on publications from the fields of artificial intelligence (AI) and medicine, both derived from the Semantic Scholar dataset. WTS achieves an Accuracy@5 of approximately 83% for AI papers and 95% in the field of medicine. It is open source and available for testing on https://wheretosubmit.ml.</abstract>
      <url hash="48f853d2">2020.findings-emnlp.78</url>
      <doi>10.18653/v1/2020.findings-emnlp.78</doi>
    </paper>
    <paper id="79">
      <title><fixed-case>A</fixed-case>ir<fixed-case>C</fixed-case>oncierge: Generating Task-Oriented Dialogue via Efficient Large-Scale Knowledge Retrieval</title>
      <author><first>Chieh-Yang</first><last>Chen</last></author>
      <author><first>Pei-Hsin</first><last>Wang</last></author>
      <author><first>Shih-Chieh</first><last>Chang</last></author>
      <author><first>Da-Cheng</first><last>Juan</last></author>
      <author><first>Wei</first><last>Wei</last></author>
      <author><first>Jia-Yu</first><last>Pan</last></author>
      <pages>884–897</pages>
      <abstract>Despite recent success in neural task-oriented dialogue systems, developing such a real-world system involves accessing large-scale knowledge bases (KBs), which cannot be simply encoded by neural approaches, such as memory network mechanisms. To alleviate the above problem, we propose , an end-to-end trainable text-to-SQL guided framework to learn a neural agent that interacts with KBs using the generated SQL queries. Specifically, the neural agent first learns to ask and confirm the customer’s intent during the multi-turn interactions, then dynamically determining when to ground the user constraints into executable SQL queries so as to fetch relevant information from KBs. With the help of our method, the agent can use less but more accurate fetched results to generate useful responses efficiently, instead of incorporating the entire KBs. We evaluate the proposed method on the AirDialogue dataset, a large corpus released by Google, containing the conversations of customers booking flight tickets from the agent. The experimental results show that significantly improves over previous work in terms of accuracy and the BLEU score, which demonstrates not only the ability to achieve the given task but also the good quality of the generated dialogues.</abstract>
      <url hash="adb265f0">2020.findings-emnlp.79</url>
      <attachment type="OptionalSupplementaryMaterial" hash="2d973a01">2020.findings-emnlp.79.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.79</doi>
    </paper>
    <paper id="80">
      <title><fixed-case>D</fixed-case>oc<fixed-case>S</fixed-case>truct: A Multimodal Method to Extract Hierarchy Structure in Document for General Form Understanding</title>
      <author><first>Zilong</first><last>Wang</last></author>
      <author><first>Mingjie</first><last>Zhan</last></author>
      <author><first>Xuebo</first><last>Liu</last></author>
      <author><first>Ding</first><last>Liang</last></author>
      <pages>898–908</pages>
      <abstract>Form understanding depends on both textual contents and organizational structure. Although modern OCR performs well, it is still challenging to realize general form understanding because forms are commonly used and of various formats. The table detection and handcrafted features in previous works cannot apply to all forms because of their requirements on formats. Therefore, we concentrate on the most elementary components, the key-value pairs, and adopt multimodal methods to extract features. We consider the form structure as a tree-like or graph-like hierarchy of text fragments. The parent-child relation corresponds to the key-value pairs in forms. We utilize the state-of-the-art models and design targeted extraction modules to extract multimodal features from semantic contents, layout information, and visual images. A hybrid fusion method of concatenation and feature shifting is designed to fuse the heterogeneous features and provide an informative joint representation. We adopt an asymmetric algorithm and negative sampling in our model as well. We validate our method on two benchmarks, MedForm and FUNSD, and extensive experiments demonstrate the effectiveness of our method.</abstract>
      <url hash="b8069edb">2020.findings-emnlp.80</url>
      <doi>10.18653/v1/2020.findings-emnlp.80</doi>
    </paper>
    <paper id="81">
      <title>Pretrained Language Models for Dialogue Generation with Multiple Input Sources</title>
      <author><first>Yu</first><last>Cao</last></author>
      <author><first>Wei</first><last>Bi</last></author>
      <author><first>Meng</first><last>Fang</last></author>
      <author><first>Dacheng</first><last>Tao</last></author>
      <pages>909–917</pages>
      <abstract>Large-scale pretrained language models have achieved outstanding performance on natural language understanding tasks. However, it is still under investigating how to apply them to dialogue generation tasks, especially those with responses conditioned on multiple sources. Previous work simply concatenates all input sources or averages information from different input sources. In this work, we study dialogue models with multiple input sources adapted from the pretrained language model GPT2. We explore various methods to fuse multiple separate attention information corresponding to different sources. Our experimental results show that proper fusion methods deliver higher relevance with dialogue history than simple fusion baselines.</abstract>
      <url hash="6941e414">2020.findings-emnlp.81</url>
      <doi>10.18653/v1/2020.findings-emnlp.81</doi>
    </paper>
    <paper id="82">
      <title>A Study in Improving <fixed-case>BLEU</fixed-case> Reference Coverage with Diverse Automatic Paraphrasing</title>
      <author><first>Rachel</first><last>Bawden</last></author>
      <author><first>Biao</first><last>Zhang</last></author>
      <author><first>Lisa</first><last>Yankovskaya</last></author>
      <author><first>Andre</first><last>Tättar</last></author>
      <author><first>Matt</first><last>Post</last></author>
      <pages>918–932</pages>
      <abstract>We investigate a long-perceived shortcoming in the typical use of BLEU: its reliance on a single reference. Using modern neural paraphrasing techniques, we study whether automatically generating additional *diverse* references can provide better coverage of the space of valid translations and thereby improve its correlation with human judgments. Our experiments on the into-English language directions of the WMT19 metrics task (at both the system and sentence level) show that using paraphrased references does generally improve BLEU, and when it does, the more diverse the better. However, we also show that better results could be achieved if those paraphrases were to specifically target the parts of the space most relevant to the MT outputs being evaluated. Moreover, the gains remain slight even when human paraphrases are used, suggesting inherent limitations to BLEU’s capacity to correctly exploit multiple references. Surprisingly, we also find that adequacy appears to be less important, as shown by the high results of a strong sampling approach, which even beats human paraphrases when used with sentence-level BLEU.</abstract>
      <url hash="b2e17989">2020.findings-emnlp.82</url>
      <attachment type="OptionalSupplementaryMaterial" hash="8410870d">2020.findings-emnlp.82.OptionalSupplementaryMaterial.txt</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.82</doi>
    </paper>
    <paper id="83">
      <title>Cross-lingual Alignment Methods for Multilingual <fixed-case>BERT</fixed-case>: A Comparative Study</title>
      <author><first>Saurabh</first><last>Kulshreshtha</last></author>
      <author><first>Jose Luis</first><last>Redondo Garcia</last></author>
      <author><first>Ching-Yun</first><last>Chang</last></author>
      <pages>933–942</pages>
      <abstract>Multilingual BERT (mBERT) has shown reasonable capability for zero-shot cross-lingual transfer when fine-tuned on downstream tasks. Since mBERT is not pre-trained with explicit cross-lingual supervision, transfer performance can further be improved by aligning mBERT with cross-lingual signal. Prior work propose several approaches to align contextualised embeddings. In this paper we analyse how different forms of cross-lingual supervision and various alignment methods influence the transfer capability of mBERT in zero-shot setting. Specifically, we compare parallel corpora vs dictionary-based supervision and rotational vs fine-tuning based alignment methods. We evaluate the performance of different alignment methodologies across eight languages on two tasks: Name Entity Recognition and Semantic Slot Filling. In addition, we propose a novel normalisation method which consistently improves the performance of rotation-based alignment including a notable 3% F1 improvement for distant and typologically dissimilar languages. Importantly we identify the biases of the alignment methods to the type of task and proximity to the transfer language. We also find that supervision from parallel corpus is generally superior to dictionary alignments.</abstract>
      <url hash="67a4d950">2020.findings-emnlp.83</url>
      <doi>10.18653/v1/2020.findings-emnlp.83</doi>
    </paper>
    <paper id="84">
      <title>Hybrid Emoji-Based Masked Language Models for Zero-Shot Abusive Language Detection</title>
      <author><first>Michele</first><last>Corazza</last></author>
      <author><first>Stefano</first><last>Menini</last></author>
      <author><first>Elena</first><last>Cabrio</last></author>
      <author><first>Sara</first><last>Tonelli</last></author>
      <author><first>Serena</first><last>Villata</last></author>
      <pages>943–949</pages>
      <abstract>Recent studies have demonstrated the effectiveness of cross-lingual language model pre-training on different NLP tasks, such as natural language inference and machine translation. In our work, we test this approach on social media data, which are particularly challenging to process within this framework, since the limited length of the textual messages and the irregularity of the language make it harder to learn meaningful encodings. More specifically, we propose a hybrid emoji-based Masked Language Model (MLM) to leverage the common information conveyed by emojis across different languages and improve the learned cross-lingual representation of short text messages, with the goal to perform zero- shot abusive language detection. We compare the results obtained with the original MLM to the ones obtained by our method, showing improved performance on German, Italian and Spanish.</abstract>
      <url hash="a578af0f">2020.findings-emnlp.84</url>
      <doi>10.18653/v1/2020.findings-emnlp.84</doi>
    </paper>
    <paper id="85">
      <title><fixed-case>S</fixed-case>e<fixed-case>N</fixed-case>s<fixed-case>ER</fixed-case>: Learning Cross-Building Sensor Metadata Tagger</title>
      <author><first>Yang</first><last>Jiao</last></author>
      <author><first>Jiacheng</first><last>Li</last></author>
      <author><first>Jiaman</first><last>Wu</last></author>
      <author><first>Dezhi</first><last>Hong</last></author>
      <author><first>Rajesh</first><last>Gupta</last></author>
      <author><first>Jingbo</first><last>Shang</last></author>
      <pages>950–960</pages>
      <abstract>Sensor metadata tagging, akin to the named entity recognition task, provides key contextual information (e.g., measurement type and location) about sensors for running smart building applications. Unfortunately, sensor metadata in different buildings often follows distinct naming conventions. Therefore, learning a tagger currently requires extensive annotations on a per building basis. In this work, we propose a novel framework, SeNsER, which learns a sensor metadata tagger for a new building based on its raw metadata and some existing fully annotated building. It leverages the commonality between different buildings: At the character level, it employs bidirectional neural language models to capture the shared underlying patterns between two buildings and thus regularizes the feature learning process; At the word level, it leverages as features the k-mers existing in the fully annotated building. During inference, we further incorporate the information obtained from sources such as Wikipedia as prior knowledge. As a result, SeNsER shows promising results in extensive experiments on multiple real-world buildings.</abstract>
      <url hash="1885d0f5">2020.findings-emnlp.85</url>
      <doi>10.18653/v1/2020.findings-emnlp.85</doi>
    </paper>
    <paper id="86">
      <title><fixed-case>P</fixed-case>ersian Ezafe Recognition Using Transformers and Its Role in Part-Of-Speech Tagging</title>
      <author><first>Ehsan</first><last>Doostmohammadi</last></author>
      <author><first>Minoo</first><last>Nassajian</last></author>
      <author><first>Adel</first><last>Rahimi</last></author>
      <pages>961–971</pages>
      <abstract>Ezafe is a grammatical particle in some Iranian languages that links two words together. Regardless of the important information it conveys, it is almost always not indicated in Persian script, resulting in mistakes in reading complex sentences and errors in natural language processing tasks. In this paper, we experiment with different machine learning methods to achieve state-of-the-art results in the task of ezafe recognition. Transformer-based methods, BERT and XLMRoBERTa, achieve the best results, the latter achieving 2.68% F1-score more than the previous state-of-the-art. We, moreover, use ezafe information to improve Persian part-of-speech tagging results and show that such information will not be useful to transformer-based methods and explain why that might be the case.</abstract>
      <url hash="d774f4e7">2020.findings-emnlp.86</url>
      <doi>10.18653/v1/2020.findings-emnlp.86</doi>
    </paper>
    <paper id="87">
      <title>Scene Graph Modification Based on Natural Language Commands</title>
      <author><first>Xuanli</first><last>He</last></author>
      <author><first>Quan Hung</first><last>Tran</last></author>
      <author><first>Gholamreza</first><last>Haffari</last></author>
      <author><first>Walter</first><last>Chang</last></author>
      <author><first>Zhe</first><last>Lin</last></author>
      <author><first>Trung</first><last>Bui</last></author>
      <author><first>Franck</first><last>Dernoncourt</last></author>
      <author><first>Nhan</first><last>Dam</last></author>
      <pages>972–990</pages>
      <abstract>Structured representations like graphs and parse trees play a crucial role in many Natural Language Processing systems. In recent years, the advancements in multi-turn user interfaces necessitate the need for controlling and updating these structured representations given new sources of information. Although there have been many efforts focusing on improving the performance of the parsers that map text to graphs or parse trees, very few have explored the problem of directly manipulating these representations. In this paper, we explore the novel problem of graph modification, where the systems need to learn how to update an existing scene graph given a new user’s command. Our novel models based on graph-based sparse transformer and cross attention information fusion outperform previous systems adapted from the machine translation and graph generation literature. We further contribute our large graph modification datasets to the research community to encourage future research for this new problem.</abstract>
      <url hash="9a5660e5">2020.findings-emnlp.87</url>
      <doi>10.18653/v1/2020.findings-emnlp.87</doi>
    </paper>
    <paper id="88">
      <title><fixed-case>L</fixed-case>i<fixed-case>M</fixed-case>i<fixed-case>T</fixed-case>: The Literal Motion in Text Dataset</title>
      <author><first>Irene</first><last>Manotas</last></author>
      <author><first>Ngoc Phuoc An</first><last>Vo</last></author>
      <author><first>Vadim</first><last>Sheinin</last></author>
      <pages>991–1000</pages>
      <abstract>Motion recognition is one of the basic cognitive capabilities of many life forms, yet identifying motion of physical entities in natural language have not been explored extensively and empirically. We present the Literal-Motion-in-Text (LiMiT) dataset, a large human-annotated collection of English text sentences describing physical occurrence of motion, with annotated physical entities in motion. We describe the annotation process for the dataset, analyze its scale and diversity, and report results of several baseline models. We also present future research directions and applications of the LiMiT dataset and share it publicly as a new resource for the research community.</abstract>
      <url hash="b3b25a5d">2020.findings-emnlp.88</url>
      <doi>10.18653/v1/2020.findings-emnlp.88</doi>
    </paper>
    <paper id="89">
      <title>Transition-based Parsing with Stack-Transformers</title>
      <author><first>Ramón</first><last>Fernandez Astudillo</last></author>
      <author><first>Miguel</first><last>Ballesteros</last></author>
      <author><first>Tahira</first><last>Naseem</last></author>
      <author><first>Austin</first><last>Blodgett</last></author>
      <author><first>Radu</first><last>Florian</last></author>
      <pages>1001–1007</pages>
      <abstract>Modeling the parser state is key to good performance in transition-based parsing. Recurrent Neural Networks considerably improved the performance of transition-based systems by modelling the global state, e.g. stack-LSTM parsers, or local state modeling of contextualized features, e.g. Bi-LSTM parsers. Given the success of Transformer architectures in recent parsing systems, this work explores modifications of the sequence-to-sequence Transformer architecture to model either global or local parser states in transition-based parsing. We show that modifications of the cross attention mechanism of the Transformer considerably strengthen performance both on dependency and Abstract Meaning Representation (AMR) parsing tasks, particularly for smaller models or limited training data.</abstract>
      <url hash="b94f3f65">2020.findings-emnlp.89</url>
      <doi>10.18653/v1/2020.findings-emnlp.89</doi>
    </paper>
    <paper id="90">
      <title>Generative Data Augmentation for Commonsense Reasoning</title>
      <author><first>Yiben</first><last>Yang</last></author>
      <author><first>Chaitanya</first><last>Malaviya</last></author>
      <author><first>Jared</first><last>Fernandez</last></author>
      <author><first>Swabha</first><last>Swayamdipta</last></author>
      <author><first>Ronan</first><last>Le Bras</last></author>
      <author><first>Ji-Ping</first><last>Wang</last></author>
      <author><first>Chandra</first><last>Bhagavatula</last></author>
      <author><first>Yejin</first><last>Choi</last></author>
      <author><first>Doug</first><last>Downey</last></author>
      <pages>1008–1025</pages>
      <abstract>Recent advances in commonsense reasoning depend on large-scale human-annotated training sets to achieve peak performance. However, manual curation of training sets is expensive and has been shown to introduce annotation artifacts that neural models can readily exploit and overfit to. We propose a novel generative data augmentation technique, G-DAUGˆC, that aims to achieve more accurate and robust learning in a low-resource setting. Our approach generates synthetic examples using pretrained language models and selects the most informative and diverse set of examples for data augmentation. On experiments with multiple commonsense reasoning benchmarks, G-DAUGˆC consistently outperforms existing data augmentation methods based on back-translation, establishing a new state-of-the-art on WinoGrande, CODAH, and CommonsenseQA, as well as enhances out-of-distribution generalization, proving to be robust against adversaries or perturbations. Our analysis demonstrates that G-DAUGˆC produces a diverse set of fluent training examples, and that its selection and training approaches are important for performance.</abstract>
      <url hash="1a2b24c2">2020.findings-emnlp.90</url>
      <attachment type="OptionalSupplementaryMaterial" hash="6448e5fb">2020.findings-emnlp.90.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.90</doi>
    </paper>
    <paper id="91">
      <title><fixed-case>H</fixed-case>ybrid<fixed-case>QA</fixed-case>: A Dataset of Multi-Hop Question Answering over Tabular and Textual Data</title>
      <author><first>Wenhu</first><last>Chen</last></author>
      <author><first>Hanwen</first><last>Zha</last></author>
      <author><first>Zhiyu</first><last>Chen</last></author>
      <author><first>Wenhan</first><last>Xiong</last></author>
      <author><first>Hong</first><last>Wang</last></author>
      <author><first>William Yang</first><last>Wang</last></author>
      <pages>1026–1036</pages>
      <abstract>Existing question answering datasets focus on dealing with homogeneous information, based either only on text or KB/Table information alone. However, as human knowledge is distributed over heterogeneous forms, using homogeneous information alone might lead to severe coverage problems. To fill in the gap, we present HybridQA, a new large-scale question-answering dataset that requires reasoning on heterogeneous information. Each question is aligned with a Wikipedia table and multiple free-form corpora linked with the entities in the table. The questions are designed to aggregate both tabular information and text information, i.e., lack of either form would render the question unanswerable. We test with three different models: 1) a table-only model. 2) text-only model. 3) a hybrid model that combines heterogeneous information to find the answer. The experimental results show that the EM scores obtained by two baselines are below 20%, while the hybrid model can achieve an EM over 40%. This gap suggests the necessity to aggregate heterogeneous information in HybridQA. However, the hybrid model’s score is still far behind human performance. Hence, HybridQA can serve as a challenging benchmark to study question answering with heterogeneous information.</abstract>
      <url hash="9367ef25">2020.findings-emnlp.91</url>
      <doi>10.18653/v1/2020.findings-emnlp.91</doi>
    </paper>
    <paper id="92">
      <title><fixed-case>P</fixed-case>ho<fixed-case>BERT</fixed-case>: Pre-trained language models for <fixed-case>V</fixed-case>ietnamese</title>
      <author><first>Dat Quoc</first><last>Nguyen</last></author>
      <author><first>Anh</first><last>Tuan Nguyen</last></author>
      <pages>1037–1042</pages>
      <abstract>We present PhoBERT with two versions, PhoBERT-base and PhoBERT-large, the first public large-scale monolingual language models pre-trained for Vietnamese. Experimental results show that PhoBERT consistently outperforms the recent best pre-trained multilingual model XLM-R (Conneau et al., 2020) and improves the state-of-the-art in multiple Vietnamese-specific NLP tasks including Part-of-speech tagging, Dependency parsing, Named-entity recognition and Natural language inference. We release PhoBERT to facilitate future research and downstream applications for Vietnamese NLP. Our PhoBERT models are available at https://github.com/VinAIResearch/PhoBERT</abstract>
      <url hash="5f3cdad6">2020.findings-emnlp.92</url>
      <doi>10.18653/v1/2020.findings-emnlp.92</doi>
    </paper>
    <paper id="93">
      <title><fixed-case>EST</fixed-case>e<fixed-case>R</fixed-case>: Combining Word Co-occurrences and Word Associations for Unsupervised Emotion Detection</title>
      <author><first>Sujatha Das</first><last>Gollapalli</last></author>
      <author><first>Polina</first><last>Rozenshtein</last></author>
      <author><first>See-Kiong</first><last>Ng</last></author>
      <pages>1043–1056</pages>
      <abstract>Accurate detection of emotions in user- generated text was shown to have several applications for e-commerce, public well-being, and disaster management. Currently, the state-of-the-art performance for emotion detection in text is obtained using complex, deep learning models trained on domain-specific, labeled data. In this paper, we propose ESTeR , an unsupervised model for identifying emotions using a novel similarity function based on random walks on graphs. Our model combines large-scale word co-occurrence information with word-associations from lexicons avoiding not only the dependence on labeled datasets, but also an explicit mapping of words to latent spaces used in emotion-enriched word embeddings. Our similarity function can also be computed efficiently. We study a range of datasets including recent tweets related to COVID-19 to illustrate the superior performance of our model and report insights on public emotions during the on-going pandemic.</abstract>
      <url hash="0f51135b">2020.findings-emnlp.93</url>
      <doi>10.18653/v1/2020.findings-emnlp.93</doi>
    </paper>
    <paper id="94">
      <title>Make Templates Smarter: A Template Based <fixed-case>D</fixed-case>ata2<fixed-case>T</fixed-case>ext System Powered by Text Stitch Model</title>
      <author><first>Bingfeng</first><last>Luo</last></author>
      <author><first>Zuo</first><last>Bai</last></author>
      <author><first>Kunfeng</first><last>Lai</last></author>
      <author><first>Jianping</first><last>Shen</last></author>
      <pages>1057–1062</pages>
      <abstract>Neural network (NN) based data2text models achieve state-of-the-art (SOTA) performance in most metrics, but they sometimes drop or modify the information in the input, and it is hard to control the generation contents. Moreover, it requires paired training data that are usually expensive to collect. Template-based methods have good fidelity and controllability but require heavy human involvement. We propose a novel template-based data2text system powered by a text stitch model. It ensures fidelity and controllability by using templates to produce the main contents. In addition, it reduces human involvement in template design by using a text stitch model to automatically stitch adjacent template units, which is a step that usually requires careful template design and limits template reusability. The text stitch model can be trained in self-supervised fashion, which only requires free texts. The experiments on a benchmark dataset show that our system outperforms SOTA NN-based systems in fidelity and surpasses template-based systems in diversity and human involvement.</abstract>
      <url hash="efc31e44">2020.findings-emnlp.94</url>
      <attachment type="OptionalSupplementaryMaterial" hash="add0a5bb">2020.findings-emnlp.94.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.94</doi>
    </paper>
    <paper id="95">
      <title><fixed-case>GCDST</fixed-case>: A Graph-based and Copy-augmented Multi-domain Dialogue State Tracking</title>
      <author><first>Peng</first><last>Wu</last></author>
      <author><first>Bowei</first><last>Zou</last></author>
      <author><first>Ridong</first><last>Jiang</last></author>
      <author><first>AiTi</first><last>Aw</last></author>
      <pages>1063–1073</pages>
      <abstract>As an essential component of task-oriented dialogue systems, Dialogue State Tracking (DST) takes charge of estimating user intentions and requests in dialogue contexts and extracting substantial goals (states) from user utterances to help the downstream modules to determine the next actions of dialogue systems. For practical usages, a major challenge to constructing a robust DST model is to process a conversation with multi-domain states. However, most existing approaches trained DST on a single domain independently, ignoring the information across domains. To tackle the multi-domain DST task, we first construct a dialogue state graph to transfer structured features among related domain-slot pairs across domains. Then, we encode the graph information of dialogue states by graph convolutional networks and utilize a hard copy mechanism to directly copy historical states from the previous conversation. Experimental results show that our model improves the performances of the multi-domain DST baseline (TRADE) with the absolute joint accuracy of 2.0% and 1.0% on the MultiWOZ 2.0 and 2.1 dialogue datasets, respectively.</abstract>
      <url hash="1e6c1ebf">2020.findings-emnlp.95</url>
      <doi>10.18653/v1/2020.findings-emnlp.95</doi>
    </paper>
    <paper id="96">
      <title>Incorporating Stylistic Lexical Preferences in Generative Language Models</title>
      <author><first>Hrituraj</first><last>Singh</last></author>
      <author><first>Gaurav</first><last>Verma</last></author>
      <author><first>Balaji Vasan</first><last>Srinivasan</last></author>
      <pages>1074–1079</pages>
      <abstract>While recent advances in language modeling has resulted in powerful generation models, their generation style remains implicitly dependent on the training data and can not emulate a specific target style. Leveraging the generative capabilities of a transformer-based language models, we present an approach to induce certain target-author attributes by incorporating continuous multi-dimensional lexical preferences of an author into generative language models. We introduce rewarding strategies in a reinforcement learning framework that encourages the use of words across multiple categorical dimensions, to varying extents. Our experiments demonstrate that the proposed approach can generate text that distinctively aligns with a given target author’s lexical style. We conduct quantitative and qualitative comparisons with competitive and relevant baselines to illustrate the benefits of the proposed approach.</abstract>
      <url hash="11e99eef">2020.findings-emnlp.96</url>
      <doi>10.18653/v1/2020.findings-emnlp.96</doi>
    </paper>
    <paper id="97">
      <title>Why do you think that? Exploring Faithful Sentence-Level Rationales Without Supervision</title>
      <author><first>Max</first><last>Glockner</last></author>
      <author><first>Ivan</first><last>Habernal</last></author>
      <author><first>Iryna</first><last>Gurevych</last></author>
      <pages>1080–1095</pages>
      <abstract>Evaluating the trustworthiness of a model’s prediction is essential for differentiating between ‘right for the right reasons’ and ‘right for the wrong reasons’. Identifying textual spans that determine the target label, known as faithful rationales, usually relies on pipeline approaches or reinforcement learning. However, such methods either require supervision and thus costly annotation of the rationales or employ non-differentiable models. We propose a differentiable training–framework to create models which output faithful rationales on a sentence level, by solely applying supervision on the target task. To achieve this, our model solves the task based on each rationale individually and learns to assign high scores to those which solved the task best. Our evaluation on three different datasets shows competitive results compared to a standard BERT blackbox while exceeding a pipeline counterpart’s performance in two cases. We further exploit the transparent decision–making process of these models to prefer selecting the correct rationales by applying direct supervision, thereby boosting the performance on the rationale–level.</abstract>
      <url hash="d609c798">2020.findings-emnlp.97</url>
      <doi>10.18653/v1/2020.findings-emnlp.97</doi>
    </paper>
    <paper id="98">
      <title>Semi-Supervised Learning for Video Captioning</title>
      <author><first>Ke</first><last>Lin</last></author>
      <author><first>Zhuoxin</first><last>Gan</last></author>
      <author><first>Liwei</first><last>Wang</last></author>
      <pages>1096–1106</pages>
      <abstract>Deep neural networks have made great success on video captioning in supervised learning setting. However, annotating videos with descriptions is very expensive and time-consuming. If the video captioning algorithm can benefit from a large number of unlabeled videos, the cost of annotation can be reduced. In the proposed study, we make the first attempt to train the video captioning model on labeled data and unlabeled data jointly, in a semi-supervised learning manner. For labeled data, we train them with the traditional cross-entropy loss. For unlabeled data, we leverage a self-critical policy gradient method with the difference between the scores obtained by Monte-Carlo sampling and greedy decoding as the reward function, while the scores are the negative K-L divergence between output distributions of original video data and augmented video data. The final loss is the weighted sum of losses obtained by labeled data and unlabeled data. Experiments conducted on VATEX, MSR-VTT and MSVD dataset demonstrate that the introduction of unlabeled data can improve the performance of the video captioning model. The proposed semi-supervised learning algorithm also outperforms several state-of-the-art semi-supervised learning approaches.</abstract>
      <url hash="b29ff23f">2020.findings-emnlp.98</url>
      <doi>10.18653/v1/2020.findings-emnlp.98</doi>
    </paper>
    <paper id="99">
      <title><fixed-case>M</fixed-case>ultiˆ2<fixed-case>OIE</fixed-case>: Multilingual Open Information Extraction Based on Multi-Head Attention with <fixed-case>BERT</fixed-case></title>
      <author><first>Youngbin</first><last>Ro</last></author>
      <author><first>Yukyung</first><last>Lee</last></author>
      <author><first>Pilsung</first><last>Kang</last></author>
      <pages>1107–1117</pages>
      <abstract>In this paper, we propose Multi<tex-math>^2</tex-math>OIE, which performs open information extraction (open IE) by combining BERT with multi-head attention. Our model is a sequence-labeling system with an efficient and effective argument extraction method. We use a query, key, and value setting inspired by the Multimodal Transformer to replace the previously used bidirectional long short-term memory architecture with multi-head attention. Multi<tex-math>^2</tex-math>OIE outperforms existing sequence-labeling systems with high computational efficiency on two benchmark evaluation datasets, Re-OIE2016 and CaRB. Additionally, we apply the proposed method to multilingual open IE using multilingual BERT. Experimental results on new benchmark datasets introduced for two languages (Spanish and Portuguese) demonstrate that our model outperforms other multilingual systems without training data for the target languages.</abstract>
      <url hash="a4554a85">2020.findings-emnlp.99</url>
      <attachment type="OptionalSupplementaryMaterial" hash="60be47e0">2020.findings-emnlp.99.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.99</doi>
    </paper>
    <paper id="100">
      <title><fixed-case>LGPS</fixed-case>olver - Solving Logic Grid Puzzles Automatically</title>
      <author><first>Elgun</first><last>Jabrayilzade</last></author>
      <author><first>Selma</first><last>Tekir</last></author>
      <pages>1118–1123</pages>
      <abstract>Logic grid puzzle (LGP) is a type of word problem where the task is to solve a problem in logic. Constraints for the problem are given in the form of textual clues. Once these clues are transformed into formal logic, a deductive reasoning process provides the solution. Solving logic grid puzzles in a fully automatic manner has been a challenge since a precise understanding of clues is necessary to develop the corresponding formal logic representation. To meet this challenge, we propose a solution that uses a DistilBERT-based classifier to classify a clue into one of the predefined predicate types for logic grid puzzles. Another novelty of the proposed solution is the recognition of comparison structures in clues. By collecting comparative adjectives from existing dictionaries and utilizing a semantic framework to catch comparative quantifiers, the semantics of clues concerning comparison structures are better understood, ensuring conversion to correct logic representation. Our approach solves logic grid puzzles in a fully automated manner with 100% accuracy on the given puzzle datasets and outperforms state-of-the-art solutions by a large margin.</abstract>
      <url hash="8a8400be">2020.findings-emnlp.100</url>
      <doi>10.18653/v1/2020.findings-emnlp.100</doi>
    </paper>
    <paper id="101">
      <title>Using the Past Knowledge to Improve Sentiment Classification</title>
      <author><first>Qi</first><last>Qin</last></author>
      <author><first>Wenpeng</first><last>Hu</last></author>
      <author><first>Bing</first><last>Liu</last></author>
      <pages>1124–1133</pages>
      <abstract>This paper studies sentiment classification in the lifelong learning setting that incrementally learns a sequence of sentiment classification tasks. It proposes a new lifelong learning model (called L2PG) that can retain and selectively transfer the knowledge learned in the past to help learn the new task. A key innovation of this proposed model is a novel parameter-gate (p-gate) mechanism that regulates the flow or transfer of the previously learned knowledge to the new task. Specifically, it can selectively use the network parameters (which represent the retained knowledge gained from the previous tasks) to assist the learning of the new task t. Knowledge distillation is also employed in the process to preserve the past knowledge by approximating the network output at the state when task t-1 was learned. Experimental results show that L2PG outperforms strong baselines, including even multiple task learning.</abstract>
      <url hash="f4ca3f3d">2020.findings-emnlp.101</url>
      <doi>10.18653/v1/2020.findings-emnlp.101</doi>
    </paper>
    <paper id="102">
      <title>High-order Semantic Role Labeling</title>
      <author><first>Zuchao</first><last>Li</last></author>
      <author><first>Hai</first><last>Zhao</last></author>
      <author><first>Rui</first><last>Wang</last></author>
      <author><first>Kevin</first><last>Parnow</last></author>
      <pages>1134–1151</pages>
      <abstract>Semantic role labeling is primarily used to identify predicates, arguments, and their semantic relationships. Due to the limitations of modeling methods and the conditions of pre-identified predicates, previous work has focused on the relationships between predicates and arguments and the correlations between arguments at most, while the correlations between predicates have been neglected for a long time. High-order features and structure learning were very common in modeling such correlations before the neural network era. In this paper, we introduce a high-order graph structure for the neural semantic role labeling model, which enables the model to explicitly consider not only the isolated predicate-argument pairs but also the interaction between the predicate-argument pairs. Experimental results on 7 languages of the CoNLL-2009 benchmark show that the high-order structural learning techniques are beneficial to the strong performing SRL models and further boost our baseline to achieve new state-of-the-art results.</abstract>
      <url hash="bb69b254">2020.findings-emnlp.102</url>
      <attachment type="OptionalSupplementaryMaterial" hash="619d45f8">2020.findings-emnlp.102.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.102</doi>
    </paper>
    <paper id="103">
      <title>Undersensitivity in Neural Reading Comprehension</title>
      <author><first>Johannes</first><last>Welbl</last></author>
      <author><first>Pasquale</first><last>Minervini</last></author>
      <author><first>Max</first><last>Bartolo</last></author>
      <author><first>Pontus</first><last>Stenetorp</last></author>
      <author><first>Sebastian</first><last>Riedel</last></author>
      <pages>1152–1165</pages>
      <abstract>Current reading comprehension methods generalise well to in-distribution test sets, yet perform poorly on adversarially selected data. Prior work on adversarial inputs typically studies model oversensitivity: semantically invariant text perturbations that cause a model’s prediction to change. Here we focus on the complementary problem: excessive prediction undersensitivity, where input text is meaningfully changed but the model’s prediction does not, even though it should. We formulate an adversarial attack which searches among semantic variations of the question for which a model erroneously predicts the same answer, and with even higher probability. We demonstrate that models trained on both SQuAD2.0 and NewsQA are vulnerable to this attack, and then investigate data augmentation and adversarial training as defences. Both substantially decrease adversarial vulnerability, which generalises to held-out data and held-out attack spaces. Addressing undersensitivity furthermore improves model robustness on the previously introduced ADDSENT and ADDONESENT datasets, and models generalise better when facing train / evaluation distribution mismatch: they are less prone to overly rely on shallow predictive cues present only in the training set, and outperform a conventional model by as much as 10.9% F1.</abstract>
      <url hash="fc512314">2020.findings-emnlp.103</url>
      <doi>10.18653/v1/2020.findings-emnlp.103</doi>
    </paper>
    <paper id="104">
      <title><fixed-case>H</fixed-case>yper<fixed-case>T</fixed-case>ext: Endowing <fixed-case>F</fixed-case>ast<fixed-case>T</fixed-case>ext with Hyperbolic Geometry</title>
      <author><first>Yudong</first><last>Zhu</last></author>
      <author><first>Di</first><last>Zhou</last></author>
      <author><first>Jinghui</first><last>Xiao</last></author>
      <author><first>Xin</first><last>Jiang</last></author>
      <author><first>Xiao</first><last>Chen</last></author>
      <author><first>Qun</first><last>Liu</last></author>
      <pages>1166–1171</pages>
      <abstract>Natural language data exhibit tree-like hierarchical structures such as the hypernym-hyponym hierarchy in WordNet. FastText, as the state-of-the-art text classifier based on shallow neural network in Euclidean space, may not represent such hierarchies precisely with limited representation capacity. Considering that hyperbolic space is naturally suitable for modelling tree-like hierarchical data, we propose a new model named HyperText for efficient text classification by endowing FastText with hyperbolic geometry. Empirically, we show that HyperText outperforms FastText on a range of text classification tasks with much reduced parameters.</abstract>
      <url hash="6b432b34">2020.findings-emnlp.104</url>
      <doi>10.18653/v1/2020.findings-emnlp.104</doi>
    </paper>
    <paper id="105">
      <title><fixed-case>A</fixed-case>uto<fixed-case>ETER</fixed-case>: Automated Entity Type Representation for Knowledge Graph Embedding</title>
      <author><first>Guanglin</first><last>Niu</last></author>
      <author id="bo-li"><first>Bo</first><last>Li</last></author>
      <author><first>Yongfei</first><last>Zhang</last></author>
      <author><first>Shiliang</first><last>Pu</last></author>
      <author><first>Jingyang</first><last>Li</last></author>
      <pages>1172–1181</pages>
      <abstract>Recent advances in Knowledge Graph Embedding (KGE) allow for representing entities and relations in continuous vector spaces. Some traditional KGE models leveraging additional type information can improve the representation of entities which however totally rely on the explicit types or neglect the diverse type representations specific to various relations. Besides, none of the existing methods is capable of inferring all the relation patterns of symmetry, inversion and composition as well as the complex properties of 1-N, N-1 and N-N relations, simultaneously. To explore the type information for any KG, we develop a novel KGE framework with Automated Entity TypE Representation (AutoETER), which learns the latent type embedding of each entity by regarding each relation as a translation operation between the types of two entities with a relation-aware projection mechanism. Particularly, our designed automated type representation learning mechanism is a pluggable module which can be easily incorporated with any KGE model. Besides, our approach could model and infer all the relation patterns and complex relations. Experiments on four datasets demonstrate the superior performance of our model compared to state-of-the-art baselines on link prediction tasks, and the visualization of type clustering provides clearly the explanation of type embeddings and verifies the effectiveness of our model.</abstract>
      <url hash="cd59099a">2020.findings-emnlp.105</url>
      <doi>10.18653/v1/2020.findings-emnlp.105</doi>
    </paper>
    <paper id="106">
      <title>Learning Robust and Multilingual Speech Representations</title>
      <author><first>Kazuya</first><last>Kawakami</last></author>
      <author><first>Luyu</first><last>Wang</last></author>
      <author><first>Chris</first><last>Dyer</last></author>
      <author><first>Phil</first><last>Blunsom</last></author>
      <author><first>Aaron</first><last>van den Oord</last></author>
      <pages>1182–1192</pages>
      <abstract>Unsupervised speech representation learning has shown remarkable success at finding representations that correlate with phonetic structures and improve downstream speech recognition performance. However, most research has been focused on evaluating the representations in terms of their ability to improve the performance of speech recognition systems on read English (e.g. Wall Street Journal and LibriSpeech). This evaluation methodology overlooks two important desiderata that speech representations should have: robustness to domain shifts and transferability to other languages. In this paper we learn representations from up to 8000 hours of diverse and noisy speech data and evaluate the representations by looking at their robustness to domain shifts and their ability to improve recognition performance in many languages. We find that our representations confer significant robustness advantages to the resulting recognition systems: we see significant improvements in out-of-domain transfer relative to baseline feature sets and the features likewise provide improvements in 25 phonetically diverse languages.</abstract>
      <url hash="7ec7f85b">2020.findings-emnlp.106</url>
      <doi>10.18653/v1/2020.findings-emnlp.106</doi>
    </paper>
    <paper id="107">
      <title><fixed-case>FQ</fixed-case>u<fixed-case>AD</fixed-case>: <fixed-case>F</fixed-case>rench Question Answering Dataset</title>
      <author><first>Martin</first><last>d’Hoffschmidt</last></author>
      <author><first>Wacim</first><last>Belblidia</last></author>
      <author><first>Quentin</first><last>Heinrich</last></author>
      <author><first>Tom</first><last>Brendlé</last></author>
      <author><first>Maxime</first><last>Vidal</last></author>
      <pages>1193–1208</pages>
      <abstract>Recent advances in the field of language modeling have improved state-of-the-art results on many Natural Language Processing tasks. Among them, Reading Comprehension has made significant progress over the past few years. However, most results are reported in English since labeled resources available in other languages, such as French, remain scarce. In the present work, we introduce the French Question Answering Dataset (FQuAD). FQuAD is a French Native Reading Comprehension dataset of questions and answers on a set of Wikipedia articles that consists of 25,000+ samples for the 1.0 version and 60,000+ samples for the 1.1 version. We train a baseline model which achieves an F1 score of 92.2 and an exact match ratio of 82.1 on the test set. In an effort to track the progress of French Question Answering models we propose a leaderboard and we have made the 1.0 version of our dataset freely available at https://illuin-tech.github.io/FQuAD-explorer/.</abstract>
      <url hash="d3e6e117">2020.findings-emnlp.107</url>
      <doi>10.18653/v1/2020.findings-emnlp.107</doi>
    </paper>
    <paper id="108">
      <title>Dynamic Semantic Matching and Aggregation Network for Few-shot Intent Detection</title>
      <author><first>Hoang</first><last>Nguyen</last></author>
      <author><first>Chenwei</first><last>Zhang</last></author>
      <author><first>Congying</first><last>Xia</last></author>
      <author><first>Philip</first><last>Yu</last></author>
      <pages>1209–1218</pages>
      <abstract>Few-shot Intent Detection is challenging due to the scarcity of available annotated utterances. Although recent works demonstrate that multi-level matching plays an important role in transferring learned knowledge from seen training classes to novel testing classes, they rely on a static similarity measure and overly fine-grained matching components. These limitations inhibit generalizing capability towards Generalized Few-shot Learning settings where both seen and novel classes are co-existent. In this paper, we propose a novel Semantic Matching and Aggregation Network where semantic components are distilled from utterances via multi-head self-attention with additional dynamic regularization constraints. These semantic components capture high-level information, resulting in more effective matching between instances. Our multi-perspective matching method provides a comprehensive matching measure to enhance representations of both labeled and unlabeled instances. We also propose a more challenging evaluation setting that considers classification on the joint all-class label space. Extensive experimental results demonstrate the effectiveness of our method. Our code and data are publicly available.</abstract>
      <url hash="0c507ba8">2020.findings-emnlp.108</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f4d7092a">2020.findings-emnlp.108.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.108</doi>
    </paper>
    <paper id="109">
      <title>Quantifying the Contextualization of Word Representations with Semantic Class Probing</title>
      <author><first>Mengjie</first><last>Zhao</last></author>
      <author><first>Philipp</first><last>Dufter</last></author>
      <author><first>Yadollah</first><last>Yaghoobzadeh</last></author>
      <author><first>Hinrich</first><last>Schütze</last></author>
      <pages>1219–1234</pages>
      <abstract>Pretrained language models achieve state-of-the-art results on many NLP tasks, but there are still many open questions about how and why they work so well. We investigate the contextualization of words in BERT. We quantify the amount of contextualization, i.e., how well words are interpreted in context, by studying the extent to which semantic classes of a word can be inferred from its contextualized embedding. Quantifying contextualization helps in understanding and utilizing pretrained language models. We show that the top layer representations support highly accurate inference of semantic classes; that the strongest contextualization effects occur in the lower layers; that local context is mostly sufficient for contextualizing words; and that top layer representations are more task-specific after finetuning while lower layer representations are more transferable. Finetuning uncovers task-related features, but pretrained knowledge about contextualization is still well preserved.</abstract>
      <url hash="cb0815b2">2020.findings-emnlp.109</url>
      <doi>10.18653/v1/2020.findings-emnlp.109</doi>
    </paper>
    <paper id="110">
      <title>Learning to Generate Clinically Coherent Chest <fixed-case>X</fixed-case>-Ray Reports</title>
      <author><first>Justin</first><last>Lovelace</last></author>
      <author><first>Bobak</first><last>Mortazavi</last></author>
      <pages>1235–1243</pages>
      <abstract>Automated radiology report generation has the potential to reduce the time clinicians spend manually reviewing radiographs and streamline clinical care. However, past work has shown that typical abstractive methods tend to produce fluent, but clinically incorrect radiology reports. In this work, we develop a radiology report generation model utilizing the transformer architecture that produces superior reports as measured by both standard language generation and clinical coherence metrics compared to competitive baselines. We then develop a method to differentiably extract clinical information from generated reports and utilize this differentiability to fine-tune our model to produce more clinically coherent reports.</abstract>
      <url hash="a81d59cd">2020.findings-emnlp.110</url>
      <doi>10.18653/v1/2020.findings-emnlp.110</doi>
    </paper>
    <paper id="111">
      <title><fixed-case>FELIX</fixed-case>: Flexible Text Editing Through Tagging and Insertion</title>
      <author><first>Jonathan</first><last>Mallinson</last></author>
      <author><first>Aliaksei</first><last>Severyn</last></author>
      <author><first>Eric</first><last>Malmi</last></author>
      <author><first>Guillermo</first><last>Garrido</last></author>
      <pages>1244–1255</pages>
      <abstract>We present FELIX – a flexible text-editing approach for generation, designed to derive maximum benefit from the ideas of decoding with bi-directional contexts and self-supervised pretraining. In contrast to conventional sequenceto-sequence (seq2seq) models, FELIX is efficient in low-resource settings and fast at inference time, while being capable of modeling flexible input-output transformations. We achieve this by decomposing the text-editing task into two sub-tasks: tagging to decide on the subset of input tokens and their order in the output text and insertion to in-fill the missing tokens in the output not present in the input. The tagging model employs a novel Pointer mechanism, while the insertion model is based on a Masked Language Model (MLM). Both of these models are chosen to be non-autoregressive to guarantee faster inference. FELIX performs favourably when compared to recent text-editing methods and strong seq2seq baselines when evaluated on four NLG tasks: Sentence Fusion, Machine Translation Automatic Post-Editing, Summarization, and Text Simplification</abstract>
      <url hash="b11a21dc">2020.findings-emnlp.111</url>
      <doi>10.18653/v1/2020.findings-emnlp.111</doi>
    </paper>
    <paper id="112">
      <title>What Can We Do to Improve Peer Review in <fixed-case>NLP</fixed-case>?</title>
      <author><first>Anna</first><last>Rogers</last></author>
      <author><first>Isabelle</first><last>Augenstein</last></author>
      <pages>1256–1262</pages>
      <abstract>Peer review is our best tool for judging the quality of conference submissions, but it is becoming increasingly spurious. We argue that a part of the problem is that the reviewers and area chairs face a poorly defined task forcing apples-to-oranges comparisons. There are several potential ways forward, but the key difficulty is creating the incentives and mechanisms for their consistent implementation in the NLP community.</abstract>
      <url hash="79c8a96b">2020.findings-emnlp.112</url>
      <doi>10.18653/v1/2020.findings-emnlp.112</doi>
    </paper>
    <paper id="113">
      <title>Unsupervised Relation Extraction from Language Models using Constrained Cloze Completion</title>
      <author><first>Ankur</first><last>Goswami</last></author>
      <author><first>Akshata</first><last>Bhat</last></author>
      <author><first>Hadar</first><last>Ohana</last></author>
      <author><first>Theodoros</first><last>Rekatsinas</last></author>
      <pages>1263–1276</pages>
      <abstract>We show that state-of-the-art self-supervised language models can be readily used to extract relations from a corpus without the need to train a fine-tuned extractive head. We introduce RE-Flex, a simple framework that performs constrained cloze completion over pretrained language models to perform unsupervised relation extraction. RE-Flex uses contextual matching to ensure that language model predictions matches supporting evidence from the input corpus that is relevant to a target relation. We perform an extensive experimental study over multiple relation extraction benchmarks and demonstrate that RE-Flex outperforms competing unsupervised relation extraction methods based on pretrained language models by up to 27.8 F1 points compared to the next-best method. Our results show that constrained inference queries against a language model can enable accurate unsupervised relation extraction.</abstract>
      <url hash="4e994e00">2020.findings-emnlp.113</url>
      <attachment type="OptionalSupplementaryMaterial" hash="7936571b">2020.findings-emnlp.113.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.113</doi>
    </paper>
    <paper id="114">
      <title>Biomedical Event Extraction with Hierarchical Knowledge Graphs</title>
      <author><first>Kung-Hsiang</first><last>Huang</last></author>
      <author><first>Mu</first><last>Yang</last></author>
      <author><first>Nanyun</first><last>Peng</last></author>
      <pages>1277–1285</pages>
      <abstract>Biomedical event extraction is critical in understanding biomolecular interactions described in scientific corpus. One of the main challenges is to identify nested structured events that are associated with non-indicative trigger words. We propose to incorporate domain knowledge from Unified Medical Language System (UMLS) to a pre-trained language model via Graph Edge-conditioned Attention Networks (GEANet) and hierarchical graph representation. To better recognize the trigger words, each sentence is first grounded to a sentence graph based on a jointly modeled hierarchical knowledge graph from UMLS. The grounded graphs are then propagated by GEANet, a novel graph neural networks for enhanced capabilities in inferring complex events. On BioNLP 2011 GENIA Event Extraction task, our approach achieved 1.41% F1 and 3.19% F1 improvements on all events and complex events, respectively. Ablation studies confirm the importance of GEANet and hierarchical KG.</abstract>
      <url hash="1cb36c3d">2020.findings-emnlp.114</url>
      <attachment type="OptionalSupplementaryMaterial" hash="b6f72c4e">2020.findings-emnlp.114.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.114</doi>
    </paper>
    <paper id="115">
      <title>Language Generation via Combinatorial Constraint Satisfaction: A Tree Search Enhanced <fixed-case>M</fixed-case>onte-<fixed-case>C</fixed-case>arlo Approach</title>
      <author><first>Maosen</first><last>Zhang</last></author>
      <author><first>Nan</first><last>Jiang</last></author>
      <author><first>Lei</first><last>Li</last></author>
      <author><first>Yexiang</first><last>Xue</last></author>
      <pages>1286–1298</pages>
      <abstract>Generating natural language under complex constraints is a principled formulation towards controllable text generation. We present a framework to allow specification of combinatorial constraints for sentence generation. We propose TSMC, an efficient method to generate high likelihood sentences with respect to a pre-trained language model while satisfying the constraints. Our approach is highly flexible, requires no task-specific train- ing, and leverages efficient constraint satisfaction solving techniques. To better handle the combinatorial constraints, a tree search algorithm is embedded into the proposal process of the Markov Chain Monte Carlo (MCMC) to explore candidates that satisfy more constraints. Compared to existing MCMC approaches, our sampling approach has a better mixing performance. Experiments show that TSMC achieves consistent and significant improvement on multiple language generation tasks.</abstract>
      <url hash="389444db">2020.findings-emnlp.115</url>
      <attachment type="OptionalSupplementaryMaterial" hash="c8744862">2020.findings-emnlp.115.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.115</doi>
    </paper>
    <paper id="116">
      <title>Examining the Ordering of Rhetorical Strategies in Persuasive Requests</title>
      <author><first>Omar</first><last>Shaikh</last></author>
      <author><first>Jiaao</first><last>Chen</last></author>
      <author><first>Jon</first><last>Saad-Falcon</last></author>
      <author><first>Polo</first><last>Chau</last></author>
      <author><first>Diyi</first><last>Yang</last></author>
      <pages>1299–1306</pages>
      <abstract>Interpreting how persuasive language influences audiences has implications across many domains like advertising, argumentation, and propaganda. Persuasion relies on more than a message’s content. Arranging the order of the message itself (i.e., ordering specific rhetorical strategies) also plays an important role. To examine how strategy orderings contribute to persuasiveness, we first utilize a Variational Autoencoder model to disentangle content and rhetorical strategies in textual requests from a large-scale loan request corpus. We then visualize interplay between content and strategy through an attentional LSTM that predicts the success of textual requests. We find that specific (orderings of) strategies interact uniquely with a request’s content to impact success rate, and thus the persuasiveness of a request.</abstract>
      <url hash="c8e23ce8">2020.findings-emnlp.116</url>
      <doi>10.18653/v1/2020.findings-emnlp.116</doi>
    </paper>
    <paper id="117">
      <title>Evaluating Models’ Local Decision Boundaries via Contrast Sets</title>
      <author><first>Matt</first><last>Gardner</last></author>
      <author><first>Yoav</first><last>Artzi</last></author>
      <author><first>Victoria</first><last>Basmov</last></author>
      <author><first>Jonathan</first><last>Berant</last></author>
      <author><first>Ben</first><last>Bogin</last></author>
      <author><first>Sihao</first><last>Chen</last></author>
      <author><first>Pradeep</first><last>Dasigi</last></author>
      <author><first>Dheeru</first><last>Dua</last></author>
      <author><first>Yanai</first><last>Elazar</last></author>
      <author><first>Ananth</first><last>Gottumukkala</last></author>
      <author><first>Nitish</first><last>Gupta</last></author>
      <author><first>Hannaneh</first><last>Hajishirzi</last></author>
      <author><first>Gabriel</first><last>Ilharco</last></author>
      <author><first>Daniel</first><last>Khashabi</last></author>
      <author><first>Kevin</first><last>Lin</last></author>
      <author><first>Jiangming</first><last>Liu</last></author>
      <author><first>Nelson F.</first><last>Liu</last></author>
      <author><first>Phoebe</first><last>Mulcaire</last></author>
      <author><first>Qiang</first><last>Ning</last></author>
      <author><first>Sameer</first><last>Singh</last></author>
      <author><first>Noah A.</first><last>Smith</last></author>
      <author><first>Sanjay</first><last>Subramanian</last></author>
      <author><first>Reut</first><last>Tsarfaty</last></author>
      <author><first>Eric</first><last>Wallace</last></author>
      <author><first>Ally</first><last>Zhang</last></author>
      <author><first>Ben</first><last>Zhou</last></author>
      <pages>1307–1323</pages>
      <abstract>Standard test sets for supervised learning evaluate in-distribution generalization. Unfortunately, when a dataset has systematic gaps (e.g., annotation artifacts), these evaluations are misleading: a model can learn simple decision rules that perform well on the test set but do not capture the abilities a dataset is intended to test. We propose a more rigorous annotation paradigm for NLP that helps to close systematic gaps in the test data. In particular, after a dataset is constructed, we recommend that the dataset authors manually perturb the test instances in small but meaningful ways that (typically) change the gold label, creating contrast sets. Contrast sets provide a local view of a model’s decision boundary, which can be used to more accurately evaluate a model’s true linguistic capabilities. We demonstrate the efficacy of contrast sets by creating them for 10 diverse NLP datasets (e.g., DROP reading comprehension, UD parsing, and IMDb sentiment analysis). Although our contrast sets are not explicitly adversarial, model performance is significantly lower on them than on the original test sets—up to 25% in some cases. We release our contrast sets as new evaluation benchmarks and encourage future dataset construction efforts to follow similar annotation processes.</abstract>
      <url hash="85e5a01e">2020.findings-emnlp.117</url>
      <doi>10.18653/v1/2020.findings-emnlp.117</doi>
    </paper>
    <paper id="118">
      <title>Parsing with Multilingual <fixed-case>BERT</fixed-case>, a Small Corpus, and a Small Treebank</title>
      <author><first>Ethan C.</first><last>Chau</last></author>
      <author><first>Lucy H.</first><last>Lin</last></author>
      <author><first>Noah A.</first><last>Smith</last></author>
      <pages>1324–1334</pages>
      <abstract>Pretrained multilingual contextual representations have shown great success, but due to the limits of their pretraining data, their benefits do not apply equally to all language varieties. This presents a challenge for language varieties unfamiliar to these models, whose labeled and unlabeled data is too limited to train a monolingual model effectively. We propose the use of additional language-specific pretraining and vocabulary augmentation to adapt multilingual models to low-resource settings. Using dependency parsing of four diverse low-resource language varieties as a case study, we show that these methods significantly improve performance over baselines, especially in the lowest-resource cases, and demonstrate the importance of the relationship between such models’ pretraining data and target language varieties.</abstract>
      <url hash="f1b9e996">2020.findings-emnlp.118</url>
      <doi>10.18653/v1/2020.findings-emnlp.118</doi>
    </paper>
    <paper id="119">
      <title><fixed-case>O</fixed-case>pt<fixed-case>SLA</fixed-case>: an Optimization-Based Approach for Sequential Label Aggregation</title>
      <author><first>Nasim</first><last>Sabetpour</last></author>
      <author><first>Adithya</first><last>Kulkarni</last></author>
      <author><first>Qi</first><last>Li</last></author>
      <pages>1335–1340</pages>
      <abstract>The need for the annotated training dataset on which data-hungry machine learning algorithms feed has increased dramatically with advanced acclaim of machine learning applications. To annotate the data, people with domain expertise are needed, but they are seldom available and expensive to hire. This has lead to the thriving of crowdsourcing platforms such as Amazon Mechanical Turk (AMT). However, the annotations provided by one worker cannot be used directly to train the model due to the lack of expertise. Existing literature in annotation aggregation focuses on binary and multi-choice problems. In contrast, little work has been done on complex tasks such as sequence labeling with imbalanced classes, a ubiquitous task in Natural Language Processing (NLP), and Bio-Informatics. We propose OptSLA, an Optimization-based Sequential Label Aggregation method, that jointly considers the characteristics of sequential labeling tasks, workers reliabilities, and advanced deep learning techniques to conquer the challenge. We evaluate our model on crowdsourced data for named entity recognition task. Our results show that the proposed OptSLA outperforms the state-of-the-art aggregation methods, and the results are easier to interpret.</abstract>
      <url hash="6e091ea4">2020.findings-emnlp.119</url>
      <doi>10.18653/v1/2020.findings-emnlp.119</doi>
    </paper>
    <paper id="120">
      <title>Optimizing Word Segmentation for Downstream Task</title>
      <author><first>Tatsuya</first><last>Hiraoka</last></author>
      <author><first>Sho</first><last>Takase</last></author>
      <author><first>Kei</first><last>Uchiumi</last></author>
      <author><first>Atsushi</first><last>Keyaki</last></author>
      <author><first>Naoaki</first><last>Okazaki</last></author>
      <pages>1341–1351</pages>
      <abstract>In traditional NLP, we tokenize a given sentence as a preprocessing, and thus the tokenization is unrelated to a target downstream task. To address this issue, we propose a novel method to explore a tokenization which is appropriate for the downstream task. Our proposed method, optimizing tokenization (OpTok), is trained to assign a high probability to such appropriate tokenization based on the downstream task loss. OpTok can be used for any downstream task which uses a vector representation of a sentence such as text classification. Experimental results demonstrate that OpTok improves the performance of sentiment analysis and textual entailment. In addition, we introduce OpTok into BERT, the state-of-the-art contextualized embeddings and report a positive effect.</abstract>
      <url hash="4d55146a">2020.findings-emnlp.120</url>
      <doi>10.18653/v1/2020.findings-emnlp.120</doi>
    </paper>
    <paper id="121">
      <title>Dynamically Updating Event Representations for Temporal Relation Classification with Multi-category Learning</title>
      <author><first>Fei</first><last>Cheng</last></author>
      <author><first>Masayuki</first><last>Asahara</last></author>
      <author><first>Ichiro</first><last>Kobayashi</last></author>
      <author><first>Sadao</first><last>Kurohashi</last></author>
      <pages>1352–1357</pages>
      <abstract>Temporal relation classification is the pair-wise task for identifying the relation of a temporal link (TLINKs) between two mentions, i.e. event, time and document creation time (DCT). It leads to two crucial limits: 1) Two TLINKs involving a common mention do not share information. 2) Existing models with independent classifiers for each TLINK category (E2E, E2T and E2D) hinder from using the whole data. This paper presents an event centric model that allows to manage dynamic event representations across multiple TLINKs. Our model deals with three TLINK categories with multi-task learning to leverage the full size of data. The experimental results show that our proposal outperforms state-of-the-art models and two strong transfer learning baselines on both the English and Japanese data.</abstract>
      <url hash="f353b197">2020.findings-emnlp.121</url>
      <doi>10.18653/v1/2020.findings-emnlp.121</doi>
    </paper>
    <paper id="122">
      <title>A Compare Aggregate Transformer for Understanding Document-grounded Dialogue</title>
      <author><first>Longxuan</first><last>Ma</last></author>
      <author><first>Wei-Nan</first><last>Zhang</last></author>
      <author><first>Runxin</first><last>Sun</last></author>
      <author><first>Ting</first><last>Liu</last></author>
      <pages>1358–1367</pages>
      <abstract>Unstructured documents serving as external knowledge of the dialogues help to generate more informative responses. Previous research focused on knowledge selection (KS) in the document with dialogue. However, dialogue history that is not related to the current dialogue may introduce noise in the KS processing. In this paper, we propose a Compare Aggregate Transformer (CAT) to jointly denoise the dialogue context and aggregate the document information for response generation. We designed two different comparison mechanisms to reduce noise (before and during decoding). In addition, we propose two metrics for evaluating document utilization efficiency based on word overlap. Experimental results on the CMU_DoG dataset show that the proposed CAT model outperforms the state-of-the-art approach and strong baselines.</abstract>
      <url hash="9021cca4">2020.findings-emnlp.122</url>
      <doi>10.18653/v1/2020.findings-emnlp.122</doi>
    </paper>
    <paper id="123">
      <title><fixed-case>T</fixed-case>ext<fixed-case>H</fixed-case>ide: Tackling Data Privacy in Language Understanding Tasks</title>
      <author><first>Yangsibo</first><last>Huang</last></author>
      <author><first>Zhao</first><last>Song</last></author>
      <author><first>Danqi</first><last>Chen</last></author>
      <author><first>Kai</first><last>Li</last></author>
      <author><first>Sanjeev</first><last>Arora</last></author>
      <pages>1368–1382</pages>
      <abstract>An unsolved challenge in distributed or federated learning is to effectively mitigate privacy risks without slowing down training or reducing accuracy. In this paper, we propose TextHide aiming at addressing this challenge for natural language understanding tasks. It requires all participants to add a simple encryption step to prevent an eavesdropping attacker from recovering private text data. Such an encryption step is efficient and only affects the task performance slightly. In addition, TextHide fits well with the popular framework of fine-tuning pre-trained language models (e.g., BERT) for any sentence or sentence-pair task. We evaluate TextHide on the GLUE benchmark, and our experiments show that TextHide can effectively defend attacks on shared gradients or representations and the averaged accuracy reduction is only 1.9%. We also present an analysis of the security of TextHide using a conjecture about the computational intractability of a mathematical problem.</abstract>
      <url hash="8dd733fd">2020.findings-emnlp.123</url>
      <doi>10.18653/v1/2020.findings-emnlp.123</doi>
    </paper>
    <paper id="124">
      <title>Modeling Intra and Inter-modality Incongruity for Multi-Modal Sarcasm Detection</title>
      <author><first>Hongliang</first><last>Pan</last></author>
      <author><first>Zheng</first><last>Lin</last></author>
      <author><first>Peng</first><last>Fu</last></author>
      <author><first>Yatao</first><last>Qi</last></author>
      <author><first>Weiping</first><last>Wang</last></author>
      <pages>1383–1392</pages>
      <abstract>Sarcasm is a pervasive phenomenon in today’s social media platforms such as Twitter and Reddit. These platforms allow users to create multi-modal messages, including texts, images, and videos. Existing multi-modal sarcasm detection methods either simply concatenate the features from multi modalities or fuse the multi modalities information in a designed manner. However, they ignore the incongruity character in sarcastic utterance, which is often manifested between modalities or within modalities. Inspired by this, we propose a BERT architecture-based model, which concentrates on both intra and inter-modality incongruity for multi-modal sarcasm detection. To be specific, we are inspired by the idea of self-attention mechanism and design inter-modality attention to capturing inter-modality incongruity. In addition, the co-attention mechanism is applied to model the contradiction within the text. The incongruity information is then used for prediction. The experimental results demonstrate that our model achieves state-of-the-art performance on a public multi-modal sarcasm detection dataset.</abstract>
      <url hash="57b757ce">2020.findings-emnlp.124</url>
      <doi>10.18653/v1/2020.findings-emnlp.124</doi>
    </paper>
    <paper id="125">
      <title>Investigating Transferability in Pretrained Language Models</title>
      <author><first>Alex</first><last>Tamkin</last></author>
      <author><first>Trisha</first><last>Singh</last></author>
      <author><first>Davide</first><last>Giovanardi</last></author>
      <author><first>Noah</first><last>Goodman</last></author>
      <pages>1393–1401</pages>
      <abstract>How does language model pretraining help transfer learning? We consider a simple ablation technique for determining the impact of each pretrained layer on transfer task performance. This method, partial reinitialization, involves replacing different layers of a pretrained model with random weights, then finetuning the entire model on the transfer task and observing the change in performance. This technique reveals that in BERT, layers with high probing performance on downstream GLUE tasks are neither necessary nor sufficient for high accuracy on those tasks. Furthermore, the benefit of using pretrained parameters for a layer varies dramatically with finetuning dataset size: parameters that provide tremendous performance improvement when data is plentiful may provide negligible benefits in data-scarce settings. These results reveal the complexity of the transfer learning process, highlighting the limitations of methods that operate on frozen models or single data samples.</abstract>
      <url hash="b2dd5db7">2020.findings-emnlp.125</url>
      <doi>10.18653/v1/2020.findings-emnlp.125</doi>
    </paper>
    <paper id="126">
      <title>Improving Knowledge-Aware Dialogue Response Generation by Using Human-Written Prototype Dialogues</title>
      <author><first>Sixing</first><last>Wu</last></author>
      <author><first>Ying</first><last>Li</last></author>
      <author><first>Dawei</first><last>Zhang</last></author>
      <author><first>Zhonghai</first><last>Wu</last></author>
      <pages>1402–1411</pages>
      <abstract>Incorporating commonsense knowledge can alleviate the issue of generating generic responses in open-domain generative dialogue systems. However, selecting knowledge facts for the dialogue context is still a challenge. The widely used approach Entity Name Matching always retrieves irrelevant facts from the view of local entity words. This paper proposes a novel knowledge selection approach, Prototype-KR, and a knowledge-aware generative model, Prototype-KRG. Given a query, our approach first retrieves a set of prototype dialogues that are relevant to the query. We find knowledge facts used in prototype dialogues usually are highly relevant to the current query; thus, Prototype-KR ranks such knowledge facts based on the semantic similarity and then selects the most appropriate facts. Subsequently, Prototype-KRG can generate an informative response using the selected knowledge facts. Experiments demonstrate that our approach has achieved notable improvements on the most metrics, compared to generative baselines. Meanwhile, compared to IR(Retrieval)-based baselines, responses generated by our approach are more relevant to the context and have comparable informativeness.</abstract>
      <url hash="0652cc54">2020.findings-emnlp.126</url>
      <doi>10.18653/v1/2020.findings-emnlp.126</doi>
    </paper>
    <paper id="127">
      <title>Filtering before Iteratively Referring for Knowledge-Grounded Response Selection in Retrieval-Based Chatbots</title>
      <author><first>Jia-Chen</first><last>Gu</last></author>
      <author><first>Zhenhua</first><last>Ling</last></author>
      <author><first>Quan</first><last>Liu</last></author>
      <author><first>Zhigang</first><last>Chen</last></author>
      <author><first>Xiaodan</first><last>Zhu</last></author>
      <pages>1412–1422</pages>
      <abstract>The challenges of building knowledge-grounded retrieval-based chatbots lie in how to ground a conversation on its background knowledge and how to match response candidates with both context and knowledge simultaneously. This paper proposes a method named Filtering before Iteratively REferring (FIRE) for this task. In this method, a context filter and a knowledge filter are first built, which derive knowledge-aware context representations and context-aware knowledge representations respectively by global and bidirectional attention. Besides, the entries irrelevant to the conversation are discarded by the knowledge filter. After that, iteratively referring is performed between context and response representations as well as between knowledge and response representations, in order to collect deep matching features for scoring response candidates. Experimental results show that FIRE outperforms previous methods by margins larger than 2.8% and 4.1% on the PERSONA-CHAT dataset with original and revised personas respectively, and margins larger than 3.1% on the CMU_DoG dataset in terms of top-1 accuracy. We also show that FIRE is more interpretable by visualizing the knowledge grounding process.</abstract>
      <url hash="ff07e960">2020.findings-emnlp.127</url>
      <doi>10.18653/v1/2020.findings-emnlp.127</doi>
    </paper>
    <paper id="128">
      <title>Privacy-Preserving News Recommendation Model Learning</title>
      <author><first>Tao</first><last>Qi</last></author>
      <author><first>Fangzhao</first><last>Wu</last></author>
      <author><first>Chuhan</first><last>Wu</last></author>
      <author><first>Yongfeng</first><last>Huang</last></author>
      <author><first>Xing</first><last>Xie</last></author>
      <pages>1423–1432</pages>
      <abstract>News recommendation aims to display news articles to users based on their personal interest. Existing news recommendation methods rely on centralized storage of user behavior data for model training, which may lead to privacy concerns and risks due to the privacy-sensitive nature of user behaviors. In this paper, we propose a privacy-preserving method for news recommendation model training based on federated learning, where the user behavior data is locally stored on user devices. Our method can leverage the useful information in the behaviors of massive number users to train accurate news recommendation models and meanwhile remove the need of centralized storage of them. More specifically, on each user device we keep a local copy of the news recommendation model, and compute gradients of the local model based on the user behaviors in this device. The local gradients from a group of randomly selected users are uploaded to server, which are further aggregated to update the global model in the server. Since the model gradients may contain some implicit private information, we apply local differential privacy (LDP) to them before uploading for better privacy protection. The updated global model is then distributed to each user device for local model update. We repeat this process for multiple rounds. Extensive experiments on a real-world dataset show the effectiveness of our method in news recommendation model training with privacy protection.</abstract>
      <url hash="35e35de2">2020.findings-emnlp.128</url>
      <doi>10.18653/v1/2020.findings-emnlp.128</doi>
    </paper>
    <paper id="129">
      <title>ex<fixed-case>BERT</fixed-case>: Extending Pre-trained Models with Domain-specific Vocabulary Under Constrained Training Resources</title>
      <author><first>Wen</first><last>Tai</last></author>
      <author><first>H. T.</first><last>Kung</last></author>
      <author><first>Xin</first><last>Dong</last></author>
      <author><first>Marcus</first><last>Comiter</last></author>
      <author><first>Chang-Fu</first><last>Kuo</last></author>
      <pages>1433–1439</pages>
      <abstract>We introduce exBERT, a training method to extend BERT pre-trained models from a general domain to a new pre-trained model for a specific domain with a new additive vocabulary under constrained training resources (i.e., constrained computation and data). exBERT uses a small extension module to learn to adapt an augmenting embedding for the new domain in the context of the original BERT’s embedding of a general vocabulary. The exBERT training method is novel in learning the new vocabulary and the extension module while keeping the weights of the original BERT model fixed, resulting in a substantial reduction in required training resources. We pre-train exBERT with biomedical articles from ClinicalKey and PubMed Central, and study its performance on biomedical downstream benchmark tasks using the MTL-Bioinformatics-2016 datasets. We demonstrate that exBERT consistently outperforms prior approaches when using limited corpus and pre-training computation resources.</abstract>
      <url hash="42e3c8c2">2020.findings-emnlp.129</url>
      <doi>10.18653/v1/2020.findings-emnlp.129</doi>
    </paper>
    <paper id="130">
      <title>Balancing via Generation for Multi-Class Text Classification Improvement</title>
      <author><first>Naama</first><last>Tepper</last></author>
      <author><first>Esther</first><last>Goldbraich</last></author>
      <author><first>Naama</first><last>Zwerdling</last></author>
      <author><first>George</first><last>Kour</last></author>
      <author><first>Ateret</first><last>Anaby Tavor</last></author>
      <author><first>Boaz</first><last>Carmeli</last></author>
      <pages>1440–1452</pages>
      <abstract>Data balancing is a known technique for improving the performance of classification tasks. In this work we define a novel balancing-viageneration framework termed BalaGen. BalaGen consists of a flexible balancing policy coupled with a text generation mechanism. Combined, these two techniques can be used to augment a dataset for more balanced distribution. We evaluate BalaGen on three publicly available semantic utterance classification (SUC) datasets. One of these is a new COVID-19 Q&amp;A dataset published here for the first time. Our work demonstrates that optimal balancing policies can significantly improve classifier performance, while augmenting just part of the classes and under-sampling others. Furthermore, capitalizing on the advantages of balancing, we show its usefulness in all relevant BalaGen framework components. We validate the superiority of BalaGen on ten semantic utterance datasets taken from real-life goaloriented dialogue systems. Based on our results we encourage using data balancing prior to training for text classification tasks.</abstract>
      <url hash="23964a14">2020.findings-emnlp.130</url>
      <doi>10.18653/v1/2020.findings-emnlp.130</doi>
    </paper>
    <paper id="131">
      <title>Conditional Neural Generation using Sub-Aspect Functions for Extractive News Summarization</title>
      <author><first>Zhengyuan</first><last>Liu</last></author>
      <author><first>Ke</first><last>Shi</last></author>
      <author><first>Nancy</first><last>Chen</last></author>
      <pages>1453–1463</pages>
      <abstract>Much progress has been made in text summarization, fueled by neural architectures using large-scale training corpora. However, in the news domain, neural models easily overfit by leveraging position-related features due to the prevalence of the inverted pyramid writing style. In addition, there is an unmet need to generate a variety of summaries for different users. In this paper, we propose a neural framework that can flexibly control summary generation by introducing a set of sub-aspect functions (i.e. importance, diversity, position). These sub-aspect functions are regulated by a set of control codes to decide which sub-aspect to focus on during summary generation. We demonstrate that extracted summaries with minimal position bias is comparable with those generated by standard models that take advantage of position preference. We also show that news summaries generated with a focus on diversity can be more preferred by human raters. These results suggest that a more flexible neural summarization framework providing more control options could be desirable in tailoring to different user preferences, which is useful since it is often impractical to articulate such preferences for different applications a priori.</abstract>
      <url hash="baae7a2f">2020.findings-emnlp.131</url>
      <attachment type="OptionalSupplementaryMaterial" hash="c486f298">2020.findings-emnlp.131.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.131</doi>
    </paper>
    <paper id="132">
      <title>Research Replication Prediction Using Weakly Supervised Learning</title>
      <author><first>Tianyi</first><last>Luo</last></author>
      <author><first>Xingyu</first><last>Li</last></author>
      <author><first>Hainan</first><last>Wang</last></author>
      <author id="yang-liu-umich"><first>Yang</first><last>Liu</last></author>
      <pages>1464–1474</pages>
      <abstract>Knowing whether a published research result can be replicated is important. Carrying out direct replication of published research incurs a high cost. There are efforts tried to use machine learning aided methods to predict scientific claims’ replicability. However, existing machine learning aided approaches use only hand-extracted statistics features such as p-value, sample size, etc. without utilizing research papers’ text information and train only on a very small size of annotated data without making the most use of a large number of unlabeled articles. Therefore, it is desirable to develop effective machine learning aided automatic methods which can automatically extract text information as features so that we can benefit from Natural Language Processing techniques. Besides, we aim for an approach that benefits from both labeled and the large number of unlabeled data. In this paper, we propose two weakly supervised learning approaches that use automatically extracted text information of research papers to improve the prediction accuracy of research replication using both labeled and unlabeled datasets. Our experiments over real-world datasets show that our approaches obtain much better prediction performance compared to the supervised models utilizing only statistic features and a small size of labeled dataset. Further, we are able to achieve an accuracy of 75.76% for predicting the replicability of research.</abstract>
      <url hash="cc511cfb">2020.findings-emnlp.132</url>
      <doi>10.18653/v1/2020.findings-emnlp.132</doi>
    </paper>
    <paper id="133">
      <title>Open Domain Question Answering based on Text Enhanced Knowledge Graph with Hyperedge Infusion</title>
      <author><first>Jiale</first><last>Han</last></author>
      <author><first>Bo</first><last>Cheng</last></author>
      <author><first>Xu</first><last>Wang</last></author>
      <pages>1475–1481</pages>
      <abstract>The incompleteness of knowledge base (KB) is a vital factor limiting the performance of question answering (QA). This paper proposes a novel QA method by leveraging text information to enhance the incomplete KB. The model enriches the entity representation through semantic information contained in the text, and employs graph convolutional networks to update the entity status. Furthermore, to exploit the latent structural information of text, we treat the text as hyperedges connecting entities among it to complement the deficient relations in KB, and hypergraph convolutional networks are further applied to reason on the hypergraph-formed text. Extensive experiments on the WebQuestionsSP benchmark with different KB settings prove the effectiveness of our model.</abstract>
      <url hash="b5c343fd">2020.findings-emnlp.133</url>
      <doi>10.18653/v1/2020.findings-emnlp.133</doi>
    </paper>
    <paper id="134">
      <title>Inexpensive Domain Adaptation of Pretrained Language Models: Case Studies on Biomedical <fixed-case>NER</fixed-case> and Covid-19 <fixed-case>QA</fixed-case></title>
      <author><first>Nina</first><last>Poerner</last></author>
      <author><first>Ulli</first><last>Waltinger</last></author>
      <author><first>Hinrich</first><last>Schütze</last></author>
      <pages>1482–1490</pages>
      <abstract>Domain adaptation of Pretrained Language Models (PTLMs) is typically achieved by unsupervised pretraining on target-domain text. While successful, this approach is expensive in terms of hardware, runtime and CO 2 emissions. Here, we propose a cheaper alternative: We train Word2Vec on target-domain text and align the resulting word vectors with the wordpiece vectors of a general-domain PTLM. We evaluate on eight English biomedical Named Entity Recognition (NER) tasks and compare against the recently proposed BioBERT model. We cover over 60% of the BioBERT - BERT F1 delta, at 5% of BioBERT’s CO 2 footprint and 2% of its cloud compute cost. We also show how to quickly adapt an existing general-domain Question Answering (QA) model to an emerging domain: the Covid-19 pandemic.</abstract>
      <url hash="7e17e67f">2020.findings-emnlp.134</url>
      <attachment type="OptionalSupplementaryMaterial" hash="c39d60b4">2020.findings-emnlp.134.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.134</doi>
    </paper>
    <paper id="135">
      <title>Semantically Driven Sentence Fusion: Modeling and Evaluation</title>
      <author><first>Eyal</first><last>Ben-David</last></author>
      <author><first>Orgad</first><last>Keller</last></author>
      <author><first>Eric</first><last>Malmi</last></author>
      <author><first>Idan</first><last>Szpektor</last></author>
      <author><first>Roi</first><last>Reichart</last></author>
      <pages>1491–1505</pages>
      <abstract>Sentence fusion is the task of joining related sentences into coherent text. Current training and evaluation schemes for this task are based on single reference ground-truths and do not account for valid fusion variants. We show that this hinders models from robustly capturing the semantic relationship between input sentences. To alleviate this, we present an approach in which ground-truth solutions are automatically expanded into multiple references via curated equivalence classes of connective phrases. We apply this method to a large-scale dataset and use the augmented dataset for both model training and evaluation. To improve the learning of semantic representation using multiple references, we enrich the model with auxiliary discourse classification tasks under a multi-tasking framework. Our experiments highlight the improvements of our approach over state-of-the-art models.</abstract>
      <url hash="37452d68">2020.findings-emnlp.135</url>
      <doi>10.18653/v1/2020.findings-emnlp.135</doi>
    </paper>
    <paper id="136">
      <title>Pseudo-Bidirectional Decoding for Local Sequence Transduction</title>
      <author><first>Wangchunshu</first><last>Zhou</last></author>
      <author><first>Tao</first><last>Ge</last></author>
      <author><first>Ke</first><last>Xu</last></author>
      <pages>1506–1511</pages>
      <abstract>Local sequence transduction (LST) tasks are sequence transduction tasks where there exists massive overlapping between the source and target sequences, such as grammatical error correction and spell or OCR correction. Motivated by this characteristic of LST tasks, we propose Pseudo-Bidirectional Decoding (PBD), a simple but versatile approach for LST tasks. PBD copies the representation of source tokens to the decoder as pseudo future context that enables the decoder self-attention to attends to its bi-directional context. In addition, the bidirectional decoding scheme and the characteristic of LST tasks motivate us to share the encoder and the decoder of LST models. Our approach provides right-side context information for the decoder, reduces the number of parameters by half, and provides good regularization effects. Experimental results on several benchmark datasets show that our approach consistently improves the performance of standard seq2seq models on LST tasks.</abstract>
      <url hash="fc35666e">2020.findings-emnlp.136</url>
      <doi>10.18653/v1/2020.findings-emnlp.136</doi>
    </paper>
    <paper id="137">
      <title>Predicting Responses to Psychological Questionnaires from Participants’ Social Media Posts and Question Text Embeddings</title>
      <author><first>Huy</first><last>Vu</last></author>
      <author><first>Suhaib</first><last>Abdurahman</last></author>
      <author><first>Sudeep</first><last>Bhatia</last></author>
      <author><first>Lyle</first><last>Ungar</last></author>
      <pages>1512–1524</pages>
      <abstract>Psychologists routinely assess people’s emotions and traits, such as their personality, by collecting their responses to survey questionnaires. Such assessments can be costly in terms of both time and money, and often lack generalizability, as existing data cannot be used to predict responses for new survey questions or participants. In this study, we propose a method for predicting a participant’s questionnaire response using their social media texts and the text of the survey question they are asked. Specifically, we use Natural Language Processing (NLP) tools such as BERT embeddings to represent both participants (via the text they write) and survey questions as embeddings vectors, allowing us to predict responses for out-of-sample participants and questions. Our novel approach can be used by researchers to integrate new participants or new questions into psychological studies without the constraint of costly data collection, facilitating novel practical applications and furthering the development of psychological theory. Finally, as a side contribution, the success of our model also suggests a new approach to study survey questions using NLP tools such as text embeddings rather than response data used in traditional methods.</abstract>
      <url hash="83cf5b5d">2020.findings-emnlp.137</url>
      <attachment type="OptionalSupplementaryMaterial" hash="d5dc8cc9">2020.findings-emnlp.137.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.137</doi>
    </paper>
    <paper id="138">
      <title>Will it Unblend?</title>
      <author><first>Yuval</first><last>Pinter</last></author>
      <author><first>Cassandra L.</first><last>Jacobs</last></author>
      <author><first>Jacob</first><last>Eisenstein</last></author>
      <pages>1525–1535</pages>
      <abstract>Natural language processing systems often struggle with out-of-vocabulary (OOV) terms, which do not appear in training data. Blends, such as “innoventor”, are one particularly challenging class of OOV, as they are formed by fusing together two or more bases that relate to the intended meaning in unpredictable manners and degrees. In this work, we run experiments on a novel dataset of English OOV blends to quantify the difficulty of interpreting the meanings of blends by large-scale contextual language models such as BERT. We first show that BERT’s processing of these blends does not fully access the component meanings, leaving their contextual representations semantically impoverished. We find this is mostly due to the loss of characters resulting from blend formation. Then, we assess how easily different models can recognize the structure and recover the origin of blends, and find that context-aware embedding systems outperform character-level and context-free embeddings, although their results are still far from satisfactory.</abstract>
      <url hash="fae0fae9">2020.findings-emnlp.138</url>
      <attachment type="OptionalSupplementaryMaterial" hash="76b1cd9a">2020.findings-emnlp.138.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.138</doi>
    </paper>
    <paper id="139">
      <title><fixed-case>C</fixed-case>ode<fixed-case>BERT</fixed-case>: A Pre-Trained Model for Programming and Natural Languages</title>
      <author><first>Zhangyin</first><last>Feng</last></author>
      <author><first>Daya</first><last>Guo</last></author>
      <author><first>Duyu</first><last>Tang</last></author>
      <author><first>Nan</first><last>Duan</last></author>
      <author><first>Xiaocheng</first><last>Feng</last></author>
      <author><first>Ming</first><last>Gong</last></author>
      <author><first>Linjun</first><last>Shou</last></author>
      <author><first>Bing</first><last>Qin</last></author>
      <author><first>Ting</first><last>Liu</last></author>
      <author><first>Daxin</first><last>Jiang</last></author>
      <author><first>Ming</first><last>Zhou</last></author>
      <pages>1536–1547</pages>
      <abstract>We present CodeBERT, a bimodal pre-trained model for programming language (PL) and natural language (NL). CodeBERT learns general-purpose representations that support downstream NL-PL applications such as natural language code search, code documentation generation, etc. We develop CodeBERT with Transformer-based neural architecture, and train it with a hybrid objective function that incorporates the pre-training task of replaced token detection, which is to detect plausible alternatives sampled from generators. This enables us to utilize both “bimodal” data of NL-PL pairs and “unimodal data, where the former provides input tokens for model training while the latter helps to learn better generators. We evaluate CodeBERT on two NL-PL applications by fine-tuning model parameters. Results show that CodeBERT achieves state-of-the-art performance on both natural language code search and code documentation generation. Furthermore, to investigate what type of knowledge is learned in CodeBERT, we construct a dataset for NL-PL probing, and evaluate in a zero-shot setting where parameters of pre-trained models are fixed. Results show that CodeBERT performs better than previous pre-trained models on NLPL probing.</abstract>
      <url hash="79b17c55">2020.findings-emnlp.139</url>
      <doi>10.18653/v1/2020.findings-emnlp.139</doi>
    </paper>
    <paper id="140">
      <title><fixed-case>S</fixed-case>tyle<fixed-case>DGPT</fixed-case>: Stylized Response Generation with Pre-trained Language Models</title>
      <author><first>Ze</first><last>Yang</last></author>
      <author><first>Wei</first><last>Wu</last></author>
      <author><first>Can</first><last>Xu</last></author>
      <author><first>Xinnian</first><last>Liang</last></author>
      <author><first>Jiaqi</first><last>Bai</last></author>
      <author><first>Liran</first><last>Wang</last></author>
      <author><first>Wei</first><last>Wang</last></author>
      <author><first>Zhoujun</first><last>Li</last></author>
      <pages>1548–1559</pages>
      <abstract>Generating responses following a desired style has great potentials to extend applications of open-domain dialogue systems, yet is refrained by lacking of parallel data for training. In this work, we explore the challenging task with pre-trained language models that have brought breakthrough to various natural language tasks. To this end, we introduce a KL loss and a style classifier to the fine-tuning step in order to steer response generation towards the target style in both a word-level and a sentence-level. Comprehensive empirical studies with two public datasets indicate that our model can significantly outperform state-of-the-art methods in terms of both style consistency and contextual coherence.</abstract>
      <url hash="b89917ce">2020.findings-emnlp.140</url>
      <doi>10.18653/v1/2020.findings-emnlp.140</doi>
    </paper>
    <paper id="141">
      <title>Enhancing Automated Essay Scoring Performance via Fine-tuning Pre-trained Language Models with Combination of Regression and Ranking</title>
      <author><first>Ruosong</first><last>Yang</last></author>
      <author><first>Jiannong</first><last>Cao</last></author>
      <author><first>Zhiyuan</first><last>Wen</last></author>
      <author><first>Youzheng</first><last>Wu</last></author>
      <author><first>Xiaodong</first><last>He</last></author>
      <pages>1560–1569</pages>
      <abstract>Automated Essay Scoring (AES) is a critical text regression task that automatically assigns scores to essays based on their writing quality. Recently, the performance of sentence prediction tasks has been largely improved by using Pre-trained Language Models via fusing representations from different layers, constructing an auxiliary sentence, using multi-task learning, etc. However, to solve the AES task, previous works utilize shallow neural networks to learn essay representations and constrain calculated scores with regression loss or ranking loss, respectively. Since shallow neural networks trained on limited samples show poor performance to capture deep semantic of texts. And without an accurate scoring function, ranking loss and regression loss measures two different aspects of the calculated scores. To improve AES’s performance, we find a new way to fine-tune pre-trained language models with multiple losses of the same task. In this paper, we propose to utilize a pre-trained language model to learn text representations first. With scores calculated from the representations, mean square error loss and the batch-wise ListNet loss with dynamic weights constrain the scores simultaneously. We utilize Quadratic Weighted Kappa to evaluate our model on the Automated Student Assessment Prize dataset. Our model outperforms not only state-of-the-art neural models near 3 percent but also the latest statistic model. Especially on the two narrative prompts, our model performs much better than all other state-of-the-art models.</abstract>
      <url hash="bf231be8">2020.findings-emnlp.141</url>
      <doi>10.18653/v1/2020.findings-emnlp.141</doi>
    </paper>
    <paper id="142">
      <title>Neural Dialogue State Tracking with Temporally Expressive Networks</title>
      <author><first>Junfan</first><last>Chen</last></author>
      <author><first>Richong</first><last>Zhang</last></author>
      <author><first>Yongyi</first><last>Mao</last></author>
      <author><first>Jie</first><last>Xu</last></author>
      <pages>1570–1579</pages>
      <abstract>Dialogue state tracking (DST) is an important part of a spoken dialogue system. Existing DST models either ignore temporal feature dependencies across dialogue turns or fail to explicitly model temporal state dependencies in a dialogue. In this work, we propose Temporally Expressive Networks (TEN) to jointly model the two types of temporal dependencies in DST. The TEN model utilizes the power of recurrent networks and probabilistic graphical models. Evaluating on standard datasets, TEN is demonstrated to improve the accuracy of turn-level-state prediction and the state aggregation.</abstract>
      <url hash="957ce101">2020.findings-emnlp.142</url>
      <doi>10.18653/v1/2020.findings-emnlp.142</doi>
    </paper>
    <paper id="143">
      <title>Inferring about fraudulent collusion risk on <fixed-case>B</fixed-case>razilian public works contracts in official texts using a <fixed-case>B</fixed-case>i-<fixed-case>LSTM</fixed-case> approach</title>
      <author><first>Marcos</first><last>Lima</last></author>
      <author><first>Roberta</first><last>Silva</last></author>
      <author><first>Felipe</first><last>Lopes de Souza Mendes</last></author>
      <author><first>Leonardo</first><last>R. de Carvalho</last></author>
      <author><first>Aleteia</first><last>Araujo</last></author>
      <author><first>Flavio</first><last>de Barros Vidal</last></author>
      <pages>1580–1588</pages>
      <abstract>Public works procurements move US$ 10 billion yearly in Brazil and are a preferred field for collusion and fraud. Federal Police and audit agencies investigate collusion (bid-rigging), over-pricing, and delivery fraud in this field and efforts have been employed to early detect fraud and collusion on public works procurements. The current automatic methods of fraud detection use structured data to classification and usually do not involve annotated data. The use of NLP for this kind of application is rare. Our work introduces a new dataset formed by public procurement calls available on Brazilian official journal (Diário Oficial da União), using by 15,132,968 textual entries of which 1,907 are annotated risky entries. Both bottleneck deep neural network and BiLSTM shown competitive compared with classical classifiers and achieved better precision (93.0% and 92.4%, respectively), which signs improvements in a criminal fraud investigation.</abstract>
      <url hash="4c56afef">2020.findings-emnlp.143</url>
      <doi>10.18653/v1/2020.findings-emnlp.143</doi>
    </paper>
    <paper id="144">
      <title>Data-to-Text Generation with Style Imitation</title>
      <author><first>Shuai</first><last>Lin</last></author>
      <author><first>Wentao</first><last>Wang</last></author>
      <author><first>Zichao</first><last>Yang</last></author>
      <author><first>Xiaodan</first><last>Liang</last></author>
      <author><first>Frank F.</first><last>Xu</last></author>
      <author><first>Eric</first><last>Xing</last></author>
      <author><first>Zhiting</first><last>Hu</last></author>
      <pages>1589–1598</pages>
      <abstract>Recent neural approaches to data-to-text generation have mostly focused on improving content fidelity while lacking explicit control over writing styles (e.g., sentence structures, word choices). More traditional systems use templates to determine the realization of text. Yet manual or automatic construction of high-quality templates is difficult, and a template acting as hard constraints could harm content fidelity when it does not match the record perfectly. We study a new way of stylistic control by using existing sentences as “soft” templates. That is, a model learns to imitate the writing style of any given exemplar sentence, with automatic adaptions to faithfully describe the record. The problem is challenging due to the lack of parallel data. We develop a neural approach that includes a hybrid attention-copy mechanism, learns with weak supervisions, and is enhanced with a new content coverage constraint. We conduct experiments in restaurants and sports domains. Results show our approach achieves stronger performance than a range of comparison methods. Our approach balances well between content fidelity and style control given exemplars that match the records to varying degrees.</abstract>
      <url hash="82e1e95c">2020.findings-emnlp.144</url>
      <doi>10.18653/v1/2020.findings-emnlp.144</doi>
    </paper>
    <paper id="145">
      <title>Teaching Machine Comprehension with Compositional Explanations</title>
      <author><first>Qinyuan</first><last>Ye</last></author>
      <author><first>Xiao</first><last>Huang</last></author>
      <author><first>Elizabeth</first><last>Boschee</last></author>
      <author><first>Xiang</first><last>Ren</last></author>
      <pages>1599–1615</pages>
      <abstract>Advances in machine reading comprehension (MRC) rely heavily on the collection of large scale human-annotated examples in the form of (question, paragraph, answer) triples. In contrast, humans are typically able to generalize with only a few examples, relying on deeper underlying world knowledge, linguistic sophistication, and/or simply superior deductive powers. In this paper, we focus on “teaching” machines reading comprehension, using a small number of semi-structured explanations that explicitly inform machines why answer spans are correct. We extract structured variables and rules from explanations and compose neural module teachers that annotate instances for training downstream MRC models. We use learnable neural modules and soft logic to handle linguistic variation and overcome sparse coverage; the modules are jointly optimized with the MRC model to improve final performance. On the SQuAD dataset, our proposed method achieves 70.14% F1 score with supervision from 26 explanations, comparable to plain supervised learning using 1,100 labeled instances, yielding a 12x speed up.</abstract>
      <url hash="789186bf">2020.findings-emnlp.145</url>
      <doi>10.18653/v1/2020.findings-emnlp.145</doi>
    </paper>
    <paper id="146">
      <title>A Knowledge-Driven Approach to Classifying Object and Attribute Coreferences in Opinion Mining</title>
      <author><first>Jiahua</first><last>Chen</last></author>
      <author><first>Shuai</first><last>Wang</last></author>
      <author><first>Sahisnu</first><last>Mazumder</last></author>
      <author><first>Bing</first><last>Liu</last></author>
      <pages>1616–1626</pages>
      <abstract>Classifying and resolving coreferences of objects (e.g., product names) and attributes (e.g., product aspects) in opinionated reviews is crucial for improving the opinion mining performance. However, the task is challenging as one often needs to consider domain-specific knowledge (e.g., iPad is a tablet and has aspect resolution) to identify coreferences in opinionated reviews. Also, compiling a handcrafted and curated domain-specific knowledge base for each domain is very time consuming and arduous. This paper proposes an approach to automatically mine and leverage domain-specific knowledge for classifying objects and attribute coreferences. The approach extracts domain-specific knowledge from unlabeled review data and trains a knowledgeaware neural coreference classification model to leverage (useful) domain knowledge together with general commonsense knowledge for the task. Experimental evaluation on realworld datasets involving five domains (product types) shows the effectiveness of the approach</abstract>
      <url hash="c3bc5f51">2020.findings-emnlp.146</url>
      <doi>10.18653/v1/2020.findings-emnlp.146</doi>
    </paper>
    <paper id="147">
      <title><fixed-case>S</fixed-case>im<fixed-case>A</fixed-case>lign: High Quality Word Alignments Without Parallel Training Data Using Static and Contextualized Embeddings</title>
      <author><first>Masoud</first><last>Jalili Sabet</last></author>
      <author><first>Philipp</first><last>Dufter</last></author>
      <author><first>François</first><last>Yvon</last></author>
      <author><first>Hinrich</first><last>Schütze</last></author>
      <pages>1627–1643</pages>
      <abstract>Word alignments are useful for tasks like statistical and neural machine translation (NMT) and cross-lingual annotation projection. Statistical word aligners perform well, as do methods that extract alignments jointly with translations in NMT. However, most approaches require parallel training data and quality decreases as less training data is available. We propose word alignment methods that require no parallel data. The key idea is to leverage multilingual word embeddings – both static and contextualized – for word alignment. Our multilingual embeddings are created from monolingual data only without relying on any parallel data or dictionaries. We find that alignments created from embeddings are superior for four and comparable for two language pairs compared to those produced by traditional statistical aligners – even with abundant parallel data; e.g., contextualized embeddings achieve a word alignment F1 for English-German that is 5 percentage points higher than eflomal, a high-quality statistical aligner, trained on 100k parallel sentences.</abstract>
      <url hash="13c6cff3">2020.findings-emnlp.147</url>
      <doi>10.18653/v1/2020.findings-emnlp.147</doi>
    </paper>
    <paper id="148">
      <title><fixed-case>T</fixed-case>weet<fixed-case>E</fixed-case>val: Unified Benchmark and Comparative Evaluation for Tweet Classification</title>
      <author><first>Francesco</first><last>Barbieri</last></author>
      <author><first>Jose</first><last>Camacho-Collados</last></author>
      <author><first>Luis</first><last>Espinosa Anke</last></author>
      <author><first>Leonardo</first><last>Neves</last></author>
      <pages>1644–1650</pages>
      <abstract>The experimental landscape in natural language processing for social media is too fragmented. Each year, new shared tasks and datasets are proposed, ranging from classics like sentiment analysis to irony detection or emoji prediction. Therefore, it is unclear what the current state of the art is, as there is no standardized evaluation protocol, neither a strong set of baselines trained on such domain-specific data. In this paper, we propose a new evaluation framework (TweetEval) consisting of seven heterogeneous Twitter-specific classification tasks. We also provide a strong set of baselines as starting point, and compare different language modeling pre-training strategies. Our initial experiments show the effectiveness of starting off with existing pre-trained generic language models, and continue training them on Twitter corpora.</abstract>
      <url hash="fd8eefe0">2020.findings-emnlp.148</url>
      <doi>10.18653/v1/2020.findings-emnlp.148</doi>
    </paper>
    <paper id="149">
      <title>Octa: Omissions and Conflicts in Target-Aspect Sentiment Analysis</title>
      <author><first>Zhe</first><last>Zhang</last></author>
      <author><first>Chung-Wei</first><last>Hang</last></author>
      <author><first>Munindar</first><last>Singh</last></author>
      <pages>1651–1662</pages>
      <abstract>Sentiments in opinionated text are often determined by both aspects and target words (or targets). We observe that targets and aspects interrelate in subtle ways, often yielding conflicting sentiments. Thus, a naive aggregation of sentiments from aspects and targets treated separately, as in existing sentiment analysis models, impairs performance. We propose Octa, an approach that jointly considers aspects and targets when inferring sentiments. To capture and quantify relationships between targets and context words, Octa uses a selective self-attention mechanism that handles implicit or missing targets. Specifically, Octa involves two layers of attention mechanisms for, respectively, selective attention between targets and context words and attention over words based on aspects. On benchmark datasets, Octa outperforms leading models by a large margin, yielding (absolute) gains in accuracy of 1.6% to 4.3%.</abstract>
      <url hash="0349b765">2020.findings-emnlp.149</url>
      <doi>10.18653/v1/2020.findings-emnlp.149</doi>
    </paper>
    <paper id="150">
      <title>On the Language Neutrality of Pre-trained Multilingual Representations</title>
      <author><first>Jindřich</first><last>Libovický</last></author>
      <author><first>Rudolf</first><last>Rosa</last></author>
      <author><first>Alexander</first><last>Fraser</last></author>
      <pages>1663–1674</pages>
      <abstract>Multilingual contextual embeddings, such as multilingual BERT and XLM-RoBERTa, have proved useful for many multi-lingual tasks. Previous work probed the cross-linguality of the representations indirectly using zero-shot transfer learning on morphological and syntactic tasks. We instead investigate the language-neutrality of multilingual contextual embeddings directly and with respect to lexical semantics. Our results show that contextual embeddings are more language-neutral and, in general, more informative than aligned static word-type embeddings, which are explicitly trained for language neutrality. Contextual embeddings are still only moderately language-neutral by default, so we propose two simple methods for achieving stronger language neutrality: first, by unsupervised centering of the representation for each language and second, by fitting an explicit projection on small parallel data. Besides, we show how to reach state-of-the-art accuracy on language identification and match the performance of statistical methods for word alignment of parallel sentences without using parallel data.</abstract>
      <url hash="8e8743a7">2020.findings-emnlp.150</url>
      <attachment type="OptionalSupplementaryMaterial" hash="a28125ab">2020.findings-emnlp.150.OptionalSupplementaryMaterial.tgz</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.150</doi>
    </paper>
    <paper id="151">
      <title>Cost-effective Selection of Pretraining Data: A Case Study of Pretraining <fixed-case>BERT</fixed-case> on Social Media</title>
      <author><first>Xiang</first><last>Dai</last></author>
      <author><first>Sarvnaz</first><last>Karimi</last></author>
      <author><first>Ben</first><last>Hachey</last></author>
      <author><first>Cecile</first><last>Paris</last></author>
      <pages>1675–1681</pages>
      <abstract>Recent studies on domain-specific BERT models show that effectiveness on downstream tasks can be improved when models are pretrained on in-domain data. Often, the pretraining data used in these models are selected based on their subject matter, e.g., biology or computer science. Given the range of applications using social media text, and its unique language variety, we pretrain two models on tweets and forum text respectively, and empirically demonstrate the effectiveness of these two resources. In addition, we investigate how similarity measures can be used to nominate in-domain pretraining data. We publicly release our pretrained models at https://bit.ly/35RpTf0.</abstract>
      <url hash="eec9d22a">2020.findings-emnlp.151</url>
      <doi>10.18653/v1/2020.findings-emnlp.151</doi>
    </paper>
    <paper id="152">
      <title><fixed-case>T</fixed-case>opic<fixed-case>BERT</fixed-case> for Energy Efficient Document Classification</title>
      <author><first>Yatin</first><last>Chaudhary</last></author>
      <author><first>Pankaj</first><last>Gupta</last></author>
      <author><first>Khushbu</first><last>Saxena</last></author>
      <author><first>Vivek</first><last>Kulkarni</last></author>
      <author><first>Thomas</first><last>Runkler</last></author>
      <author><first>Hinrich</first><last>Schütze</last></author>
      <pages>1682–1690</pages>
      <abstract>Prior research notes that BERT’s computational cost grows quadratically with sequence length thus leading to longer training times, higher GPU memory constraints and carbon emissions. While recent work seeks to address these scalability issues at pre-training, these issues are also prominent in fine-tuning especially for long sequence tasks like document classification. Our work thus focuses on optimizing the computational cost of fine-tuning for document classification. We achieve this by complementary learning of both topic and language models in a unified framework, named TopicBERT. This significantly reduces the number of self-attention operations – a main performance bottleneck. Consequently, our model achieves a 1.4x ( 40%) speedup with 40% reduction in CO2 emission while retaining 99.9% performance over 5 datasets.</abstract>
      <url hash="c065d94b">2020.findings-emnlp.152</url>
      <doi>10.18653/v1/2020.findings-emnlp.152</doi>
    </paper>
    <paper id="153">
      <title>Improving Constituency Parsing with Span Attention</title>
      <author><first>Yuanhe</first><last>Tian</last></author>
      <author><first>Yan</first><last>Song</last></author>
      <author><first>Fei</first><last>Xia</last></author>
      <author><first>Tong</first><last>Zhang</last></author>
      <pages>1691–1703</pages>
      <abstract>Constituency parsing is a fundamental and important task for natural language understanding, where a good representation of contextual information can help this task. N-grams, which is a conventional type of feature for contextual information, have been demonstrated to be useful in many tasks, and thus could also be beneficial for constituency parsing if they are appropriately modeled. In this paper, we propose span attention for neural chart-based constituency parsing to leverage n-gram information. Considering that current chart-based parsers with Transformer-based encoder represent spans by subtraction of the hidden states at the span boundaries, which may cause information loss especially for long spans, we incorporate n-grams into span representations by weighting them according to their contributions to the parsing process. Moreover, we propose categorical span attention to further enhance the model by weighting n-grams within different length categories, and thus benefit long-sentence parsing. Experimental results on three widely used benchmark datasets demonstrate the effectiveness of our approach in parsing Arabic, Chinese, and English, where state-of-the-art performance is obtained by our approach on all of them.</abstract>
      <url hash="9d7fd658">2020.findings-emnlp.153</url>
      <attachment type="OptionalSupplementaryMaterial" hash="ec16aa93">2020.findings-emnlp.153.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.153</doi>
    </paper>
    <paper id="154">
      <title><fixed-case>R</fixed-case>eco<fixed-case>BERT</fixed-case>: A Catalog Language Model for Text-Based Recommendations</title>
      <author><first>Itzik</first><last>Malkiel</last></author>
      <author><first>Oren</first><last>Barkan</last></author>
      <author><first>Avi</first><last>Caciularu</last></author>
      <author><first>Noam</first><last>Razin</last></author>
      <author><first>Ori</first><last>Katz</last></author>
      <author><first>Noam</first><last>Koenigstein</last></author>
      <pages>1704–1714</pages>
      <abstract>Language models that utilize extensive self-supervised pre-training from unlabeled text, have recently shown to significantly advance the state-of-the-art performance in a variety of language understanding tasks. However, it is yet unclear if and how these recent models can be harnessed for conducting text-based recommendations. In this work, we introduce RecoBERT, a BERT-based approach for learning catalog-specialized language models for text-based item recommendations. We suggest novel training and inference procedures for scoring similarities between pairs of items, that don’t require item similarity labels. Both the training and the inference techniques were designed to utilize the unlabeled structure of textual catalogs, and minimize the discrepancy between them. By incorporating four scores during inference, RecoBERT can infer text-based item-to-item similarities more accurately than other techniques. In addition, we introduce a new language understanding task for wine recommendations using similarities based on professional wine reviews. As an additional contribution, we publish annotated recommendations dataset crafted by human wine experts. Finally, we evaluate RecoBERT and compare it to various state-of-the-art NLP models on wine and fashion recommendations tasks.</abstract>
      <url hash="57014973">2020.findings-emnlp.154</url>
      <attachment type="OptionalSupplementaryMaterial" hash="911400a1">2020.findings-emnlp.154.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.154</doi>
    </paper>
    <paper id="155">
      <title>Multi-Agent Mutual Learning at Sentence-Level and Token-Level for Neural Machine Translation</title>
      <author><first>Baohao</first><last>Liao</last></author>
      <author><first>Yingbo</first><last>Gao</last></author>
      <author><first>Hermann</first><last>Ney</last></author>
      <pages>1715–1724</pages>
      <abstract>Mutual learning, where multiple agents learn collaboratively and teach one another, has been shown to be an effective way to distill knowledge for image classification tasks. In this paper, we extend mutual learning to the machine translation task and operate at both the sentence-level and the token-level. Firstly, we co-train multiple agents by using the same parallel corpora. After convergence, each agent selects and learns its poorly predicted tokens from other agents. The poorly predicted tokens are determined by the acceptance-rejection sampling algorithm. Our experiments show that sequential mutual learning at the sentence-level and the token-level improves the results cumulatively. Absolute improvements compared to strong baselines are obtained on various translation tasks. On the IWSLT’14 German-English task, we get a new state-of-the-art BLEU score of 37.0. We also report a competitive result, 29.9 BLEU score, on the WMT’14 English-German task.</abstract>
      <url hash="2f569459">2020.findings-emnlp.155</url>
      <doi>10.18653/v1/2020.findings-emnlp.155</doi>
    </paper>
    <paper id="156">
      <title><fixed-case>D</fixed-case>om<fixed-case>BERT</fixed-case>: Domain-oriented Language Model for Aspect-based Sentiment Analysis</title>
      <author><first>Hu</first><last>Xu</last></author>
      <author><first>Bing</first><last>Liu</last></author>
      <author><first>Lei</first><last>Shu</last></author>
      <author><first>Philip</first><last>Yu</last></author>
      <pages>1725–1731</pages>
      <abstract>This paper focuses on learning domain-oriented language models driven by end tasks, which aims to combine the worlds of both general-purpose language models (such as ELMo and BERT) and domain-specific language understanding. We propose DomBERT, an extension of BERT to learn from both in-domain corpus and relevant domain corpora. This helps in learning domain language models with low-resources. Experiments are conducted on an assortment of tasks in aspect-based sentiment analysis (ABSA), demonstrating promising results.</abstract>
      <url hash="2cfbd858">2020.findings-emnlp.156</url>
      <doi>10.18653/v1/2020.findings-emnlp.156</doi>
    </paper>
    <paper id="157">
      <title><fixed-case>RMM</fixed-case>: A Recursive Mental Model for Dialogue Navigation</title>
      <author><first>Homero</first><last>Roman Roman</last></author>
      <author><first>Yonatan</first><last>Bisk</last></author>
      <author><first>Jesse</first><last>Thomason</last></author>
      <author><first>Asli</first><last>Celikyilmaz</last></author>
      <author><first>Jianfeng</first><last>Gao</last></author>
      <pages>1732–1745</pages>
      <abstract>Language-guided robots must be able to both ask humans questions and understand answers. Much existing work focuses only on the latter. In this paper, we go beyond instruction following and introduce a two-agent task where one agent navigates and asks questions that a second, guiding agent answers. Inspired by theory of mind, we propose the Recursive Mental Model (RMM). The navigating agent models the guiding agent to simulate answers given candidate generated questions. The guiding agent in turn models the navigating agent to simulate navigation steps it would take to generate answers. We use the progress agents make towards the goal as a reinforcement learning reward signal to directly inform not only navigation actions, but also both question and answer generation. We demonstrate that RMM enables better generalization to novel environments. Interlocutor modelling may be a way forward for human-agent RMM where robots need to both ask and answer questions.</abstract>
      <url hash="dd4ff542">2020.findings-emnlp.157</url>
      <doi>10.18653/v1/2020.findings-emnlp.157</doi>
    </paper>
    <paper id="158">
      <title>Will This Idea Spread Beyond Academia? Understanding Knowledge Transfer of Scientific Concepts across Text Corpora</title>
      <author><first>Hancheng</first><last>Cao</last></author>
      <author><first>Mengjie</first><last>Cheng</last></author>
      <author><first>Zhepeng</first><last>Cen</last></author>
      <author><first>Daniel</first><last>McFarland</last></author>
      <author><first>Xiang</first><last>Ren</last></author>
      <pages>1746–1757</pages>
      <abstract>What kind of basic research ideas are more likely to get applied in practice? There is a long line of research investigating patterns of knowledge transfer, but it generally focuses on documents as the unit of analysis and follow their transfer into practice for a specific scientific domain. Here we study translational research at the level of scientific concepts for all scientific fields. We do this through text mining and predictive modeling using three corpora: 38.6 million paper abstracts, 4 million patent documents, and 0.28 million clinical trials. We extract scientific concepts (i.e., phrases) from corpora as instantiations of “research ideas”, create concept-level features as motivated by literature, and then follow the trajectories of over 450,000 new concepts (emerged from 1995-2014) to identify factors that lead only a small proportion of these ideas to be used in inventions and drug trials. Results from our analysis suggest several mechanisms that distinguish which scientific concept will be adopted in practice, and which will not. We also demonstrate that our derived features can be used to explain and predict knowledge transfer with high accuracy. Our work provides greater understanding of knowledge transfer for researchers, practitioners, and government agencies interested in encouraging translational research.</abstract>
      <url hash="50f0f5a9">2020.findings-emnlp.158</url>
      <doi>10.18653/v1/2020.findings-emnlp.158</doi>
    </paper>
    <paper id="159">
      <title>Recurrent Inference in Text Editing</title>
      <author><first>Ning</first><last>Shi</last></author>
      <author><first>Ziheng</first><last>Zeng</last></author>
      <author><first>Haotian</first><last>Zhang</last></author>
      <author><first>Yichen</first><last>Gong</last></author>
      <pages>1758–1769</pages>
      <abstract>In neural text editing, prevalent sequence-to-sequence based approaches directly map the unedited text either to the edited text or the editing operations, in which the performance is degraded by the limited source text encoding and long, varying decoding steps. To address this problem, we propose a new inference method, Recurrence, that iteratively performs editing actions, significantly narrowing the problem space. In each iteration, encoding the partially edited text, Recurrence decodes the latent representation, generates an action of short, fixed-length, and applies the action to complete a single edit. For a comprehensive comparison, we introduce three types of text editing tasks: Arithmetic Operators Restoration (AOR), Arithmetic Equation Simplification (AES), Arithmetic Equation Correction (AEC). Extensive experiments on these tasks with varying difficulties demonstrate that Recurrence achieves improvements over conventional inference methods.</abstract>
      <url hash="22da648d">2020.findings-emnlp.159</url>
      <attachment type="OptionalSupplementaryMaterial" hash="5ac90db6">2020.findings-emnlp.159.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.159</doi>
    </paper>
    <paper id="160">
      <title>An Empirical Exploration of Local Ordering Pre-training for Structured Prediction</title>
      <author><first>Zhisong</first><last>Zhang</last></author>
      <author><first>Xiang</first><last>Kong</last></author>
      <author><first>Lori</first><last>Levin</last></author>
      <author><first>Eduard</first><last>Hovy</last></author>
      <pages>1770–1783</pages>
      <abstract>Recently, pre-training contextualized encoders with language model (LM) objectives has been shown an effective semi-supervised method for structured prediction. In this work, we empirically explore an alternative pre-training method for contextualized encoders. Instead of predicting words in LMs, we “mask out” and predict word order information, with a local ordering strategy and word-selecting objectives. With evaluations on three typical structured prediction tasks (dependency parsing, POS tagging, and NER) over four languages (English, Finnish, Czech, and Italian), we show that our method is consistently beneficial. We further conduct detailed error analysis, including one that examines a specific type of parsing error where the head is misidentified. The results show that pre-trained contextual encoders can bring improvements in a structured way, suggesting that they may be able to capture higher-order patterns and feature combinations from unlabeled data.</abstract>
      <url hash="afd95395">2020.findings-emnlp.160</url>
      <doi>10.18653/v1/2020.findings-emnlp.160</doi>
    </paper>
    <paper id="161">
      <title>Unsupervised Extractive Summarization by Pre-training Hierarchical Transformers</title>
      <author><first>Shusheng</first><last>Xu</last></author>
      <author><first>Xingxing</first><last>Zhang</last></author>
      <author><first>Yi</first><last>Wu</last></author>
      <author><first>Furu</first><last>Wei</last></author>
      <author><first>Ming</first><last>Zhou</last></author>
      <pages>1784–1795</pages>
      <abstract>Unsupervised extractive document summarization aims to select important sentences from a document without using labeled summaries during training. Existing methods are mostly graph-based with sentences as nodes and edge weights measured by sentence similarities. In this work, we find that transformer attentions can be used to rank sentences for unsupervised extractive summarization. Specifically, we first pre-train a hierarchical transformer model using unlabeled documents only. Then we propose a method to rank sentences using sentence-level self-attentions and pre-training objectives. Experiments on CNN/DailyMail and New York Times datasets show our model achieves state-of-the-art performance on unsupervised summarization. We also find in experiments that our model is less dependent on sentence positions. When using a linear combination of our model and a recent unsupervised model explicitly modeling sentence positions, we obtain even better results.</abstract>
      <url hash="820e6e2c">2020.findings-emnlp.161</url>
      <attachment type="OptionalSupplementaryMaterial" hash="acbcd3d1">2020.findings-emnlp.161.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.161</doi>
    </paper>
    <paper id="162">
      <title>Active Learning Approaches to Enhancing Neural Machine Translation</title>
      <author><first>Yuekai</first><last>Zhao</last></author>
      <author><first>Haoran</first><last>Zhang</last></author>
      <author><first>Shuchang</first><last>Zhou</last></author>
      <author><first>Zhihua</first><last>Zhang</last></author>
      <pages>1796–1806</pages>
      <abstract>Active learning is an efficient approach for mitigating data dependency when training neural machine translation (NMT) models. In this paper, we explore new training frameworks by incorporating active learning into various techniques such as transfer learning and iterative back-translation (IBT) under a limited human translation budget. We design a word frequency based acquisition function and combine it with a strong uncertainty based method. The combined method steadily outperforms all other acquisition functions in various scenarios. As far as we know, we are the first to do a large-scale study on actively training Transformer for NMT. Specifically, with a human translation budget of only 20% of the original parallel corpus, we manage to surpass Transformer trained on the entire parallel corpus in three language pairs.</abstract>
      <url hash="318b5e93">2020.findings-emnlp.162</url>
      <attachment type="OptionalSupplementaryMaterial" hash="899f419a">2020.findings-emnlp.162.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.162</doi>
    </paper>
    <paper id="163">
      <title><fixed-case>AGIF</fixed-case>: An Adaptive Graph-Interactive Framework for Joint Multiple Intent Detection and Slot Filling</title>
      <author><first>Libo</first><last>Qin</last></author>
      <author><first>Xiao</first><last>Xu</last></author>
      <author><first>Wanxiang</first><last>Che</last></author>
      <author><first>Ting</first><last>Liu</last></author>
      <pages>1807–1816</pages>
      <abstract>In real-world scenarios, users usually have multiple intents in the same utterance. Unfortunately, most spoken language understanding (SLU) models either mainly focused on the single intent scenario, or simply incorporated an overall intent context vector for all tokens, ignoring the fine-grained multiple intents information integration for token-level slot prediction. In this paper, we propose an Adaptive Graph-Interactive Framework (AGIF) for joint multiple intent detection and slot filling, where we introduce an intent-slot graph interaction layer to model the strong correlation between the slot and intents. Such an interaction layer is applied to each token adaptively, which has the advantage to automatically extract the relevant intents information, making a fine-grained intent information integration for the token-level slot prediction. Experimental results on three multi-intent datasets show that our framework obtains substantial improvement and achieves the state-of-the-art performance. In addition, our framework achieves new state-of-the-art performance on two single-intent datasets.</abstract>
      <url hash="2c95c22a">2020.findings-emnlp.163</url>
      <doi>10.18653/v1/2020.findings-emnlp.163</doi>
    </paper>
    <paper id="164">
      <title>Continual Learning Long Short Term Memory</title>
      <author><first>Xin</first><last>Guo</last></author>
      <author><first>Yu</first><last>Tian</last></author>
      <author><first>Qinghan</first><last>Xue</last></author>
      <author><first>Panos</first><last>Lampropoulos</last></author>
      <author><first>Steven</first><last>Eliuk</last></author>
      <author><first>Kenneth</first><last>Barner</last></author>
      <author><first>Xiaolong</first><last>Wang</last></author>
      <pages>1817–1822</pages>
      <abstract>Catastrophic forgetting in neural networks indicates the performance decreasing of deep learning models on previous tasks while learning new tasks. To address this problem, we propose a novel Continual Learning Long Short Term Memory (CL-LSTM) cell in Recurrent Neural Network (RNN) in this paper. CL-LSTM considers not only the state of each individual task’s output gates but also the correlation of the states between tasks, so that the deep learning models can incrementally learn new tasks without catastrophically forgetting previously tasks. Experimental results demonstrate significant improvements of CL-LSTM over state-of-the-art approaches on spoken language understanding (SLU) tasks.</abstract>
      <url hash="44fe9cbc">2020.findings-emnlp.164</url>
      <attachment type="OptionalSupplementaryMaterial" hash="3055a9f5">2020.findings-emnlp.164.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.164</doi>
    </paper>
    <paper id="165">
      <title><fixed-case>C</fixed-case>ommon<fixed-case>G</fixed-case>en: A Constrained Text Generation Challenge for Generative Commonsense Reasoning</title>
      <author><first>Bill Yuchen</first><last>Lin</last></author>
      <author><first>Wangchunshu</first><last>Zhou</last></author>
      <author><first>Ming</first><last>Shen</last></author>
      <author><first>Pei</first><last>Zhou</last></author>
      <author><first>Chandra</first><last>Bhagavatula</last></author>
      <author><first>Yejin</first><last>Choi</last></author>
      <author><first>Xiang</first><last>Ren</last></author>
      <pages>1823–1840</pages>
      <abstract>Recently, large-scale pre-trained language models have demonstrated impressive performance on several commonsense-reasoning benchmark datasets. However, building machines with commonsense to compose realistically plausible sentences remains challenging. In this paper, we present a constrained text generation task, CommonGen associated with a benchmark dataset, to explicitly test machines for the ability of generative commonsense reasoning. Given a set of common concepts (e.g., dog, frisbee, catch, throw); the task is to generate a coherent sentence describing an everyday scenario using these concepts (e.g., “a man throws a frisbee and his dog catches it”). The CommonGen task is challenging because it inherently requires 1) relational reasoning with background commonsense knowledge and 2) compositional generalization ability to work on unseen concept combinations. Our dataset, constructed through a combination of crowdsourced and existing caption corpora, consists of 77k commonsense descriptions over 35k unique concept-sets. Experiments show that there is a large gap between state-of-the-art text generation models (e.g., T5) and human performance (31.6% v.s. 63.5% in SPICE metric). Furthermore, we demonstrate that the learned generative commonsense reasoning capability can be transferred to improve downstream tasks such as CommonsenseQA (76.9% to 78.4 in dev accuracy) by generating additional context.</abstract>
      <url hash="0b9ef604">2020.findings-emnlp.165</url>
      <doi>10.18653/v1/2020.findings-emnlp.165</doi>
    </paper>
    <paper id="166">
      <title>Constrained Decoding for Computationally Efficient Named Entity Recognition Taggers</title>
      <author><first>Brian</first><last>Lester</last></author>
      <author><first>Daniel</first><last>Pressel</last></author>
      <author><first>Amy</first><last>Hemmeter</last></author>
      <author><first>Sagnik</first><last>Ray Choudhury</last></author>
      <author><first>Srinivas</first><last>Bangalore</last></author>
      <pages>1841–1848</pages>
      <abstract>Current state-of-the-art models for named entity recognition (NER) are neural models with a conditional random field (CRF) as the final layer. Entities are represented as per-token labels with a special structure in order to decode them into spans. Current work eschews prior knowledge of how the span encoding scheme works and relies on the CRF learning which transitions are illegal and which are not to facilitate global coherence. We find that by constraining the output to suppress illegal transitions we can train a tagger with a cross-entropy loss twice as fast as a CRF with differences in F1 that are statistically insignificant, effectively eliminating the need for a CRF. We analyze the dynamics of tag co-occurrence to explain when these constraints are most effective and provide open source implementations of our tagger in both PyTorch and TensorFlow.</abstract>
      <url hash="87619392">2020.findings-emnlp.166</url>
      <attachment type="OptionalSupplementaryMaterial" hash="000864b8">2020.findings-emnlp.166.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.166</doi>
    </paper>
    <paper id="167">
      <title>On the Potential of Lexico-logical Alignments for Semantic Parsing to <fixed-case>SQL</fixed-case> Queries</title>
      <author><first>Tianze</first><last>Shi</last></author>
      <author><first>Chen</first><last>Zhao</last></author>
      <author><first>Jordan</first><last>Boyd-Graber</last></author>
      <author><first>Hal</first><last>Daumé III</last></author>
      <author><first>Lillian</first><last>Lee</last></author>
      <pages>1849–1864</pages>
      <abstract>Large-scale semantic parsing datasets annotated with logical forms have enabled major advances in supervised approaches. But can richer supervision help even more? To explore the utility of fine-grained, lexical-level supervision, we introduce SQUALL, a dataset that enriches 11,276 WIKITABLEQUESTIONS English-language questions with manually created SQL equivalents plus alignments between SQL and question fragments. Our annotation enables new training possibilities for encoderdecoder models, including approaches from machine translation previously precluded by the absence of alignments. We propose and test two methods: (1) supervised attention; (2) adopting an auxiliary objective of disambiguating references in the input queries to table columns. In 5-fold cross validation, these strategies improve over strong baselines by 4.4% execution accuracy. Oracle experiments suggest that annotated alignments can support further accuracy gains of up to 23.9%.</abstract>
      <url hash="52f51d24">2020.findings-emnlp.167</url>
      <doi>10.18653/v1/2020.findings-emnlp.167</doi>
    </paper>
    <paper id="168">
      <title><fixed-case>TED</fixed-case>: A Pretrained Unsupervised Summarization Model with Theme Modeling and Denoising</title>
      <author><first>Ziyi</first><last>Yang</last></author>
      <author><first>Chenguang</first><last>Zhu</last></author>
      <author><first>Robert</first><last>Gmyr</last></author>
      <author><first>Michael</first><last>Zeng</last></author>
      <author><first>Xuedong</first><last>Huang</last></author>
      <author><first>Eric</first><last>Darve</last></author>
      <pages>1865–1874</pages>
      <abstract>Text summarization aims to extract essential information from a piece of text and transform the text into a concise version. Existing unsupervised abstractive summarization models leverage recurrent neural networks framework while the recently proposed transformer exhibits much more capability. Moreover, most of previous summarization models ignore abundant unlabeled corpora resources available for pretraining. In order to address these issues, we propose TED, a transformer-based unsupervised abstractive summarization system with pretraining on large-scale data. We first leverage the lead bias in news articles to pretrain the model on millions of unlabeled corpora. Next, we finetune TED on target domains through theme modeling and a denoising autoencoder to enhance the quality of generated summaries. Notably, TED outperforms all unsupervised abstractive baselines on NYT, CNN/DM and English Gigaword datasets with various document styles. Further analysis shows that the summaries generated by TED are highly abstractive, and each component in the objective function of TED is highly effective.</abstract>
      <url hash="b329f2c8">2020.findings-emnlp.168</url>
      <attachment type="OptionalSupplementaryMaterial" hash="710b9bcb">2020.findings-emnlp.168.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.168</doi>
    </paper>
    <paper id="169">
      <title>Improving End-to-End <fixed-case>B</fixed-case>angla Speech Recognition with Semi-supervised Training</title>
      <author><first>Nafis</first><last>Sadeq</last></author>
      <author><first>Nafis Tahmid</first><last>Chowdhury</last></author>
      <author><first>Farhan Tanvir</first><last>Utshaw</last></author>
      <author><first>Shafayat</first><last>Ahmed</last></author>
      <author><first>Muhammad Abdullah</first><last>Adnan</last></author>
      <pages>1875–1883</pages>
      <abstract>Automatic speech recognition systems usually require large annotated speech corpus for training. The manual annotation of a large corpus is very difficult. It can be very helpful to use unsupervised and semi-supervised learning methods in addition to supervised learning. In this work, we focus on using a semi-supervised training approach for Bangla Speech Recognition that can exploit large unpaired audio and text data. We encode speech and text data in an intermediate domain and propose a novel loss function based on the global encoding distance between encoded data to guide the semi-supervised training. Our proposed method reduces the Word Error Rate (WER) of the system from 37% to 31.9%.</abstract>
      <url hash="bb28d585">2020.findings-emnlp.169</url>
      <doi>10.18653/v1/2020.findings-emnlp.169</doi>
    </paper>
    <paper id="170">
      <title>No Gestures Left Behind: Learning Relationships between Spoken Language and Freeform Gestures</title>
      <author><first>Chaitanya</first><last>Ahuja</last></author>
      <author><first>Dong Won</first><last>Lee</last></author>
      <author><first>Ryo</first><last>Ishii</last></author>
      <author><first>Louis-Philippe</first><last>Morency</last></author>
      <pages>1884–1895</pages>
      <abstract>We study relationships between spoken language and co-speech gestures in context of two key challenges. First, distributions of text and gestures are inherently skewed making it important to model the long tail. Second, gesture predictions are made at a subword level, making it important to learn relationships between language and acoustic cues. We introduce AISLe, which combines adversarial learning with importance sampling to strike a balance between precision and coverage. We propose the use of a multimodal multiscale attention block to perform subword alignment without the need of explicit alignment between language and acoustic cues. Finally, to empirically study the importance of language in this task, we extend the dataset proposed in Ahuja et al. (2020) with automatically extracted transcripts for audio signals. We substantiate the effectiveness of our approach through large-scale quantitative and user studies, which show that our proposed methodology significantly outperforms previous state-of-the-art approaches for gesture generation. Link to code, data and videos: https://github.com/chahuja/aisle</abstract>
      <url hash="54f989b9">2020.findings-emnlp.170</url>
      <attachment type="OptionalSupplementaryMaterial" hash="c0a75a6f">2020.findings-emnlp.170.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.170</doi>
    </paper>
    <paper id="171">
      <title><fixed-case>UNIFIEDQA</fixed-case>: Crossing Format Boundaries with a Single <fixed-case>QA</fixed-case> System</title>
      <author><first>Daniel</first><last>Khashabi</last></author>
      <author><first>Sewon</first><last>Min</last></author>
      <author><first>Tushar</first><last>Khot</last></author>
      <author><first>Ashish</first><last>Sabharwal</last></author>
      <author><first>Oyvind</first><last>Tafjord</last></author>
      <author><first>Peter</first><last>Clark</last></author>
      <author><first>Hannaneh</first><last>Hajishirzi</last></author>
      <pages>1896–1907</pages>
      <abstract>Question answering (QA) tasks have been posed using a variety of formats, such as extractive span selection, multiple choice, etc. This has led to format-specialized models, and even to an implicit division in the QA community. We argue that such boundaries are artificial and perhaps unnecessary, given the reasoning abilities we seek to teach are not governed by the format. As evidence, we use the latest advances in language modeling to build a single pre-trained QA model, UNIFIEDQA, that performs well across 19 QA datasets spanning 4 diverse formats. UNIFIEDQA performs on par with 8 different models that were trained on individual datasets themselves. Even when faced with 12 unseen datasets of observed formats, UNIFIEDQA performs surprisingly well, showing strong generalization from its outof-format training data. Finally, simply finetuning this pre trained QA model into specialized models results in a new state of the art on 10 factoid and commonsense question answering datasets, establishing UNIFIEDQA as a strong starting point for building QA systems.</abstract>
      <url hash="2c90bd29">2020.findings-emnlp.171</url>
      <attachment type="OptionalSupplementaryMaterial" hash="640db1a9">2020.findings-emnlp.171.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.171</doi>
    </paper>
    <paper id="172">
      <title>Robust and Interpretable Grounding of Spatial References with Relation Networks</title>
      <author><first>Tsung-Yen</first><last>Yang</last></author>
      <author><first>Andrew</first><last>Lan</last></author>
      <author><first>Karthik</first><last>Narasimhan</last></author>
      <pages>1908–1923</pages>
      <abstract>Learning representations of spatial references in natural language is a key challenge in tasks like autonomous navigation and robotic manipulation. Recent work has investigated various neural architectures for learning multi-modal representations for spatial concepts. However, the lack of explicit reasoning over entities makes such approaches vulnerable to noise in input text or state observations. In this paper, we develop effective models for understanding spatial references in text that are robust and interpretable, without sacrificing performance. We design a text-conditioned relation network whose parameters are dynamically computed with a cross-modal attention module to capture fine-grained spatial relations between entities. This design choice provides interpretability of learned intermediate outputs. Experiments across three tasks demonstrate that our model achieves superior performance, with a 17% improvement in predicting goal locations and a 15% improvement in robustness compared to state-of-the-art systems.</abstract>
      <url hash="f1c2e4cc">2020.findings-emnlp.172</url>
      <doi>10.18653/v1/2020.findings-emnlp.172</doi>
    </paper>
    <paper id="173">
      <title>Pragmatic Issue-Sensitive Image Captioning</title>
      <author><first>Allen</first><last>Nie</last></author>
      <author><first>Reuben</first><last>Cohn-Gordon</last></author>
      <author><first>Christopher</first><last>Potts</last></author>
      <pages>1924–1938</pages>
      <abstract>Image captioning systems need to produce texts that are not only true but also relevant in that they are properly aligned with the current issues. For instance, in a newspaper article about a sports event, a caption that not only identifies the player in a picture but also comments on their ethnicity could create unwanted reader reactions. To address this, we propose Issue-Sensitive Image Captioning (ISIC). In ISIC, the captioner is given a target image and an issue, which is a set of images partitioned in a way that specifies what information is relevant. For the sports article, we could construct a partition that places images into equivalence classes based on player position. To model this task, we use an extension of the Rational Speech Acts model. Our extension is built on top of state-of-the-art pretrained neural image captioners and explicitly uses image partitions to control caption generation. In both automatic and human evaluations, we show that these models generate captions that are descriptive and issue-sensitive. Finally, we show how ISIC can complement and enrich the related task of Visual Question Answering.</abstract>
      <url hash="337ff291">2020.findings-emnlp.173</url>
      <doi>10.18653/v1/2020.findings-emnlp.173</doi>
    </paper>
    <paper id="174">
      <title><fixed-case>PTUM</fixed-case>: Pre-training User Model from Unlabeled User Behaviors via Self-supervision</title>
      <author><first>Chuhan</first><last>Wu</last></author>
      <author><first>Fangzhao</first><last>Wu</last></author>
      <author><first>Tao</first><last>Qi</last></author>
      <author><first>Jianxun</first><last>Lian</last></author>
      <author><first>Yongfeng</first><last>Huang</last></author>
      <author><first>Xing</first><last>Xie</last></author>
      <pages>1939–1944</pages>
      <abstract>User modeling is critical for many personalized web services. Many existing methods model users based on their behaviors and the labeled data of target tasks. However, these methods cannot exploit useful information in unlabeled user behavior data, and their performance may be not optimal when labeled data is scarce. Motivated by pre-trained language models which are pre-trained on large-scale unlabeled corpus to empower many downstream tasks, in this paper we propose to pre-train user models from large-scale unlabeled user behaviors data. We propose two self-supervision tasks for user model pre-training. The first one is masked behavior prediction, which can model the relatedness between historical behaviors. The second one is next K behavior prediction, which can model the relatedness between past and future behaviors. The pre-trained user models are finetuned in downstream tasks to learn task-specific user representations. Experimental results on two real-world datasets validate the effectiveness of our proposed user model pre-training method.</abstract>
      <url hash="e5101132">2020.findings-emnlp.174</url>
      <doi>10.18653/v1/2020.findings-emnlp.174</doi>
    </paper>
    <paper id="175">
      <title>Adversarial Subword Regularization for Robust Neural Machine Translation</title>
      <author><first>Jungsoo</first><last>Park</last></author>
      <author><first>Mujeen</first><last>Sung</last></author>
      <author><first>Jinhyuk</first><last>Lee</last></author>
      <author><first>Jaewoo</first><last>Kang</last></author>
      <pages>1945–1953</pages>
      <abstract>Exposing diverse subword segmentations to neural machine translation (NMT) models often improves the robustness of machine translation as NMT models can experience various subword candidates. However, the diversification of subword segmentations mostly relies on the pre-trained subword language models from which erroneous segmentations of unseen words are less likely to be sampled. In this paper, we present adversarial subword regularization (ADVSR) to study whether gradient signals during training can be a substitute criterion for exposing diverse subword segmentations. We experimentally show that our model-based adversarial samples effectively encourage NMT models to be less sensitive to segmentation errors and improve the performance of NMT models in low-resource and out-domain datasets.</abstract>
      <url hash="6a8f3bd4">2020.findings-emnlp.175</url>
      <doi>10.18653/v1/2020.findings-emnlp.175</doi>
    </paper>
    <paper id="176">
      <title>Learning Visual-Semantic Embeddings for Reporting Abnormal Findings on Chest <fixed-case>X</fixed-case>-rays</title>
      <author><first>Jianmo</first><last>Ni</last></author>
      <author><first>Chun-Nan</first><last>Hsu</last></author>
      <author><first>Amilcare</first><last>Gentili</last></author>
      <author><first>Julian</first><last>McAuley</last></author>
      <pages>1954–1960</pages>
      <abstract>Automatic medical image report generation has drawn growing attention due to its potential to alleviate radiologists’ workload. Existing work on report generation often trains encoder-decoder networks to generate complete reports. However, such models are affected by data bias (e.g. label imbalance) and face common issues inherent in text generation models (e.g. repetition). In this work, we focus on reporting abnormal findings on radiology images; instead of training on complete radiology reports, we propose a method to identify abnormal findings from the reports in addition to grouping them with unsupervised clustering and minimal rules. We formulate the task as cross-modal retrieval and propose Conditional Visual-Semantic Embeddings to align images and fine-grained abnormal findings in a joint embedding space. We demonstrate that our method is able to retrieve abnormal findings and outperforms existing generation models on both clinical correctness and text generation metrics.</abstract>
      <url hash="94095774">2020.findings-emnlp.176</url>
      <doi>10.18653/v1/2020.findings-emnlp.176</doi>
    </paper>
    <paper id="177">
      <title><fixed-case>S</fixed-case>yn<fixed-case>ET</fixed-case>: Synonym Expansion using Transitivity</title>
      <author><first>Jiale</first><last>Yu</last></author>
      <author><first>Yongliang</first><last>Shen</last></author>
      <author><first>Xinyin</first><last>Ma</last></author>
      <author><first>Chenghao</first><last>Jia</last></author>
      <author><first>Chen</first><last>Chen</last></author>
      <author><first>Weiming</first><last>Lu</last></author>
      <pages>1961–1970</pages>
      <abstract>In this paper, we study a new task of synonym expansion using transitivity, and propose a novel approach named SynET, which considers both the contexts of two given synonym pairs. It introduces an auxiliary task to reduce the impact of noisy sentences, and proposes a Multi-Perspective Entity Matching Network to match entities from multiple perspectives. Extensive experiments on a real-world dataset show the effectiveness of our approach.</abstract>
      <url hash="1a8249f2">2020.findings-emnlp.177</url>
      <doi>10.18653/v1/2020.findings-emnlp.177</doi>
    </paper>
    <paper id="178">
      <title>Scheduled <fixed-case>D</fixed-case>rop<fixed-case>H</fixed-case>ead: A Regularization Method for Transformer Models</title>
      <author><first>Wangchunshu</first><last>Zhou</last></author>
      <author><first>Tao</first><last>Ge</last></author>
      <author><first>Furu</first><last>Wei</last></author>
      <author><first>Ming</first><last>Zhou</last></author>
      <author><first>Ke</first><last>Xu</last></author>
      <pages>1971–1980</pages>
      <abstract>We introduce DropHead, a structured dropout method specifically designed for regularizing the multi-head attention mechanism which is a key component of transformer. In contrast to the conventional dropout mechanism which randomly drops units or connections, DropHead drops entire attention heads during training to prevent the multi-head attention model from being dominated by a small portion of attention heads. It can help reduce the risk of overfitting and allow the models to better benefit from the multi-head attention. Given the interaction between multi-headedness and training dynamics, we further propose a novel dropout rate scheduler to adjust the dropout rate of DropHead throughout training, which results in a better regularization effect. Experimental results demonstrate that our proposed approach can improve transformer models by 0.9 BLEU score on WMT14 En-De translation task and around 1.0 accuracy for various text classification tasks.</abstract>
      <url hash="3fc170c9">2020.findings-emnlp.178</url>
      <doi>10.18653/v1/2020.findings-emnlp.178</doi>
    </paper>
    <paper id="179">
      <title>Multi-Turn Dialogue Generation in <fixed-case>E</fixed-case>-Commerce Platform with the Context of Historical Dialogue</title>
      <author><first>WeiSheng</first><last>Zhang</last></author>
      <author><first>Kaisong</first><last>Song</last></author>
      <author><first>Yangyang</first><last>Kang</last></author>
      <author><first>Zhongqing</first><last>Wang</last></author>
      <author><first>Changlong</first><last>Sun</last></author>
      <author><first>Xiaozhong</first><last>Liu</last></author>
      <author><first>Shoushan</first><last>Li</last></author>
      <author><first>Min</first><last>Zhang</last></author>
      <author><first>Luo</first><last>Si</last></author>
      <pages>1981–1990</pages>
      <abstract>As an important research topic, customer service dialogue generation tends to generate generic seller responses by leveraging current dialogue information. In this study, we propose a novel and extensible dialogue generation method by leveraging sellers’ historical dialogue information, which can be both accessible and informative. By utilizing innovative historical dialogue representation learning and historical dialogue selection mechanism, the proposed model is capable of detecting most related responses from sellers’ historical dialogues, which can further enhance the current dialogue generation quality. Unlike prior dialogue generation efforts, we treat each seller’s historical dialogues as a list of Customer-Seller utterance pairs and allow the model to measure their different importance, and copy words directly from most relevant pairs. Extensive experimental results show that the proposed approach can generate high-quality responses that cater to specific sellers’ characteristics and exhibit consistent superiority over baselines on a real-world multi-turn customer service dialogue dataset.</abstract>
      <url hash="01c6fe47">2020.findings-emnlp.179</url>
      <doi>10.18653/v1/2020.findings-emnlp.179</doi>
    </paper>
    <paper id="180">
      <title>Automatically Identifying Gender Issues in Machine Translation using Perturbations</title>
      <author><first>Hila</first><last>Gonen</last></author>
      <author><first>Kellie</first><last>Webster</last></author>
      <pages>1991–1995</pages>
      <abstract>The successful application of neural methods to machine translation has realized huge quality advances for the community. With these improvements, many have noted outstanding challenges, including the modeling and treatment of gendered language. While previous studies have identified issues using synthetic examples, we develop a novel technique to mine examples from real world data to explore challenges for deployed systems. We use our method to compile an evaluation benchmark spanning examples for four languages from three language families, which we publicly release to facilitate research. The examples in our benchmark expose where model representations are gendered, and the unintended consequences these gendered representations can have in downstream application.</abstract>
      <url hash="b481f716">2020.findings-emnlp.180</url>
      <doi>10.18653/v1/2020.findings-emnlp.180</doi>
    </paper>
    <paper id="181">
      <title>Ruler: Data Programming by Demonstration for Document Labeling</title>
      <author><first>Sara</first><last>Evensen</last></author>
      <author><first>Chang</first><last>Ge</last></author>
      <author><first>Cagatay</first><last>Demiralp</last></author>
      <pages>1996–2005</pages>
      <abstract>Data programming aims to reduce the cost of curating training data by encoding domain knowledge as labeling functions over source data. As such it not only requires domain expertise but also programming experience, a skill that many subject matter experts lack. Additionally, generating functions by enumerating rules is not only time consuming but also inherently difficult, even for people with programming experience. In this paper we introduce Ruler, an interactive system that synthesizes labeling rules using span-level interactive demonstrations over document examples. Ruler is a first-of-a-kind implementation of data programming by demonstration (DPBD). This new framework aims to relieve users from the burden of writing labeling functions, enabling them to focus on higher-level semantic analysis, such as identifying relevant signals for the labeling task. We compare Ruler with conventional data programming through a user study conducted with 10 data scientists who were asked to create labeling functions for sentiment and spam classification tasks. Results show Ruler is easier to learn and to use, and that it offers higher overall user-satisfaction while providing model performances comparable to those achieved by conventional data programming.</abstract>
      <url hash="7a5e99af">2020.findings-emnlp.181</url>
      <doi>10.18653/v1/2020.findings-emnlp.181</doi>
    </paper>
    <paper id="182">
      <title>Dual Reconstruction: a Unifying Objective for Semi-Supervised Neural Machine Translation</title>
      <author><first>Weijia</first><last>Xu</last></author>
      <author><first>Xing</first><last>Niu</last></author>
      <author><first>Marine</first><last>Carpuat</last></author>
      <pages>2006–2020</pages>
      <abstract>While Iterative Back-Translation and Dual Learning effectively incorporate monolingual training data in neural machine translation, they use different objectives and heuristic gradient approximation strategies, and have not been extensively compared. We introduce a novel dual reconstruction objective that provides a unified view of Iterative Back-Translation and Dual Learning. It motivates a theoretical analysis and controlled empirical study on German-English and Turkish-English tasks, which both suggest that Iterative Back-Translation is more effective than Dual Learning despite its relative simplicity.</abstract>
      <url hash="c7e7a7a6">2020.findings-emnlp.182</url>
      <doi>10.18653/v1/2020.findings-emnlp.182</doi>
    </paper>
    <paper id="183">
      <title>Focus-Constrained Attention Mechanism for <fixed-case>CVAE</fixed-case>-based Response Generation</title>
      <author><first>Zhi</first><last>Cui</last></author>
      <author><first>Yanran</first><last>Li</last></author>
      <author><first>Jiayi</first><last>Zhang</last></author>
      <author><first>Jianwei</first><last>Cui</last></author>
      <author><first>Chen</first><last>Wei</last></author>
      <author><first>Bin</first><last>Wang</last></author>
      <pages>2021–2030</pages>
      <abstract>To model diverse responses for a given post, one promising way is to introduce a latent variable into Seq2Seq models. The latent variable is supposed to capture the discourse-level information and encourage the informativeness of target responses. However, such discourse-level information is often too coarse for the decoder to be utilized. To tackle it, our idea is to transform the coarse-grained discourse-level information into fine-grained word-level information. Specifically, we firstly measure the semantic concentration of corresponding target response on the post words by introducing a fine-grained focus signal. Then, we propose a focus-constrained attention mechanism to take full advantage of focus in well aligning the input to the target response. The experimental results demonstrate that by exploiting the fine-grained signal, our model can generate more diverse and informative responses compared with several state-of-the-art models.</abstract>
      <url hash="abfec9de">2020.findings-emnlp.183</url>
      <doi>10.18653/v1/2020.findings-emnlp.183</doi>
    </paper>
    <paper id="184">
      <title>Chunk-based <fixed-case>C</fixed-case>hinese Spelling Check with Global Optimization</title>
      <author><first>Zuyi</first><last>Bao</last></author>
      <author><first>Chen</first><last>Li</last></author>
      <author><first>Rui</first><last>Wang</last></author>
      <pages>2031–2040</pages>
      <abstract>Chinese spelling check is a challenging task due to the characteristics of the Chinese language, such as the large character set, no word boundary, and short word length. On the one hand, most of the previous works only consider corrections with similar character pronunciation or shape, failing to correct visually and phonologically irrelevant typos. On the other hand, pipeline-style architectures are widely adopted to deal with different types of spelling errors in individual modules, which is difficult to optimize. In order to handle these issues, in this work, 1) we extend the traditional confusion sets with semantical candidates to cover different types of errors; 2) we propose a chunk-based framework to correct single-character and multi-character word errors uniformly; and 3) we adopt a global optimization strategy to enable a sentence-level correction selection. The experimental results show that the proposed approach achieves a new state-of-the-art performance on three benchmark datasets, as well as an optical character recognition dataset.</abstract>
      <url hash="d3f66397">2020.findings-emnlp.184</url>
      <doi>10.18653/v1/2020.findings-emnlp.184</doi>
    </paper>
    <paper id="185">
      <title>Multi-pretraining for Large-scale Text Classification</title>
      <author><first>Kang-Min</first><last>Kim</last></author>
      <author><first>Bumsu</first><last>Hyeon</last></author>
      <author><first>Yeachan</first><last>Kim</last></author>
      <author><first>Jun-Hyung</first><last>Park</last></author>
      <author><first>SangKeun</first><last>Lee</last></author>
      <pages>2041–2050</pages>
      <abstract>Deep neural network-based pretraining methods have achieved impressive results in many natural language processing tasks including text classification. However, their applicability to large-scale text classification with numerous categories (e.g., several thousands) is yet to be well-studied, where the training data is insufficient and skewed in terms of categories. In addition, existing pretraining methods usually involve excessive computation and memory overheads. In this paper, we develop a novel multi-pretraining framework for large-scale text classification. This multi-pretraining framework includes both a self-supervised pretraining and a weakly supervised pretraining. We newly introduce an out-of-context words detection task on the unlabeled data as the self-supervised pretraining. It captures the topic-consistency of words used in sentences, which is proven to be useful for text classification. In addition, we propose a weakly supervised pretraining, where labels for text classification are obtained automatically from an existing approach. Experimental results clearly show that both pretraining approaches are effective for large-scale text classification task. The proposed scheme exhibits significant improvements as much as 3.8% in terms of macro-averaging F1-score over strong pretraining methods, while being computationally efficient.</abstract>
      <url hash="607147f9">2020.findings-emnlp.185</url>
      <doi>10.18653/v1/2020.findings-emnlp.185</doi>
    </paper>
    <paper id="186">
      <title>End-to-End Speech Recognition and Disfluency Removal</title>
      <author><first>Paria</first><last>Jamshid Lou</last></author>
      <author><first>Mark</first><last>Johnson</last></author>
      <pages>2051–2061</pages>
      <abstract>Disfluency detection is usually an intermediate step between an automatic speech recognition (ASR) system and a downstream task. By contrast, this paper aims to investigate the task of end-to-end speech recognition and disfluency removal. We specifically explore whether it is possible to train an ASR model to directly map disfluent speech into fluent transcripts, without relying on a separate disfluency detection model. We show that end-to-end models do learn to directly generate fluent transcripts; however, their performance is slightly worse than a baseline pipeline approach consisting of an ASR system and a specialized disfluency detection model. We also propose two new metrics for evaluating integrated ASR and disfluency removal models. The findings of this paper can serve as a benchmark for further research on the task of end-to-end speech recognition and disfluency removal in the future.</abstract>
      <url hash="ab06b85d">2020.findings-emnlp.186</url>
      <doi>10.18653/v1/2020.findings-emnlp.186</doi>
    </paper>
    <paper id="187">
      <title>Characterizing the Value of Information in Medical Notes</title>
      <author><first>Chao-Chun</first><last>Hsu</last></author>
      <author><first>Shantanu</first><last>Karnwal</last></author>
      <author><first>Sendhil</first><last>Mullainathan</last></author>
      <author><first>Ziad</first><last>Obermeyer</last></author>
      <author><first>Chenhao</first><last>Tan</last></author>
      <pages>2062–2072</pages>
      <abstract>Machine learning models depend on the quality of input data. As electronic health records are widely adopted, the amount of data in health care is growing, along with complaints about the quality of medical notes. We use two prediction tasks, readmission prediction and in-hospital mortality prediction, to characterize the value of information in medical notes. We show that as a whole, medical notes only provide additional predictive power over structured information in readmission prediction. We further propose a probing framework to select parts of notes that enable more accurate predictions than using all notes, despite that the selected information leads to a distribution shift from the training data (“all notes”). Finally, we demonstrate that models trained on the selected valuable information achieve even better predictive performance, with only 6.8%of all the tokens for readmission prediction.</abstract>
      <url hash="0a75833f">2020.findings-emnlp.187</url>
      <attachment type="OptionalSupplementaryMaterial" hash="a4914f6a">2020.findings-emnlp.187.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.187</doi>
    </paper>
    <paper id="188">
      <title><fixed-case>KL</fixed-case>earn: Background Knowledge Inference from Summarization Data</title>
      <author><first>Maxime</first><last>Peyrard</last></author>
      <author><first>Robert</first><last>West</last></author>
      <pages>2073–2085</pages>
      <abstract>The goal of text summarization is to compress documents to the relevant information while excluding background information already known to the receiver. So far, summarization researchers have given considerably more attention to relevance than to background knowledge. In contrast, this work puts background knowledge in the foreground. Building on the realization that the choices made by human summarizers and annotators contain implicit information about their background knowledge, we develop and compare techniques for inferring background knowledge from summarization data. Based on this framework, we define summary scoring functions that explicitly model background knowledge, and show that these scoring functions fit human judgments significantly better than baselines. We illustrate some of the many potential applications of our framework. First, we provide insights into human information importance priors. Second, we demonstrate that averaging the background knowledge of multiple, potentially biased annotators or corpora greatly improves summaryscoring performance. Finally, we discuss potential applications of our framework beyond summarization.</abstract>
      <url hash="9127f499">2020.findings-emnlp.188</url>
      <doi>10.18653/v1/2020.findings-emnlp.188</doi>
    </paper>
    <paper id="189">
      <title>Extracting Chemical-Protein Interactions via Calibrated Deep Neural Network and Self-training</title>
      <author><first>Dongha</first><last>Choi</last></author>
      <author><first>Hyunju</first><last>Lee</last></author>
      <pages>2086–2095</pages>
      <abstract>The extraction of interactions between chemicals and proteins from several biomedical articles is important in many fields of biomedical research such as drug development and prediction of drug side effects. Several natural language processing methods, including deep neural network (DNN) models, have been applied to address this problem. However, these methods were trained with hard-labeled data, which tend to become over-confident, leading to degradation of the model reliability. To estimate the data uncertainty and improve the reliability, “calibration” techniques have been applied to deep learning models. In this study, to extract chemical–protein interactions, we propose a DNN-based approach incorporating uncertainty information and calibration techniques. Our model first encodes the input sequence using a pre-trained language-understanding model, following which it is trained using two calibration methods: mixup training and addition of a confidence penalty loss. Finally, the model is re-trained with augmented data that are extracted using the estimated uncertainties. Our approach has achieved state-of-the-art performance with regard to the Biocreative VI ChemProt task, while preserving higher calibration abilities than those of previous approaches. Furthermore, our approach also presents the possibilities of using uncertainty estimation for performance improvement.</abstract>
      <url hash="89dd250c">2020.findings-emnlp.189</url>
      <attachment type="OptionalSupplementaryMaterial" hash="5a80a0bb">2020.findings-emnlp.189.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.189</doi>
    </paper>
    <paper id="190">
      <title><fixed-case>L</fixed-case>ogic2<fixed-case>T</fixed-case>ext: High-Fidelity Natural Language Generation from Logical Forms</title>
      <author><first>Zhiyu</first><last>Chen</last></author>
      <author><first>Wenhu</first><last>Chen</last></author>
      <author><first>Hanwen</first><last>Zha</last></author>
      <author><first>Xiyou</first><last>Zhou</last></author>
      <author><first>Yunkai</first><last>Zhang</last></author>
      <author><first>Sairam</first><last>Sundaresan</last></author>
      <author><first>William Yang</first><last>Wang</last></author>
      <pages>2096–2111</pages>
      <abstract>Previous studies on Natural Language Generation (NLG) from structured data have primarily focused on surface-level descriptions of record sequences. However, for complex structured data, e.g., multi-row tables, it is often desirable for an NLG system to describe interesting facts from logical inferences across records. If only provided with the table, it is hard for existing models to produce controllable and high-fidelity logical generations. In this work, we formulate high-fidelity NLG as generation from logical forms in order to obtain controllable and faithful generations. We present a new large-scale dataset, Logic2Text, with 10,753 descriptions involving common logic types paired with the underlying logical forms. The logical forms show diversified graph structure of free schema, which pose great challenges on the model’s ability to understand the semantics. We experiment on (1) Fully-supervised training with the full datasets, and (2) Few-shot setting, provided with hundreds of paired examples; We compare several popular generation models and analyze their performances. We hope our dataset can encourage research towards building an advanced NLG system capable of natural, faithful, and human-like generation. The dataset and code is available at https://github.com/czyssrs/Logic2Text.</abstract>
      <url hash="ce73ea8e">2020.findings-emnlp.190</url>
      <doi>10.18653/v1/2020.findings-emnlp.190</doi>
    </paper>
    <paper id="191">
      <title><fixed-case>M</fixed-case>ed<fixed-case>IC</fixed-case>a<fixed-case>T</fixed-case>: A Dataset of Medical Images, Captions, and Textual References</title>
      <author><first>Sanjay</first><last>Subramanian</last></author>
      <author><first>Lucy Lu</first><last>Wang</last></author>
      <author><first>Ben</first><last>Bogin</last></author>
      <author><first>Sachin</first><last>Mehta</last></author>
      <author><first>Madeleine</first><last>van Zuylen</last></author>
      <author><first>Sravanthi</first><last>Parasa</last></author>
      <author><first>Sameer</first><last>Singh</last></author>
      <author><first>Matt</first><last>Gardner</last></author>
      <author><first>Hannaneh</first><last>Hajishirzi</last></author>
      <pages>2112–2120</pages>
      <abstract>Understanding the relationship between figures and text is key to scientific document understanding. Medical figures in particular are quite complex, often consisting of several subfigures (75% of figures in our dataset), with detailed text describing their content. Previous work studying figures in scientific papers focused on classifying figure content rather than understanding how images relate to the text. To address challenges in figure retrieval and figure-to-text alignment, we introduce MedICaT, a dataset of medical images in context. MedICaT consists of 217K images from 131K open access biomedical papers, and includes captions, inline references for 74% of figures, and manually annotated subfigures and subcaptions for a subset of figures. Using MedICaT, we introduce the task of subfigure to subcaption alignment in compound figures and demonstrate the utility of inline references in image-text matching. Our data and code can be accessed at https://github.com/allenai/medicat.</abstract>
      <url hash="01907b89">2020.findings-emnlp.191</url>
      <attachment type="OptionalSupplementaryMaterial" hash="563fa86b">2020.findings-emnlp.191.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.191</doi>
    </paper>
    <paper id="192">
      <title><fixed-case>TSDG</fixed-case>: Content-aware Neural Response Generation with Two-stage Decoding Process</title>
      <author><first>Junsheng</first><last>Kong</last></author>
      <author><first>Zhicheng</first><last>Zhong</last></author>
      <author><first>Yi</first><last>Cai</last></author>
      <author><first>Xin</first><last>Wu</last></author>
      <author><first>Da</first><last>Ren</last></author>
      <pages>2121–2126</pages>
      <abstract>Neural response generative models have achieved remarkable progress in recent years but tend to yield irrelevant and uninformative responses. One of the reasons is that encoder-decoder based models always use a single decoder to generate a complete response at a stroke. This tends to generate high-frequency function words with less semantic information rather than low-frequency content words with more semantic information. To address this issue, we propose a content-aware model with two-stage decoding process named Two-stage Dialogue Generation (TSDG). We separate the decoding process of content words and function words so that content words can be generated independently without the interference of function words. Experimental results on two datasets indicate that our model significantly outperforms several competitive generative models in terms of automatic and human evaluation.</abstract>
      <url hash="10b262ba">2020.findings-emnlp.192</url>
      <doi>10.18653/v1/2020.findings-emnlp.192</doi>
    </paper>
    <paper id="193">
      <title>Unsupervised Cross-Lingual Adaptation of Dependency Parsers Using <fixed-case>CRF</fixed-case> Autoencoders</title>
      <author><first>Zhao</first><last>Li</last></author>
      <author><first>Kewei</first><last>Tu</last></author>
      <pages>2127–2133</pages>
      <abstract>We consider the task of cross-lingual adaptation of dependency parsers without annotated target corpora and parallel corpora. Previous work either directly applies a discriminative source parser to the target language, ignoring unannotated target corpora, or employs an unsupervised generative parser that can leverage unannotated target data but has weaker representational power than discriminative parsers. In this paper, we propose to utilize unsupervised discriminative parsers based on the CRF autoencoder framework for this task. We train a source parser and use it to initialize and regularize a target parser that is trained on unannotated target data. We conduct experiments that transfer an English parser to 20 target languages. The results show that our method significantly outperforms previous methods.</abstract>
      <url hash="1a856696">2020.findings-emnlp.193</url>
      <doi>10.18653/v1/2020.findings-emnlp.193</doi>
    </paper>
    <paper id="194">
      <title>Diversify Question Generation with Continuous Content Selectors and Question Type Modeling</title>
      <author><first>Zhen</first><last>Wang</last></author>
      <author><first>Siwei</first><last>Rao</last></author>
      <author><first>Jie</first><last>Zhang</last></author>
      <author><first>Zhen</first><last>Qin</last></author>
      <author><first>Guangjian</first><last>Tian</last></author>
      <author><first>Jun</first><last>Wang</last></author>
      <pages>2134–2143</pages>
      <abstract>Generating questions based on answers and relevant contexts is a challenging task. Recent work mainly pays attention to the quality of a single generated question. However, question generation is actually a one-to-many problem, as it is possible to raise questions with different focuses on contexts and various means of expression. In this paper, we explore the diversity of question generation and come up with methods from these two aspects. Specifically, we relate contextual focuses with content selectors, which are modeled by a continuous latent variable with the technique of conditional variational auto-encoder (CVAE). In the realization of CVAE, a multimodal prior distribution is adopted to allow for more diverse content selectors. To take into account various means of expression, question types are explicitly modeled and a diversity-promoting algorithm is proposed further. Experimental results on public datasets show that our proposed method can significantly improve the diversity of generated questions, especially from the perspective of using different question types. Overall, our proposed method achieves a better trade-off between generation quality and diversity compared with existing approaches.</abstract>
      <url hash="08220343">2020.findings-emnlp.194</url>
      <doi>10.18653/v1/2020.findings-emnlp.194</doi>
    </paper>
    <paper id="195">
      <title>Participatory Research for Low-resourced Machine Translation: A Case Study in <fixed-case>A</fixed-case>frican Languages</title>
      <author><first>Wilhelmina</first><last>Nekoto</last></author>
      <author><first>Vukosi</first><last>Marivate</last></author>
      <author><first>Tshinondiwa</first><last>Matsila</last></author>
      <author><first>Timi</first><last>Fasubaa</last></author>
      <author><first>Taiwo</first><last>Fagbohungbe</last></author>
      <author><first>Solomon Oluwole</first><last>Akinola</last></author>
      <author><first>Shamsuddeen</first><last>Muhammad</last></author>
      <author><first>Salomon</first><last>Kabongo Kabenamualu</last></author>
      <author><first>Salomey</first><last>Osei</last></author>
      <author><first>Freshia</first><last>Sackey</last></author>
      <author><first>Rubungo Andre</first><last>Niyongabo</last></author>
      <author><first>Ricky</first><last>Macharm</last></author>
      <author><first>Perez</first><last>Ogayo</last></author>
      <author><first>Orevaoghene</first><last>Ahia</last></author>
      <author><first>Musie Meressa</first><last>Berhe</last></author>
      <author><first>Mofetoluwa</first><last>Adeyemi</last></author>
      <author><first>Masabata</first><last>Mokgesi-Selinga</last></author>
      <author><first>Lawrence</first><last>Okegbemi</last></author>
      <author><first>Laura</first><last>Martinus</last></author>
      <author><first>Kolawole</first><last>Tajudeen</last></author>
      <author><first>Kevin</first><last>Degila</last></author>
      <author><first>Kelechi</first><last>Ogueji</last></author>
      <author><first>Kathleen</first><last>Siminyu</last></author>
      <author><first>Julia</first><last>Kreutzer</last></author>
      <author><first>Jason</first><last>Webster</last></author>
      <author><first>Jamiil Toure</first><last>Ali</last></author>
      <author><first>Jade</first><last>Abbott</last></author>
      <author><first>Iroro</first><last>Orife</last></author>
      <author><first>Ignatius</first><last>Ezeani</last></author>
      <author><first>Idris Abdulkadir</first><last>Dangana</last></author>
      <author><first>Herman</first><last>Kamper</last></author>
      <author><first>Hady</first><last>Elsahar</last></author>
      <author><first>Goodness</first><last>Duru</last></author>
      <author><first>Ghollah</first><last>Kioko</last></author>
      <author><first>Murhabazi</first><last>Espoir</last></author>
      <author><first>Elan</first><last>van Biljon</last></author>
      <author><first>Daniel</first><last>Whitenack</last></author>
      <author><first>Christopher</first><last>Onyefuluchi</last></author>
      <author><first>Chris Chinenye</first><last>Emezue</last></author>
      <author><first>Bonaventure F. P.</first><last>Dossou</last></author>
      <author><first>Blessing</first><last>Sibanda</last></author>
      <author><first>Blessing</first><last>Bassey</last></author>
      <author><first>Ayodele</first><last>Olabiyi</last></author>
      <author><first>Arshath</first><last>Ramkilowan</last></author>
      <author><first>Alp</first><last>Öktem</last></author>
      <author><first>Adewale</first><last>Akinfaderin</last></author>
      <author><first>Abdallah</first><last>Bashir</last></author>
      <pages>2144–2160</pages>
      <abstract>Research in NLP lacks geographic diversity, and the question of how NLP can be scaled to low-resourced languages has not yet been adequately solved. ‘Low-resourced’-ness is a complex problem going beyond data availability and reflects systemic problems in society. In this paper, we focus on the task of Machine Translation (MT), that plays a crucial role for information accessibility and communication worldwide. Despite immense improvements in MT over the past decade, MT is centered around a few high-resourced languages. As MT researchers cannot solve the problem of low-resourcedness alone, we propose participatory research as a means to involve all necessary agents required in the MT development process. We demonstrate the feasibility and scalability of participatory research with a case study on MT for African languages. Its implementation leads to a collection of novel translation datasets, MT benchmarks for over 30 languages, with human evaluations for a third of them, and enables participants without formal training to make a unique scientific contribution. Benchmarks, models, data, code, and evaluation results are released at https://github.com/masakhane-io/masakhane-mt.</abstract>
      <url hash="3e1a105e">2020.findings-emnlp.195</url>
      <doi>10.18653/v1/2020.findings-emnlp.195</doi>
    </paper>
    <paper id="196">
      <title><fixed-case>C</fixed-case>onve<fixed-case>RT</fixed-case>: Efficient and Accurate Conversational Representations from Transformers</title>
      <author><first>Matthew</first><last>Henderson</last></author>
      <author><first>Iñigo</first><last>Casanueva</last></author>
      <author><first>Nikola</first><last>Mrkšić</last></author>
      <author><first>Pei-Hao</first><last>Su</last></author>
      <author><first>Tsung-Hsien</first><last>Wen</last></author>
      <author><first>Ivan</first><last>Vulić</last></author>
      <pages>2161–2174</pages>
      <abstract>General-purpose pretrained sentence encoders such as BERT are not ideal for real-world conversational AI applications; they are computationally heavy, slow, and expensive to train. We propose ConveRT (Conversational Representations from Transformers), a pretraining framework for conversational tasks satisfying all the following requirements: it is effective, affordable, and quick to train. We pretrain using a retrieval-based response selection task, effectively leveraging quantization and subword-level parameterization in the dual encoder to build a lightweight memory- and energy-efficient model. We show that ConveRT achieves state-of-the-art performance across widely established response selection tasks. We also demonstrate that the use of extended dialog history as context yields further performance gains. Finally, we show that pretrained representations from the proposed encoder can be transferred to the intent classification task, yielding strong results across three diverse data sets. ConveRT trains substantially faster than standard sentence encoders or previous state-of-the-art dual encoders. With its reduced size and superior performance, we believe this model promises wider portability and scalability for Conversational AI applications.</abstract>
      <url hash="39f0e4ce">2020.findings-emnlp.196</url>
      <doi>10.18653/v1/2020.findings-emnlp.196</doi>
    </paper>
    <paper id="197">
      <title>Computer Assisted Translation with Neural Quality Estimation and Automatic Post-Editing</title>
      <author><first>Ke</first><last>Wang</last></author>
      <author><first>Jiayi</first><last>Wang</last></author>
      <author><first>Niyu</first><last>Ge</last></author>
      <author><first>Yangbin</first><last>Shi</last></author>
      <author><first>Yu</first><last>Zhao</last></author>
      <author><first>Kai</first><last>Fan</last></author>
      <pages>2175–2186</pages>
      <abstract>With the advent of neural machine translation, there has been a marked shift towards leveraging and consuming the machine translation results. However, the gap between machine translation systems and human translators needs to be manually closed by post-editing. In this paper, we propose an end-to-end deep learning framework of the quality estimation and automatic post-editing of the machine translation output. Our goal is to provide error correction suggestions and to further relieve the burden of human translators through an interpretable model. To imitate the behavior of human translators, we design three efficient delegation modules – quality estimation, generative post-editing, and atomic operation post-editing and construct a hierarchical model based on them. We examine this approach with the English–German dataset from WMT 2017 APE shared task and our experimental results can achieve the state-of-the-art performance. We also verify that the certified translators can significantly expedite their post-editing processing with our model in human evaluation.</abstract>
      <url hash="f9f2948b">2020.findings-emnlp.197</url>
      <attachment type="OptionalSupplementaryMaterial" hash="ff2db648">2020.findings-emnlp.197.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.197</doi>
    </paper>
    <paper id="198">
      <title>Zero-Shot Rationalization by Multi-Task Transfer Learning from Question Answering</title>
      <author><first>Po-Nien</first><last>Kung</last></author>
      <author><first>Tse-Hsuan</first><last>Yang</last></author>
      <author><first>Yi-Cheng</first><last>Chen</last></author>
      <author><first>Sheng-Siang</first><last>Yin</last></author>
      <author><first>Yun-Nung</first><last>Chen</last></author>
      <pages>2187–2197</pages>
      <abstract>Extracting rationales can help human understand which information the model utilizes and how it makes the prediction towards better interpretability. However, annotating rationales requires much effort and only few datasets contain such labeled rationales, making supervised learning for rationalization difficult. In this paper, we propose a novel approach that leverages the benefits of both multi-task learning and transfer learning for generating rationales through question answering in a zero-shot fashion. For two benchmark rationalization datasets, the proposed method achieves comparable or even better performance of rationalization without any supervised signal, demonstrating the great potential of zero-shot rationalization for better interpretability.</abstract>
      <url hash="bc4caf57">2020.findings-emnlp.198</url>
      <doi>10.18653/v1/2020.findings-emnlp.198</doi>
    </paper>
    <paper id="199">
      <title>The Role of Reentrancies in <fixed-case>A</fixed-case>bstract <fixed-case>M</fixed-case>eaning <fixed-case>R</fixed-case>epresentation Parsing</title>
      <author><first>Ida</first><last>Szubert</last></author>
      <author><first>Marco</first><last>Damonte</last></author>
      <author><first>Shay B.</first><last>Cohen</last></author>
      <author><first>Mark</first><last>Steedman</last></author>
      <pages>2198–2207</pages>
      <abstract>Abstract Meaning Representation (AMR) parsing aims at converting sentences into AMR representations. These are graphs and not trees because AMR supports reentrancies (nodes with more than one parent). Following previous findings on the importance of reen- trancies for AMR, we empirically find and discuss several linguistic phenomena respon- sible for reentrancies in AMR, some of which have not received attention before. We cate- gorize the types of errors AMR parsers make with respect to reentrancies. Furthermore, we find that correcting these errors provides an in- crease of up to 5% Smatch in parsing perfor- mance and 20% in reentrancy prediction</abstract>
      <url hash="0e7f5436">2020.findings-emnlp.199</url>
      <attachment type="OptionalSupplementaryMaterial" hash="109a2ae7">2020.findings-emnlp.199.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.199</doi>
    </paper>
    <paper id="200">
      <title>Cross-Lingual Suicidal-Oriented Word Embedding toward Suicide Prevention</title>
      <author><first>Daeun</first><last>Lee</last></author>
      <author><first>Soyoung</first><last>Park</last></author>
      <author><first>Jiwon</first><last>Kang</last></author>
      <author><first>Daejin</first><last>Choi</last></author>
      <author><first>Jinyoung</first><last>Han</last></author>
      <pages>2208–2217</pages>
      <abstract>Early intervention for suicide risks with social media data has increasingly received great attention. Using a suicide dictionary created by mental health experts is one of the effective ways to detect suicidal ideation. However, little attention has been paid to validate whether and how the existing dictionaries for other languages (i.e., English and Chinese) can be used for predicting suicidal ideation for a low-resource language (i.e., Korean) where a knowledge-based suicide dictionary has not yet been developed. To this end, we propose a cross-lingual suicidal ideation detection model that can identify whether a given social media post includes suicidal ideation or not. To utilize the existing suicide dictionaries developed for other languages (i.e., English and Chinese) in word embedding, our model translates a post written in the target language (i.e., Korean) into English and Chinese, and then uses the separate suicidal-oriented word embeddings developed for English and Chinese, respectively. By applying an ensemble approach for different languages, the model achieves high accuracy, over 87%. We believe our model is useful in accessing suicidal ideation using social media data for preventing potential suicide risk in an early stage.</abstract>
      <url hash="3aca5db3">2020.findings-emnlp.200</url>
      <doi>10.18653/v1/2020.findings-emnlp.200</doi>
    </paper>
    <paper id="201">
      <title>Service-oriented Text-to-<fixed-case>SQL</fixed-case> Parsing</title>
      <author><first>Wangsu</first><last>Hu</last></author>
      <author><first>Jilei</first><last>Tian</last></author>
      <pages>2218–2222</pages>
      <abstract>The information retrieval from relational database requires professionals who has an understanding of structural query language such as SQL. TEXT2SQL models apply natural language inference to enable user interacting the database via natural language utterance. Current TEXT2SQL models normally focus on generating complex SQL query in a precise and complete fashion while certain features of real-world application in the production environment is not fully addressed. This paper is aimed to develop a service-oriented Text-to-SQL parser that translates natural language utterance to structural and executable SQL query. We introduce a algorithmic framework named Semantic-Enriched SQL generator (SE-SQL) that enables flexibly access database than rigid API in the application while keeping the performance quality for the most commonly used cases. The qualitative result shows that the proposed model achieves 88.3% execution accuracy on WikiSQL task, outperforming baseline by 13% error reduction. Moreover, the framework considers several service-oriented needs including low-complexity inference, out-of-table rejection, and text normalization.</abstract>
      <url hash="165a02b0">2020.findings-emnlp.201</url>
      <attachment type="OptionalSupplementaryMaterial" hash="72f5f330">2020.findings-emnlp.201.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.201</doi>
    </paper>
    <paper id="202">
      <title>Reinforcement Learning with Imbalanced Dataset for Data-to-Text Medical Report Generation</title>
      <author><first>Toru</first><last>Nishino</last></author>
      <author><first>Ryota</first><last>Ozaki</last></author>
      <author><first>Yohei</first><last>Momoki</last></author>
      <author><first>Tomoki</first><last>Taniguchi</last></author>
      <author><first>Ryuji</first><last>Kano</last></author>
      <author><first>Norihisa</first><last>Nakano</last></author>
      <author><first>Yuki</first><last>Tagawa</last></author>
      <author><first>Motoki</first><last>Taniguchi</last></author>
      <author><first>Tomoko</first><last>Ohkuma</last></author>
      <author><first>Keigo</first><last>Nakamura</last></author>
      <pages>2223–2236</pages>
      <abstract>Automated generation of medical reports that describe the findings in the medical images helps radiologists by alleviating their workload. Medical report generation system should generate correct and concise reports. However, data imbalance makes it difficult to train models accurately. Medical datasets are commonly imbalanced in their finding labels because incidence rates differ among diseases; moreover, the ratios of abnormalities to normalities are significantly imbalanced. We propose a novel reinforcement learning method with a reconstructor to improve the clinical correctness of generated reports to train the data-to-text module with a highly imbalanced dataset. Moreover, we introduce a novel data augmentation strategy for reinforcement learning to additionally train the model on infrequent findings. From the perspective of a practical use, we employ a Two-Stage Medical Report Generator (TS-MRGen) for controllable report generation from input images. TS-MRGen consists of two separated stages: an image diagnosis module and a data-to-text module. Radiologists can modify the image diagnosis module results to control the reports that the data-to-text module generates. We conduct an experiment with two medical datasets to assess the data-to-text module and the entire two-stage model. Results demonstrate that the reports generated by our model describe the findings in the input image more correctly.</abstract>
      <url hash="dcbce74d">2020.findings-emnlp.202</url>
      <attachment type="OptionalSupplementaryMaterial" hash="d0253092">2020.findings-emnlp.202.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.202</doi>
    </paper>
    <paper id="203">
      <title>Reducing Quantity Hallucinations in Abstractive Summarization</title>
      <author><first>Zheng</first><last>Zhao</last></author>
      <author><first>Shay B.</first><last>Cohen</last></author>
      <author><first>Bonnie</first><last>Webber</last></author>
      <pages>2237–2249</pages>
      <abstract>It is well-known that abstractive summaries are subject to hallucination—including material that is not supported by the original text. While summaries can be made hallucination-free by limiting them to general phrases, such summaries would fail to be very informative. Alternatively, one can try to avoid hallucinations by verifying that any specific entities in the summary appear in the original text in a similar context. This is the approach taken by our system, Herman. The system learns to recognize and verify quantity entities (dates, numbers, sums of money, etc.) in a beam-worth of abstractive summaries produced by state-of-the-art models, in order to up-rank those summaries whose quantity terms are supported by the original text. Experimental results demonstrate that the ROUGE scores of such up-ranked summaries have a higher Precision than summaries that have not been up-ranked, without a comparable loss in Recall, resulting in higher F1. Preliminary human evaluation of up-ranked vs. original summaries shows people’s preference for the former.</abstract>
      <url hash="e74fe8aa">2020.findings-emnlp.203</url>
      <doi>10.18653/v1/2020.findings-emnlp.203</doi>
    </paper>
    <paper id="204">
      <title>Rethinking Topic Modelling: From Document-Space to Term-Space</title>
      <author><first>Magnus</first><last>Sahlgren</last></author>
      <pages>2250–2259</pages>
      <abstract>This paper problematizes the reliance on documents as the basic notion for defining term interactions in standard topic models. As an alternative to this practice, we reformulate topic distributions as latent factors in term similarity space. We exemplify the idea using a number of standard word embeddings built with very wide context windows. The embedding spaces are transformed to sparse similarity spaces, and topics are extracted in standard fashion by factorizing to a lower-dimensional space. We use a number of different factorization techniques, and evaluate the various models using a large set of evaluation metrics, including previously published coherence measures, as well as a number of novel measures that we suggest better correspond to real-world applications of topic models. Our results clearly demonstrate that term-based models outperform standard document-based models by a large margin.</abstract>
      <url hash="ba4786a4">2020.findings-emnlp.204</url>
      <doi>10.18653/v1/2020.findings-emnlp.204</doi>
    </paper>
    <paper id="205">
      <title>Sparse and Decorrelated Representations for Stable Zero-shot <fixed-case>NMT</fixed-case></title>
      <author><first>Bokyung</first><last>Son</last></author>
      <author><first>Sungwon</first><last>Lyu</last></author>
      <pages>2260–2266</pages>
      <abstract>Using a single encoder and decoder for all directions and training with English-centric data is a popular scheme for multilingual NMT. However, zero-shot translation under this scheme is vulnerable to changes in training conditions, as the model degenerates by decoding non-English texts into English regardless of the target specifier token. We present that enforcing both sparsity and decorrelation on encoder intermediate representations with the SLNI regularizer (Aljundi et al., 2019) efficiently mitigates this problem, without performance loss in supervised directions. Notably, effects of SLNI turns out to be irrelevant to promoting language-invariance in encoder representations.</abstract>
      <url hash="640737ec">2020.findings-emnlp.205</url>
      <doi>10.18653/v1/2020.findings-emnlp.205</doi>
    </paper>
    <paper id="206">
      <title>A Semi-supervised Approach to Generate the Code-Mixed Text using Pre-trained Encoder and Transfer Learning</title>
      <author><first>Deepak</first><last>Gupta</last></author>
      <author><first>Asif</first><last>Ekbal</last></author>
      <author><first>Pushpak</first><last>Bhattacharyya</last></author>
      <pages>2267–2280</pages>
      <abstract>Code-mixing, the interleaving of two or more languages within a sentence or discourse is ubiquitous in multilingual societies. The lack of code-mixed training data is one of the major concerns for the development of end-to-end neural network-based models to be deployed for a variety of natural language processing (NLP) applications. A potential solution is to either manually create or crowd-source the code-mixed labelled data for the task at hand, but that requires much human efforts and often not feasible because of the language specific diversity in the code-mixed text. To circumvent the data scarcity issue, we propose an effective deep learning approach for automatically generating the code-mixed text from English to multiple languages without any parallel data. In order to train the neural network, we create synthetic code-mixed texts from the available parallel corpus by modelling various linguistic properties of code-mixing. Our codemixed text generator is built upon the encoder-decoder framework, where the encoder is augmented with the linguistic and task-agnostic features obtained from the transformer based language model. We also transfer the knowledge from a neural machine translation (NMT) to warm-start the training of code-mixed generator. Experimental results and in-depth analysis show the effectiveness of our proposed code-mixed text generation on eight diverse language pairs.</abstract>
      <url hash="94a42506">2020.findings-emnlp.206</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f9b14caa">2020.findings-emnlp.206.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.206</doi>
    </paper>
    <paper id="207">
      <title><fixed-case>BERT</fixed-case>-<fixed-case>MK</fixed-case>: Integrating Graph Contextualized Knowledge into Pre-trained Language Models</title>
      <author><first>Bin</first><last>He</last></author>
      <author><first>Di</first><last>Zhou</last></author>
      <author><first>Jinghui</first><last>Xiao</last></author>
      <author><first>Xin</first><last>Jiang</last></author>
      <author><first>Qun</first><last>Liu</last></author>
      <author><first>Nicholas Jing</first><last>Yuan</last></author>
      <author><first>Tong</first><last>Xu</last></author>
      <pages>2281–2290</pages>
      <abstract>Complex node interactions are common in knowledge graphs (KGs), and these interactions can be considered as contextualized knowledge exists in the topological structure of KGs. Traditional knowledge representation learning (KRL) methods usually treat a single triple as a training unit, neglecting the usage of graph contextualized knowledge. To utilize these unexploited graph-level knowledge, we propose an approach to model subgraphs in a medical KG. Then, the learned knowledge is integrated with a pre-trained language model to do the knowledge generalization. Experimental results demonstrate that our model achieves the state-of-the-art performance on several medical NLP tasks, and the improvement above MedERNIE indicates that graph contextualized knowledge is beneficial.</abstract>
      <url hash="cee8bbf8">2020.findings-emnlp.207</url>
      <doi>10.18653/v1/2020.findings-emnlp.207</doi>
    </paper>
    <paper id="208">
      <title>Recursive Top-Down Production for Sentence Generation with Latent Trees</title>
      <author><first>Shawn</first><last>Tan</last></author>
      <author><first>Yikang</first><last>Shen</last></author>
      <author><first>Alessandro</first><last>Sordoni</last></author>
      <author><first>Aaron</first><last>Courville</last></author>
      <author><first>Timothy J.</first><last>O’Donnell</last></author>
      <pages>2291–2307</pages>
      <abstract>We model the recursive production property of context-free grammars for natural and synthetic languages. To this end, we present a dynamic programming algorithm that marginalises over latent binary tree structures with N leaves, allowing us to compute the likelihood of a sequence of N tokens under a latent tree model, which we maximise to train a recursive neural function. We demonstrate performance on two synthetic tasks: SCAN, where it outperforms previous models on the LENGTH split, and English question formation, where it performs comparably to decoders with the ground-truth tree structure. We also present experimental results on German-English translation on the Multi30k dataset, and qualitatively analyse the induced tree structures our model learns for the SCAN tasks and the German-English translation task.</abstract>
      <url hash="c65d586c">2020.findings-emnlp.208</url>
      <attachment type="OptionalSupplementaryMaterial" hash="8b847342">2020.findings-emnlp.208.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.208</doi>
    </paper>
    <paper id="209">
      <title>Guided Dialogue Policy Learning without Adversarial Learning in the Loop</title>
      <author><first>Ziming</first><last>Li</last></author>
      <author><first>Sungjin</first><last>Lee</last></author>
      <author><first>Baolin</first><last>Peng</last></author>
      <author><first>Jinchao</first><last>Li</last></author>
      <author><first>Julia</first><last>Kiseleva</last></author>
      <author><first>Maarten</first><last>de Rijke</last></author>
      <author><first>Shahin</first><last>Shayandeh</last></author>
      <author><first>Jianfeng</first><last>Gao</last></author>
      <pages>2308–2317</pages>
      <abstract>Reinforcement learning methods have emerged as a popular choice for training an efficient and effective dialogue policy. However, these methods suffer from sparse and unstable reward signals returned by a user simulator only when a dialogue finishes. Besides, the reward signal is manually designed by human experts, which requires domain knowledge. Recently, a number of adversarial learning methods have been proposed to learn the reward function together with the dialogue policy. However, to alternatively update the dialogue policy and the reward model on the fly, we are limited to policy-gradient-based algorithms, such as REINFORCE and PPO. Moreover, the alternating training of a dialogue agent and the reward model can easily get stuck in local optima or result in mode collapse. To overcome the listed issues, we propose to decompose the adversarial training into two steps. First, we train the discriminator with an auxiliary dialogue generator and then incorporate a derived reward model into a common reinforcement learning method to guide the dialogue policy learning. This approach is applicable to both on-policy and off-policy reinforcement learning methods. Based on our extensive experimentation, we can conclude the proposed method: (1) achieves a remarkable task success rate using both on-policy and off-policy reinforcement learning methods; and (2) has potential to transfer knowledge from existing domains to a new domain.</abstract>
      <url hash="f230bb5a">2020.findings-emnlp.209</url>
      <doi>10.18653/v1/2020.findings-emnlp.209</doi>
    </paper>
    <paper id="210">
      <title><fixed-case>M</fixed-case>ulti<fixed-case>DM</fixed-case>-<fixed-case>GCN</fixed-case>: Aspect-guided Response Generation in Multi-domain Multi-modal Dialogue System using Graph Convolutional Network</title>
      <author><first>Mauajama</first><last>Firdaus</last></author>
      <author><first>Nidhi</first><last>Thakur</last></author>
      <author><first>Asif</first><last>Ekbal</last></author>
      <pages>2318–2328</pages>
      <abstract>In the recent past, dialogue systems have gained immense popularity and have become ubiquitous. During conversations, humans not only rely on languages but seek contextual information through visual contents as well. In every task-oriented dialogue system, the user is guided by the different aspects of a product or service that regulates the conversation towards selecting the product or service. In this work, we present a multi-modal conversational framework for a task-oriented dialogue setup that generates the responses following the different aspects of a product or service to cater to the user’s needs. We show that the responses guided by the aspect information provide more interactive and informative responses for better communication between the agent and the user. We first create a Multi-domain Multi-modal Dialogue (MDMMD) dataset having conversations involving both text and images belonging to the three different domains, such as restaurants, electronics, and furniture. We implement a Graph Convolutional Network (GCN) based framework that generates appropriate textual responses from the multi-modal inputs. The multi-modal information having both textual and image representation is fed to the decoder and the aspect information for generating aspect guided responses. Quantitative and qualitative analyses show that the proposed methodology outperforms several baselines for the proposed task of aspect-guided response generation.</abstract>
      <url hash="ca6e079f">2020.findings-emnlp.210</url>
      <doi>10.18653/v1/2020.findings-emnlp.210</doi>
    </paper>
    <paper id="211">
      <title>Edge-Enhanced Graph Convolution Networks for Event Detection with Syntactic Relation</title>
      <author><first>Shiyao</first><last>Cui</last></author>
      <author><first>Bowen</first><last>Yu</last></author>
      <author><first>Tingwen</first><last>Liu</last></author>
      <author><first>Zhenyu</first><last>Zhang</last></author>
      <author><first>Xuebin</first><last>Wang</last></author>
      <author><first>Jinqiao</first><last>Shi</last></author>
      <pages>2329–2339</pages>
      <abstract>Event detection (ED), a key subtask of information extraction, aims to recognize instances of specific event types in text. Previous studies on the task have verified the effectiveness of integrating syntactic dependency into graph convolutional networks. However, these methods usually ignore dependency label information, which conveys rich and useful linguistic knowledge for ED. In this paper, we propose a novel architecture named Edge-Enhanced Graph Convolution Networks (EE-GCN), which simultaneously exploits syntactic structure and typed dependency label information to perform ED. Specifically, an edge-aware node update module is designed to generate expressive word representations by aggregating syntactically-connected words through specific dependency types. Furthermore, to fully explore clues hidden from dependency edges, a node-aware edge update module is introduced, which refines the relation representations with contextual information.These two modules are complementary to each other and work in a mutual promotion way. We conduct experiments on the widely used ACE2005 dataset and the results show significant improvement over competitive baseline methods.</abstract>
      <url hash="2a5e0605">2020.findings-emnlp.211</url>
      <doi>10.18653/v1/2020.findings-emnlp.211</doi>
    </paper>
    <paper id="212">
      <title>Semi-supervised Formality Style Transfer using Language Model Discriminator and Mutual Information Maximization</title>
      <author><first>Kunal</first><last>Chawla</last></author>
      <author><first>Diyi</first><last>Yang</last></author>
      <pages>2340–2354</pages>
      <abstract>Formality style transfer is the task of converting informal sentences to grammatically-correct formal sentences, which can be used to improve performance of many downstream NLP tasks. In this work, we propose a semi-supervised formality style transfer model that utilizes a language model-based discriminator to maximize the likelihood of the output sentence being formal, which allows us to use maximization of token-level conditional probabilities for training. We further propose to maximize mutual information between source and target styles as our training objective instead of maximizing the regular likelihood that often leads to repetitive and trivial generated responses. Experiments showed that our model outperformed previous state-of-the-art baselines significantly in terms of both automated metrics and human judgement. We further generalized our model to unsupervised text style transfer task, and achieved significant improvements on two benchmark sentiment style transfer datasets.</abstract>
      <url hash="5c032a4c">2020.findings-emnlp.212</url>
      <attachment type="OptionalSupplementaryMaterial" hash="7d5860cf">2020.findings-emnlp.212.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.212</doi>
    </paper>
    <paper id="213">
      <title>Differentially Private Representation for <fixed-case>NLP</fixed-case>: Formal Guarantee and An Empirical Study on Privacy and Fairness</title>
      <author><first>Lingjuan</first><last>Lyu</last></author>
      <author><first>Xuanli</first><last>He</last></author>
      <author><first>Yitong</first><last>Li</last></author>
      <pages>2355–2365</pages>
      <abstract>It has been demonstrated that hidden representation learned by deep model can encode private information of the input, hence can be exploited to recover such information with reasonable accuracy. To address this issue, we propose a novel approach called Differentially Private Neural Representation (DPNR) to preserve privacy of the extracted representation from text. DPNR utilises Differential Privacy (DP) to provide formal privacy guarantee. Further, we show that masking words via dropout can further enhance privacy. To maintain utility of the learned representation, we integrate DP-noisy representation into a robust training process to derive a robust target model, which also helps for model fairness over various demographic variables. Experimental results on benchmark datasets under various parameter settings demonstrate that DPNR largely reduces privacy leakage without significantly sacrificing the main task performance.</abstract>
      <url hash="f8207cad">2020.findings-emnlp.213</url>
      <doi>10.18653/v1/2020.findings-emnlp.213</doi>
    </paper>
    <paper id="214">
      <title>Helpful or Hierarchical? Predicting the Communicative Strategies of Chat Participants, and their Impact on Success</title>
      <author><first>Farzana</first><last>Rashid</last></author>
      <author><first>Tommaso</first><last>Fornaciari</last></author>
      <author><first>Dirk</first><last>Hovy</last></author>
      <author><first>Eduardo</first><last>Blanco</last></author>
      <author><first>Fernando</first><last>Vega-Redondo</last></author>
      <pages>2366–2371</pages>
      <abstract>When interacting with each other, we motivate, advise, inform, show love or power towards our peers. However, the way we interact may also hold some indication on how successful we are, as people often try to help each other to achieve their goals. We study the chat interactions of thousands of aspiring entrepreneurs who discuss and develop business models. We manually annotate a set of about 5,500 chat interactions with four dimensions of interaction styles (motivation, cooperation, equality, advice). We find that these styles can be reliably predicted, and that the communication styles can be used to predict a number of indices of business success. Our findings indicate that successful communicators are also successful in other domains.</abstract>
      <url hash="d23a9dc0">2020.findings-emnlp.214</url>
      <doi>10.18653/v1/2020.findings-emnlp.214</doi>
    </paper>
    <paper id="215">
      <title>Learning Knowledge Bases with Parameters for Task-Oriented Dialogue Systems</title>
      <author><first>Andrea</first><last>Madotto</last></author>
      <author><first>Samuel</first><last>Cahyawijaya</last></author>
      <author><first>Genta Indra</first><last>Winata</last></author>
      <author><first>Yan</first><last>Xu</last></author>
      <author><first>Zihan</first><last>Liu</last></author>
      <author><first>Zhaojiang</first><last>Lin</last></author>
      <author><first>Pascale</first><last>Fung</last></author>
      <pages>2372–2394</pages>
      <abstract>Task-oriented dialogue systems are either modularized with separate dialogue state tracking (DST) and management steps or end-to-end trainable. In either case, the knowledge base (KB) plays an essential role in fulfilling user requests. Modularized systems rely on DST to interact with the KB, which is expensive in terms of annotation and inference time. End-to-end systems, instead, use the KB directly as input, but they cannot scale when the KB is larger than a few hundred entries. In this paper, we propose a method to embed the KB, of any size, directly into the model parameters. The resulting model does not require any DST or template responses, nor the KB as input, and it can dynamically update its KB via fine-tuning. We evaluate our solution in five task-oriented dialogue datasets with small, medium, and large KB size. Our experiments show that end-to-end models can effectively embed knowledge bases in their parameters and achieve competitive performance in all evaluated datasets.</abstract>
      <url hash="e3f276b7">2020.findings-emnlp.215</url>
      <attachment type="OptionalSupplementaryMaterial" hash="7a886fac">2020.findings-emnlp.215.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.215</doi>
    </paper>
    <paper id="216">
      <title>Adapting Open Domain Fact Extraction and Verification to <fixed-case>COVID</fixed-case>-<fixed-case>FACT</fixed-case> through In-Domain Language Modeling</title>
      <author><first>Zhenghao</first><last>Liu</last></author>
      <author><first>Chenyan</first><last>Xiong</last></author>
      <author><first>Zhuyun</first><last>Dai</last></author>
      <author><first>Si</first><last>Sun</last></author>
      <author><first>Maosong</first><last>Sun</last></author>
      <author><first>Zhiyuan</first><last>Liu</last></author>
      <pages>2395–2400</pages>
      <abstract>With the epidemic of COVID-19, verifying the scientifically false online information, such as fake news and maliciously fabricated statements, has become crucial. However, the lack of training data in the scientific domain limits the performance of fact verification models. This paper proposes an in-domain language modeling method for fact extraction and verification systems. We come up with SciKGAT to combine the advantages of open-domain literature search, state-of-the-art fact verification systems and in-domain medical knowledge through language modeling. Our experiments on SCIFACT, a dataset of expert-written scientific fact verification, show that SciKGAT achieves 30% absolute improvement on precision. Our analyses show that such improvement thrives from our in-domain language model by picking up more related evidence pieces and accurate fact verification. Our codes and data are released via Github.</abstract>
      <url hash="f059044a">2020.findings-emnlp.216</url>
      <doi>10.18653/v1/2020.findings-emnlp.216</doi>
    </paper>
    <paper id="217">
      <title><fixed-case>P</fixed-case>rophet<fixed-case>N</fixed-case>et: Predicting Future N-gram for Sequence-to-<fixed-case>S</fixed-case>equence<fixed-case>P</fixed-case>re-training</title>
      <author><first>Weizhen</first><last>Qi</last></author>
      <author><first>Yu</first><last>Yan</last></author>
      <author><first>Yeyun</first><last>Gong</last></author>
      <author><first>Dayiheng</first><last>Liu</last></author>
      <author><first>Nan</first><last>Duan</last></author>
      <author><first>Jiusheng</first><last>Chen</last></author>
      <author><first>Ruofei</first><last>Zhang</last></author>
      <author><first>Ming</first><last>Zhou</last></author>
      <pages>2401–2410</pages>
      <abstract>This paper presents a new sequence-to-sequence pre-training model called ProphetNet, which introduces a novel self-supervised objective named future n-gram prediction and the proposed n-stream self-attention mechanism. Instead of optimizing one-step-ahead prediction in the traditional sequence-to-sequence model, the ProphetNet is optimized by n-step ahead prediction that predicts the next n tokens simultaneously based on previous context tokens at each time step. The future n-gram prediction explicitly encourages the model to plan for the future tokens and prevent overfitting on strong local correlations. We pre-train ProphetNet using a base scale dataset (16GB) and a large-scale dataset (160GB), respectively. Then we conduct experiments on CNN/DailyMail, Gigaword, and SQuAD 1.1 benchmarks for abstractive summarization and question generation tasks. Experimental results show that ProphetNet achieves new state-of-the-art results on all these datasets compared to the models using the same scale pre-training corpus.</abstract>
      <url hash="bf35b73c">2020.findings-emnlp.217</url>
      <doi>10.18653/v1/2020.findings-emnlp.217</doi>
    </paper>
    <paper id="218">
      <title><fixed-case>D</fixed-case>iv<fixed-case>GAN</fixed-case>: Towards Diverse Paraphrase Generation via Diversified Generative Adversarial Network</title>
      <author><first>Yue</first><last>Cao</last></author>
      <author><first>Xiaojun</first><last>Wan</last></author>
      <pages>2411–2421</pages>
      <abstract>Paraphrases refer to texts that convey the same meaning with different expression forms. Traditional seq2seq-based models on paraphrase generation mainly focus on the fidelity while ignoring the diversity of outputs. In this paper, we propose a deep generative model to generate diverse paraphrases. We build our model based on the conditional generative adversarial network, and propose to incorporate a simple yet effective diversity loss term into the model in order to improve the diversity of outputs. The proposed diversity loss maximizes the ratio of pairwise distance between the generated texts and their corresponding latent codes, forcing the generator to focus more on the latent codes and produce diverse samples. Experimental results on benchmarks of paraphrase generation show that our proposed model can generate more diverse paraphrases compared with baselines.</abstract>
      <url hash="be6d68ee">2020.findings-emnlp.218</url>
      <doi>10.18653/v1/2020.findings-emnlp.218</doi>
    </paper>
    <paper id="219">
      <title>Plug-and-Play Conversational Models</title>
      <author><first>Andrea</first><last>Madotto</last></author>
      <author><first>Etsuko</first><last>Ishii</last></author>
      <author><first>Zhaojiang</first><last>Lin</last></author>
      <author><first>Sumanth</first><last>Dathathri</last></author>
      <author><first>Pascale</first><last>Fung</last></author>
      <pages>2422–2433</pages>
      <abstract>There has been considerable progress made towards conversational models that generate coherent and fluent responses; however, this often involves training large language models on large dialogue datasets, such as Reddit. These large conversational models provide little control over the generated responses, and this control is further limited in the absence of annotated conversational datasets for attribute specific generation that can be used for fine-tuning the model. In this paper, we first propose and evaluate plug-and-play methods for controllable response generation, which does not require dialogue specific datasets and does not rely on fine-tuning a large model. While effective, the decoding procedure induces considerable computational overhead, rendering the conversational model unsuitable for interactive usage. To overcome this, we introduce an approach that does not require further computation at decoding time, while also does not require any fine-tuning of a large language model. We demonstrate, through extensive automatic and human evaluation, a high degree of control over the generated conversational responses with regard to multiple desired attributes, while being fluent.</abstract>
      <url hash="1b32473d">2020.findings-emnlp.219</url>
      <attachment type="OptionalSupplementaryMaterial" hash="dd0c54dc">2020.findings-emnlp.219.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.219</doi>
    </paper>
    <paper id="220">
      <title>Event-Driven Learning of Systematic Behaviours in Stock Markets</title>
      <author><first>Xianchao</first><last>Wu</last></author>
      <pages>2434–2444</pages>
      <abstract>It is reported that financial news, especially financial events expressed in news, provide information to investors’ long/short decisions and influence the movements of stock markets. Motivated by this, we leverage financial event streams to train a classification neural network that detects latent event-stock linkages and stock markets’ systematic behaviours in the U.S. stock market. Our proposed pipeline includes (1) a combined event extraction method that utilizes Open Information Extraction and neural co-reference resolution, (2) a BERT/ALBERT enhanced representation of events, and (3) an extended hierarchical attention network that includes attentions on event, news and temporal levels. Our pipeline achieves significantly better accuracies and higher simulated annualized returns than state-of-the-art models when being applied to predicting Standard&amp;Poor 500, Dow Jones, Nasdaq indices and 10 individual stocks.</abstract>
      <url hash="72c0a59c">2020.findings-emnlp.220</url>
      <doi>10.18653/v1/2020.findings-emnlp.220</doi>
    </paper>
    <paper id="221">
      <title>Learning Improvised Chatbots from Adversarial Modifications of Natural Language Feedback</title>
      <author><first>Makesh Narsimhan</first><last>Sreedhar</last></author>
      <author><first>Kun</first><last>Ni</last></author>
      <author><first>Siva</first><last>Reddy</last></author>
      <pages>2445–2453</pages>
      <abstract>The ubiquitous nature of dialogue systems and their interaction with users generate an enormous amount of data. Can we improve chatbots using this data? A self-feeding chatbot improves itself by asking natural language feedback when a user is dissatisfied with its response and uses this feedback as an additional training sample. However, user feedback in most cases contains extraneous sequences hindering their usefulness as a training sample. In this work, we propose a generative adversarial model that converts noisy feedback into a plausible natural response in a conversation. The generator’s goal is to convert the feedback into a response that answers the user’s previous utterance and to fool the discriminator which distinguishes feedback from natural responses. We show that augmenting original training data with these modified feedback responses improves the original chatbot performance from 69.94%to 75.96% in ranking correct responses on the PERSONACHATdataset, a large improvement given that the original model is already trained on 131k samples.</abstract>
      <url hash="a34701f2">2020.findings-emnlp.221</url>
      <doi>10.18653/v1/2020.findings-emnlp.221</doi>
    </paper>
    <paper id="222">
      <title>Adapting Coreference Resolution to <fixed-case>T</fixed-case>witter Conversations</title>
      <author><first>Berfin</first><last>Aktaş</last></author>
      <author><first>Veronika</first><last>Solopova</last></author>
      <author><first>Annalena</first><last>Kohnert</last></author>
      <author><first>Manfred</first><last>Stede</last></author>
      <pages>2454–2460</pages>
      <abstract>The performance of standard coreference resolution is known to drop significantly on Twitter texts. We improve the performance of the (Lee et al., 2018) system, which is originally trained on OntoNotes, by retraining on manually-annotated Twitter conversation data. Further experiments by combining different portions of OntoNotes with Twitter data show that selecting text genres for the training data can beat the mere maximization of training data amount. In addition, we inspect several phenomena such as the role of deictic pronouns in conversational data, and present additional results for variant settings. Our best configuration improves the performance of the”out of the box” system by 21.6%.</abstract>
      <url hash="92395a7c">2020.findings-emnlp.222</url>
      <doi>10.18653/v1/2020.findings-emnlp.222</doi>
    </paper>
    <paper id="223">
      <title>On <fixed-case>R</fixed-case>omanization for Model Transfer Between Scripts in Neural Machine Translation</title>
      <author><first>Chantal</first><last>Amrhein</last></author>
      <author><first>Rico</first><last>Sennrich</last></author>
      <pages>2461–2469</pages>
      <abstract>Transfer learning is a popular strategy to improve the quality of low-resource machine translation. For an optimal transfer of the embedding layer, the child and parent model should share a substantial part of the vocabulary. This is not the case when transferring to languages with a different script. We explore the benefit of romanization in this scenario. Our results show that romanization entails information loss and is thus not always superior to simpler vocabulary transfer methods, but can improve the transfer between related languages with different scripts. We compare two romanization tools and find that they exhibit different degrees of information loss, which affects translation quality. Finally, we extend romanization to the target side, showing that this can be a successful strategy when coupled with a simple deromanization model.</abstract>
      <url hash="31748a4d">2020.findings-emnlp.223</url>
      <doi>10.18653/v1/2020.findings-emnlp.223</doi>
    </paper>
    <paper id="224">
      <title><fixed-case>COSMIC</fixed-case>: <fixed-case>CO</fixed-case>mmon<fixed-case>S</fixed-case>ense knowledge for e<fixed-case>M</fixed-case>otion Identification in Conversations</title>
      <author><first>Deepanway</first><last>Ghosal</last></author>
      <author><first>Navonil</first><last>Majumder</last></author>
      <author><first>Alexander</first><last>Gelbukh</last></author>
      <author><first>Rada</first><last>Mihalcea</last></author>
      <author><first>Soujanya</first><last>Poria</last></author>
      <pages>2470–2481</pages>
      <abstract>In this paper, we address the task of utterance level emotion recognition in conversations using commonsense knowledge. We propose COSMIC, a new framework that incorporates different elements of commonsense such as mental states, events, and causal relations, and build upon them to learn interactions between interlocutors participating in a conversation. Current state-of-theart methods often encounter difficulties in context propagation, emotion shift detection, and differentiating between related emotion classes. By learning distinct commonsense representations, COSMIC addresses these challenges and achieves new state-of-the-art results for emotion recognition on four different benchmark conversational datasets. Our code is available at https://github.com/declare-lab/conv-emotion.</abstract>
      <url hash="afc847c6">2020.findings-emnlp.224</url>
      <doi>10.18653/v1/2020.findings-emnlp.224</doi>
    </paper>
    <paper id="225">
      <title>Improving Compositional Generalization in Semantic Parsing</title>
      <author><first>Inbar</first><last>Oren</last></author>
      <author><first>Jonathan</first><last>Herzig</last></author>
      <author><first>Nitish</first><last>Gupta</last></author>
      <author><first>Matt</first><last>Gardner</last></author>
      <author><first>Jonathan</first><last>Berant</last></author>
      <pages>2482–2495</pages>
      <abstract>Generalization of models to out-of-distribution (OOD) data has captured tremendous attention recently. Specifically, compositional generalization, i.e., whether a model generalizes to new structures built of components observed during training, has sparked substantial interest. In this work, we investigate compositional generalization in semantic parsing, a natural test-bed for compositional generalization, as output programs are constructed from sub-components. We analyze a wide variety of models and propose multiple extensions to the attention module of the semantic parser, aiming to improve compositional generalization. We find that the following factors improve compositional generalization: (a) using contextual representations, such as ELMo and BERT, (b) informing the decoder what input tokens have previously been attended to, (c) training the decoder attention to agree with pre-computed token alignments, and (d) downsampling examples corresponding to frequent program templates. While we substantially reduce the gap between in-distribution and OOD generalization, performance on OOD compositions is still substantially lower.</abstract>
      <url hash="0348d62f">2020.findings-emnlp.225</url>
      <doi>10.18653/v1/2020.findings-emnlp.225</doi>
    </paper>
    <paper id="226">
      <title>Answer Span Correction in Machine Reading Comprehension</title>
      <author><first>Revanth</first><last>Gangi Reddy</last></author>
      <author><first>Md Arafat</first><last>Sultan</last></author>
      <author><first>Efsun</first><last>Sarioglu Kayi</last></author>
      <author><first>Rong</first><last>Zhang</last></author>
      <author><first>Vittorio</first><last>Castelli</last></author>
      <author><first>Avi</first><last>Sil</last></author>
      <pages>2496–2501</pages>
      <abstract>Answer validation in machine reading comprehension (MRC) consists of verifying an extracted answer against an input context and question pair. Previous work has looked at re-assessing the “answerability” of the question given the extracted answer. Here we address a different problem: the tendency of existing MRC systems to produce partially correct answers when presented with answerable questions. We explore the nature of such errors and propose a post-processing correction method that yields statistically significant performance improvements over state-of-the-art MRC systems in both monolingual and multilingual evaluation.</abstract>
      <url hash="6a6bc511">2020.findings-emnlp.226</url>
      <doi>10.18653/v1/2020.findings-emnlp.226</doi>
    </paper>
    <paper id="227">
      <title><fixed-case>O</fixed-case>n the <fixed-case>I</fixed-case>nterplay <fixed-case>B</fixed-case>etween <fixed-case>F</fixed-case>ine-tuning and <fixed-case>S</fixed-case>entence-level <fixed-case>P</fixed-case>robing for <fixed-case>L</fixed-case>inguistic <fixed-case>K</fixed-case>nowledge in <fixed-case>P</fixed-case>re-trained <fixed-case>T</fixed-case>ransformers</title>
      <author><first>Marius</first><last>Mosbach</last></author>
      <author><first>Anna</first><last>Khokhlova</last></author>
      <author><first>Michael A.</first><last>Hedderich</last></author>
      <author><first>Dietrich</first><last>Klakow</last></author>
      <pages>2502–2516</pages>
      <abstract>Fine-tuning pre-trained contextualized embedding models has become an integral part of the NLP pipeline. At the same time, probing has emerged as a way to investigate the linguistic knowledge captured by pre-trained models. Very little is, however, understood about how fine-tuning affects the representations of pre-trained models and thereby the linguistic knowledge they encode. This paper contributes towards closing this gap. We study three different pre-trained models: BERT, RoBERTa, and ALBERT, and investigate through sentence-level probing how fine-tuning affects their representations. We find that for some probing tasks fine-tuning leads to substantial changes in accuracy, possibly suggesting that fine-tuning introduces or even removes linguistic knowledge from a pre-trained model. These changes, however, vary greatly across different models, fine-tuning and probing tasks. Our analysis reveals that while fine-tuning indeed changes the representations of a pre-trained model and these changes are typically larger for higher layers, only in very few cases, fine-tuning has a positive effect on probing accuracy that is larger than just using the pre-trained model with a strong pooling method. Based on our findings, we argue that both positive and negative effects of fine-tuning on probing require a careful interpretation.</abstract>
      <url hash="4a3e0cc2">2020.findings-emnlp.227</url>
      <doi>10.18653/v1/2020.findings-emnlp.227</doi>
    </paper>
    <paper id="228">
      <title>Zero-shot Entity Linking with Efficient Long Range Sequence Modeling</title>
      <author><first>Zonghai</first><last>Yao</last></author>
      <author><first>Liangliang</first><last>Cao</last></author>
      <author><first>Huapu</first><last>Pan</last></author>
      <pages>2517–2522</pages>
      <abstract>This paper considers the problem of zero-shot entity linking, in which a link in the test time may not present in training. Following the prevailing BERT-based research efforts, we find a simple yet effective way is to expand the long-range sequence modeling. Unlike many previous methods, our method does not require expensive pre-training of BERT with long position embeddings. Instead, we propose an efficient position embeddings initialization method called Embedding-repeat, which initializes larger position embeddings based on BERT-Base. On the zero-shot entity linking dataset, our method improves the STOA from 76.06% to 79.08%, and for its long data, the corresponding improvement is from 74.57% to 82.14%. Our experiments suggest the effectiveness of long-range sequence modeling without retraining the BERT model.</abstract>
      <url hash="6be09f03">2020.findings-emnlp.228</url>
      <doi>10.18653/v1/2020.findings-emnlp.228</doi>
    </paper>
    <paper id="229">
      <title>How Does Context Matter? On the Robustness of Event Detection with Context-Selective Mask Generalization</title>
      <author><first>Jian</first><last>Liu</last></author>
      <author><first>Yubo</first><last>Chen</last></author>
      <author><first>Kang</first><last>Liu</last></author>
      <author><first>Yantao</first><last>Jia</last></author>
      <author><first>Zhicheng</first><last>Sheng</last></author>
      <pages>2523–2532</pages>
      <abstract>Event detection (ED) aims to identify and classify event triggers in texts, which is a crucial subtask of event extraction (EE). Despite many advances in ED, the existing studies are typically centered on improving the overall performance of an ED model, which rarely consider the robustness of an ED model. This paper aims to fill this research gap by stressing the importance of robustness modeling in ED models. We first pinpoint three stark cases demonstrating the brittleness of the existing ED models. After analyzing the underlying reason, we propose a new training mechanism, called context-selective mask generalization for ED, which can effectively mine context-specific patterns for learning and robustify an ED model. The experimental results have confirmed the effectiveness of our model regarding defending against adversarial attacks, exploring unseen predicates, and tackling ambiguity cases. Moreover, a deeper analysis suggests that our approach can learn a complementary predictive bias with most ED models that use full context for feature learning.</abstract>
      <url hash="ed0cce5e">2020.findings-emnlp.229</url>
      <doi>10.18653/v1/2020.findings-emnlp.229</doi>
    </paper>
    <paper id="230">
      <title>Adaptive Feature Selection for End-to-End Speech Translation</title>
      <author><first>Biao</first><last>Zhang</last></author>
      <author><first>Ivan</first><last>Titov</last></author>
      <author><first>Barry</first><last>Haddow</last></author>
      <author><first>Rico</first><last>Sennrich</last></author>
      <pages>2533–2544</pages>
      <abstract>Information in speech signals is not evenly distributed, making it an additional challenge for end-to-end (E2E) speech translation (ST) to learn to focus on informative features. In this paper, we propose adaptive feature selection (AFS) for encoder-decoder based E2E ST. We first pre-train an ASR encoder and apply AFS to dynamically estimate the importance of each encoded speech feature to ASR. A ST encoder, stacked on top of the ASR encoder, then receives the filtered features from the (frozen) ASR encoder. We take L0DROP (Zhang et al., 2020) as the backbone for AFS, and adapt it to sparsify speech features with respect to both temporal and feature dimensions. Results on LibriSpeech EnFr and MuST-C benchmarks show that AFS facilitates learning of ST by pruning out ~84% temporal features, yielding an average translation gain of ~1.3-1.6 BLEU and a decoding speedup of ~1.4x. In particular, AFS reduces the performance gap compared to the cascade baseline, and outperforms it on LibriSpeech En-Fr with a BLEU score of 18.56 (without data augmentation).</abstract>
      <url hash="cb9b56da">2020.findings-emnlp.230</url>
      <doi>10.18653/v1/2020.findings-emnlp.230</doi>
    </paper>
    <paper id="231">
      <title>Abstractive Multi-Document Summarization via Joint Learning with Single-Document Summarization</title>
      <author><first>Hanqi</first><last>Jin</last></author>
      <author><first>Xiaojun</first><last>Wan</last></author>
      <pages>2545–2554</pages>
      <abstract>Single-document and multi-document summarizations are very closely related in both task definition and solution method. In this work, we propose to improve neural abstractive multi-document summarization by jointly learning an abstractive single-document summarizer. We build a unified model for single-document and multi-document summarizations by fully sharing the encoder and decoder and utilizing a decoding controller to aggregate the decoder’s outputs for multiple input documents. We evaluate our model on two multi-document summarization datasets: Multi-News and DUC-04. Experimental results show the efficacy of our approach, and it can substantially outperform several strong baselines. We also verify the helpfulness of single-document summarization to abstractive multi-document summarization task.</abstract>
      <url hash="9df4d38b">2020.findings-emnlp.231</url>
      <doi>10.18653/v1/2020.findings-emnlp.231</doi>
    </paper>
    <paper id="232">
      <title>Blockwise Self-Attention for Long Document Understanding</title>
      <author><first>Jiezhong</first><last>Qiu</last></author>
      <author><first>Hao</first><last>Ma</last></author>
      <author><first>Omer</first><last>Levy</last></author>
      <author><first>Wen-tau</first><last>Yih</last></author>
      <author><first>Sinong</first><last>Wang</last></author>
      <author><first>Jie</first><last>Tang</last></author>
      <pages>2555–2565</pages>
      <abstract>We present BlockBERT, a lightweight and efficient BERT model for better modeling long-distance dependencies. Our model extends BERT by introducing sparse block structures into the attention matrix to reduce both memory consumption and training/inference time, which also enables attention heads to capture either short- or long-range contextual information. We conduct experiments on language model pre-training and several benchmark question answering datasets with various paragraph lengths. BlockBERT uses 18.7-36.1% less memory and 12.0-25.1% less time to learn the model. During testing, BlockBERT saves 27.8% inference time, while having comparable and sometimes better prediction accuracy, compared to an advanced BERT-based model, RoBERTa.</abstract>
      <url hash="fad6ed9f">2020.findings-emnlp.232</url>
      <doi>10.18653/v1/2020.findings-emnlp.232</doi>
    </paper>
    <paper id="233">
      <title>Unsupervised Few-Bits Semantic Hashing with Implicit Topics Modeling</title>
      <author><first>Fanghua</first><last>Ye</last></author>
      <author><first>Jarana</first><last>Manotumruksa</last></author>
      <author><first>Emine</first><last>Yilmaz</last></author>
      <pages>2566–2575</pages>
      <abstract>Semantic hashing is a powerful paradigm for representing texts as compact binary hash codes. The explosion of short text data has spurred the demand of few-bits hashing. However, the performance of existing semantic hashing methods cannot be guaranteed when applied to few-bits hashing because of severe information loss. In this paper, we present a simple but effective unsupervised neural generative semantic hashing method with a focus on few-bits hashing. Our model is built upon variational autoencoder and represents each hash bit as a Bernoulli variable, which allows the model to be end-to-end trainable. To address the issue of information loss, we introduce a set of auxiliary implicit topic vectors. With the aid of these topic vectors, the generated hash codes are not only low-dimensional representations of the original texts but also capture their implicit topics. We conduct comprehensive experiments on four datasets. The results demonstrate that our approach achieves significant improvements over state-of-the-art semantic hashing methods in few-bits hashing.</abstract>
      <url hash="05c4fc42">2020.findings-emnlp.233</url>
      <doi>10.18653/v1/2020.findings-emnlp.233</doi>
    </paper>
    <paper id="234">
      <title>Grid Tagging Scheme for Aspect-oriented Fine-grained Opinion Extraction</title>
      <author><first>Zhen</first><last>Wu</last></author>
      <author><first>Chengcan</first><last>Ying</last></author>
      <author><first>Fei</first><last>Zhao</last></author>
      <author><first>Zhifang</first><last>Fan</last></author>
      <author><first>Xinyu</first><last>Dai</last></author>
      <author><first>Rui</first><last>Xia</last></author>
      <pages>2576–2585</pages>
      <abstract>Aspect-oriented Fine-grained Opinion Extraction (AFOE) aims at extracting aspect terms and opinion terms from review in the form of opinion pairs or additionally extracting sentiment polarity of aspect term to form opinion triplet. Because of containing several opinion factors, the complete AFOE task is usually divided into multiple subtasks and achieved in the pipeline. However, pipeline approaches easily suffer from error propagation and inconvenience in real-world scenarios. To this end, we propose a novel tagging scheme, Grid Tagging Scheme (GTS), to address the AFOE task in an end-to-end fashion only with one unified grid tagging task. Additionally, we design an effective inference strategy on GTS to exploit mutual indication between different opinion factors for more accurate extractions. To validate the feasibility and compatibility of GTS, we implement three different GTS models respectively based on CNN, BiLSTM, and BERT, and conduct experiments on the aspect-oriented opinion pair extraction and opinion triplet extraction datasets. Extensive experimental results indicate that GTS models outperform strong baselines significantly and achieve state-of-the-art performance.</abstract>
      <url hash="608cf928">2020.findings-emnlp.234</url>
      <doi>10.18653/v1/2020.findings-emnlp.234</doi>
    </paper>
    <paper id="235">
      <title>Learning Numeral Embedding</title>
      <author><first>Chengyue</first><last>Jiang</last></author>
      <author><first>Zhonglin</first><last>Nian</last></author>
      <author><first>Kaihao</first><last>Guo</last></author>
      <author><first>Shanbo</first><last>Chu</last></author>
      <author><first>Yinggong</first><last>Zhao</last></author>
      <author><first>Libin</first><last>Shen</last></author>
      <author><first>Kewei</first><last>Tu</last></author>
      <pages>2586–2599</pages>
      <abstract>Word embedding is an essential building block for deep learning methods for natural language processing. Although word embedding has been extensively studied over the years, the problem of how to effectively embed numerals, a special subset of words, is still underexplored. Existing word embedding methods do not learn numeral embeddings well because there are an infinite number of numerals and their individual appearances in training corpora are highly scarce. In this paper, we propose two novel numeral embedding methods that can handle the out-of-vocabulary (OOV) problem for numerals. We first induce a finite set of prototype numerals using either a self-organizing map or a Gaussian mixture model. We then represent the embedding of a numeral as a weighted average of the prototype number embeddings. Numeral embeddings represented in this manner can be plugged into existing word embedding learning approaches such as skip-gram for training. We evaluated our methods and showed its effectiveness on four intrinsic and extrinsic tasks: word similarity, embedding numeracy, numeral prediction, and sequence labeling.</abstract>
      <url hash="ce57e6a4">2020.findings-emnlp.235</url>
      <doi>10.18653/v1/2020.findings-emnlp.235</doi>
    </paper>
    <paper id="236">
      <title>An Investigation of Potential Function Designs for Neural <fixed-case>CRF</fixed-case></title>
      <author><first>Zechuan</first><last>Hu</last></author>
      <author><first>Yong</first><last>Jiang</last></author>
      <author><first>Nguyen</first><last>Bach</last></author>
      <author><first>Tao</first><last>Wang</last></author>
      <author><first>Zhongqiang</first><last>Huang</last></author>
      <author><first>Fei</first><last>Huang</last></author>
      <author><first>Kewei</first><last>Tu</last></author>
      <pages>2600–2609</pages>
      <abstract>The neural linear-chain CRF model is one of the most widely-used approach to sequence labeling. In this paper, we investigate a series of increasingly expressive potential functions for neural CRF models, which not only integrate the emission and transition functions, but also explicitly take the representations of the contextual words as input. Our extensive experiments show that the decomposed quadrilinear potential function based on the vector representations of two neighboring labels and two neighboring words consistently achieves the best performance.</abstract>
      <url hash="4e832a54">2020.findings-emnlp.236</url>
      <doi>10.18653/v1/2020.findings-emnlp.236</doi>
    </paper>
    <paper id="237">
      <title>Fast End-to-end Coreference Resolution for <fixed-case>K</fixed-case>orean</title>
      <author><first>Cheoneum</first><last>Park</last></author>
      <author><first>Jamin</first><last>Shin</last></author>
      <author><first>Sungjoon</first><last>Park</last></author>
      <author><first>Joonho</first><last>Lim</last></author>
      <author><first>Changki</first><last>Lee</last></author>
      <pages>2610–2624</pages>
      <abstract>Recently, end-to-end neural network-based approaches have shown significant improvements over traditional pipeline-based models in English coreference resolution. However, such advancements came at a cost of computational complexity and recent works have not focused on tackling this problem. Hence, in this paper, to cope with this issue, we propose BERT-SRU-based Pointer Networks that leverages the linguistic property of head-final languages. Applying this model to the Korean coreference resolution, we significantly reduce the coreference linking search space. Combining this with Ensemble Knowledge Distillation, we maintain state-of-the-art performance 66.9% of CoNLL F1 on ETRI test set while achieving 2x speedup (30 doc/sec) in document processing time.</abstract>
      <url hash="937a0a49">2020.findings-emnlp.237</url>
      <attachment type="OptionalSupplementaryMaterial" hash="8017b0d0">2020.findings-emnlp.237.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.237</doi>
    </paper>
    <paper id="238">
      <title>Toward Stance-based Personas for Opinionated Dialogues</title>
      <author><first>Thomas</first><last>Scialom</last></author>
      <author><first>Serra Sinem</first><last>Tekiroğlu</last></author>
      <author><first>Jacopo</first><last>Staiano</last></author>
      <author><first>Marco</first><last>Guerini</last></author>
      <pages>2625–2635</pages>
      <abstract>In the context of chit-chat dialogues it has been shown that endowing systems with a persona profile is important to produce more coherent and meaningful conversations. Still, the representation of such personas has thus far been limited to a fact-based representation (e.g. “I have two cats.”). We argue that these representations remain superficial w.r.t. the complexity of human personality. In this work, we propose to make a step forward and investigate stance-based persona, trying to grasp more profound characteristics, such as opinions, values, and beliefs to drive language generation. To this end, we introduce a novel dataset allowing to explore different stance-based persona representations and their impact on claim generation, showing that they are able to grasp abstract and profound aspects of the author persona.</abstract>
      <url hash="2cf0cdb8">2020.findings-emnlp.238</url>
      <doi>10.18653/v1/2020.findings-emnlp.238</doi>
    </paper>
    <paper id="239">
      <title>Hierarchical Pre-training for Sequence Labelling in Spoken Dialog</title>
      <author><first>Emile</first><last>Chapuis</last></author>
      <author><first>Pierre</first><last>Colombo</last></author>
      <author><first>Matteo</first><last>Manica</last></author>
      <author><first>Matthieu</first><last>Labeau</last></author>
      <author><first>Chloé</first><last>Clavel</last></author>
      <pages>2636–2648</pages>
      <abstract>Sequence labelling tasks like Dialog Act and Emotion/Sentiment identification are a key component of spoken dialog systems. In this work, we propose a new approach to learn generic representations adapted to spoken dialog, which we evaluate on a new benchmark we call Sequence labellIng evaLuatIon benChmark fOr spoken laNguagE benchmark (SILICONE). SILICONE is model-agnostic and contains 10 different datasets of various sizes. We obtain our representations with a hierarchical encoder based on transformer architectures, for which we extend two well-known pre-training objectives. Pre-training is performed on OpenSubtitles: a large corpus of spoken dialog containing over 2.3 billion of tokens. We demonstrate how hierarchical encoders achieve competitive results with consistently fewer parameters compared to state-of-the-art models and we show their importance for both pre-training and fine-tuning.</abstract>
      <url hash="c9e5b242">2020.findings-emnlp.239</url>
      <attachment type="OptionalSupplementaryMaterial" hash="783f9968">2020.findings-emnlp.239.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.239</doi>
    </paper>
    <paper id="240">
      <title>Extending Multilingual <fixed-case>BERT</fixed-case> to Low-Resource Languages</title>
      <author><first>Zihan</first><last>Wang</last></author>
      <author><first>Karthikeyan</first><last>K</last></author>
      <author><first>Stephen</first><last>Mayhew</last></author>
      <author><first>Dan</first><last>Roth</last></author>
      <pages>2649–2656</pages>
      <abstract>Multilingual BERT (M-BERT) has been a huge success in both supervised and zero-shot cross-lingual transfer learning. However, this success is focused only on the top 104 languages in Wikipedia it was trained on. In this paper, we propose a simple but effective approach to extend M-BERT E-MBERT so it can benefit any new language, and show that our approach aids languages that are already in M-BERT as well. We perform an extensive set of experiments with Named Entity Recognition (NER) on 27 languages, only 16 of which are in M-BERT, and show an average increase of about 6% F1 on M-BERT languages and 23% F1 increase on new languages. We release models and code at http://cogcomp.org/page/publication_view/912.</abstract>
      <url hash="86cf23f4">2020.findings-emnlp.240</url>
      <doi>10.18653/v1/2020.findings-emnlp.240</doi>
    </paper>
    <paper id="241">
      <title>Out-of-Sample Representation Learning for Knowledge Graphs</title>
      <author><first>Marjan</first><last>Albooyeh</last></author>
      <author><first>Rishab</first><last>Goel</last></author>
      <author><first>Seyed Mehran</first><last>Kazemi</last></author>
      <pages>2657–2666</pages>
      <abstract>Many important problems can be formulated as reasoning in knowledge graphs. Representation learning has proved extremely effective for transductive reasoning, in which one needs to make new predictions for already observed entities. This is true for both attributed graphs(where each entity has an initial feature vector) and non-attributed graphs (where the only initial information derives from known relations with other entities). For out-of-sample reasoning, where one needs to make predictions for entities that were unseen at training time, much prior work considers attributed graph. However, this problem is surprisingly under-explored for non-attributed graphs. In this paper, we study the out-of-sample representation learning problem for non-attributed knowledge graphs, create benchmark datasets for this task, develop several models and baselines, and provide empirical analyses and comparisons of the proposed models and baselines.</abstract>
      <url hash="e48fdee2">2020.findings-emnlp.241</url>
      <doi>10.18653/v1/2020.findings-emnlp.241</doi>
    </paper>
    <paper id="242">
      <title>Fine-Grained Grounding for Multimodal Speech Recognition</title>
      <author><first>Tejas</first><last>Srinivasan</last></author>
      <author><first>Ramon</first><last>Sanabria</last></author>
      <author><first>Florian</first><last>Metze</last></author>
      <author><first>Desmond</first><last>Elliott</last></author>
      <pages>2667–2677</pages>
      <abstract>Multimodal automatic speech recognition systems integrate information from images to improve speech recognition quality, by grounding the speech in the visual context. While visual signals have been shown to be useful for recovering entities that have been masked in the audio, these models should be capable of recovering a broader range of word types. Existing systems rely on global visual features that represent the entire image, but localizing the relevant regions of the image will make it possible to recover a larger set of words, such as adjectives and verbs. In this paper, we propose a model that uses finer-grained visual information from different parts of the image, using automatic object proposals. In experiments on the Flickr8K Audio Captions Corpus, we find that our model improves over approaches that use global visual features, that the proposals enable the model to recover entities and other related words, such as adjectives, and that improvements are due to the model’s ability to localize the correct proposals.</abstract>
      <url hash="364ebacb">2020.findings-emnlp.242</url>
      <doi>10.18653/v1/2020.findings-emnlp.242</doi>
    </paper>
    <paper id="243">
      <title>Unsupervised Expressive Rules Provide Explainability and Assist Human Experts Grasping New Domains</title>
      <author><first>Eyal</first><last>Shnarch</last></author>
      <author><first>Leshem</first><last>Choshen</last></author>
      <author><first>Guy</first><last>Moshkowich</last></author>
      <author><first>Ranit</first><last>Aharonov</last></author>
      <author><first>Noam</first><last>Slonim</last></author>
      <pages>2678–2697</pages>
      <abstract>Approaching new data can be quite deterrent; you do not know how your categories of interest are realized in it, commonly, there is no labeled data at hand, and the performance of domain adaptation methods is unsatisfactory. Aiming to assist domain experts in their first steps into a new task over a new corpus, we present an unsupervised approach to reveal complex rules which cluster the unexplored corpus by its prominent categories (or facets). These rules are human-readable, thus providing an important ingredient which has become in short supply lately - explainability. Each rule provides an explanation for the commonality of all the texts it clusters together. The experts can then identify which rules best capture texts of their categories of interest, and utilize them to deepen their understanding of these categories. These rules can also bootstrap the process of data labeling by pointing at a subset of the corpus which is enriched with texts demonstrating the target categories. We present an extensive evaluation of the usefulness of these rules in identifying target categories, as well as a user study which assesses their interpretability.</abstract>
      <url hash="014b5e3d">2020.findings-emnlp.243</url>
      <attachment type="OptionalSupplementaryMaterial" hash="5889c2fb">2020.findings-emnlp.243.OptionalSupplementaryMaterial.txt</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.243</doi>
    </paper>
    <paper id="244">
      <title>Textual <fixed-case>S</fixed-case>upervision for <fixed-case>V</fixed-case>isually <fixed-case>G</fixed-case>rounded <fixed-case>S</fixed-case>poken <fixed-case>L</fixed-case>anguage <fixed-case>U</fixed-case>nderstanding</title>
      <author><first>Bertrand</first><last>Higy</last></author>
      <author><first>Desmond</first><last>Elliott</last></author>
      <author><first>Grzegorz</first><last>Chrupała</last></author>
      <pages>2698–2709</pages>
      <abstract>Visually-grounded models of spoken language understanding extract semantic information directly from speech, without relying on transcriptions. This is useful for low-resource languages, where transcriptions can be expensive or impossible to obtain. Recent work showed that these models can be improved if transcriptions are available at training time. However, it is not clear how an end-to-end approach compares to a traditional pipeline-based approach when one has access to transcriptions. Comparing different strategies, we find that the pipeline approach works better when enough text is available. With low-resource languages in mind, we also show that translations can be effectively used in place of transcriptions but more data is needed to obtain similar results.</abstract>
      <url hash="62d3c038">2020.findings-emnlp.244</url>
      <doi>10.18653/v1/2020.findings-emnlp.244</doi>
    </paper>
    <paper id="245">
      <title><fixed-case>U</fixed-case>niversal <fixed-case>D</fixed-case>ependencies <fixed-case>A</fixed-case>ccording to <fixed-case>BERT</fixed-case>: <fixed-case>B</fixed-case>oth <fixed-case>M</fixed-case>ore <fixed-case>S</fixed-case>pecific and <fixed-case>M</fixed-case>ore <fixed-case>G</fixed-case>eneral</title>
      <author><first>Tomasz</first><last>Limisiewicz</last></author>
      <author><first>David</first><last>Mareček</last></author>
      <author><first>Rudolf</first><last>Rosa</last></author>
      <pages>2710–2722</pages>
      <abstract>This work focuses on analyzing the form and extent of syntactic abstraction captured by BERT by extracting labeled dependency trees from self-attentions. Previous work showed that individual BERT heads tend to encode particular dependency relation types. We extend these findings by explicitly comparing BERT relations to Universal Dependencies (UD) annotations, showing that they often do not match one-to-one. We suggest a method for relation identification and syntactic tree construction. Our approach produces significantly more consistent dependency trees than previous work, showing that it better explains the syntactic abstractions in BERT. At the same time, it can be successfully applied with only a minimal amount of supervision and generalizes well across languages.</abstract>
      <url hash="04546947">2020.findings-emnlp.245</url>
      <doi>10.18653/v1/2020.findings-emnlp.245</doi>
    </paper>
    <paper id="246">
      <title>Visual Objects As Context: Exploiting Visual Objects for Lexical Entailment</title>
      <author><first>Masayasu</first><last>Muraoka</last></author>
      <author><first>Tetsuya</first><last>Nasukawa</last></author>
      <author><first>Bishwaranjan</first><last>Bhattacharjee</last></author>
      <pages>2723–2735</pages>
      <abstract>We propose a new word representation method derived from visual objects in associated images to tackle the lexical entailment task. Although it has been shown that the <i>Distributional Informativeness Hypothesis</i> (DIH) holds on text, in which the DIH assumes that a context surrounding a hyponym is more informative than that of a hypernym, it has never been tested on visual objects. Since our perception is tightly associated with language, it is meaningful to explore whether the DIH holds on visual objects. To this end, we consider visual objects as the context of a word and represent a word as a bag of visual objects found in images associated with the word. This allows us to test the feasibility of the visual DIH. To better distinguish word pairs in a hypernym relation from other relations such as co-hypernyms, we also propose a new measurable function that takes into account both the difference in the generality of meaning and similarity of meaning between words. Our experimental results show that the DIH holds on visual objects and that the proposed method combined with the proposed function outperforms existing unsupervised representation methods.</abstract>
      <url hash="e3456af1">2020.findings-emnlp.246</url>
      <doi>10.18653/v1/2020.findings-emnlp.246</doi>
    </paper>
    <paper id="247">
      <title>Learning to Plan and Realize Separately for Open-Ended Dialogue Systems</title>
      <author><first>Sashank</first><last>Santhanam</last></author>
      <author><first>Zhuo</first><last>Cheng</last></author>
      <author><first>Brodie</first><last>Mather</last></author>
      <author><first>Bonnie</first><last>Dorr</last></author>
      <author><first>Archna</first><last>Bhatia</last></author>
      <author><first>Bryanna</first><last>Hebenstreit</last></author>
      <author><first>Alan</first><last>Zemel</last></author>
      <author><first>Adam</first><last>Dalton</last></author>
      <author><first>Tomek</first><last>Strzalkowski</last></author>
      <author><first>Samira</first><last>Shaikh</last></author>
      <pages>2736–2750</pages>
      <abstract>Achieving true human-like ability to conduct a conversation remains an elusive goal for open-ended dialogue systems. We posit this is because extant approaches towards natural language generation (NLG) are typically construed as end-to-end architectures that do not adequately model human generation processes. To investigate, we decouple generation into two separate phases: planning and realization. In the planning phase, we train two planners to generate plans for response utterances. The realization phase uses response plans to produce an appropriate response. Through rigorous evaluations, both automated and human, we demonstrate that decoupling the process into planning and realization performs better than an end-to-end approach.</abstract>
      <url hash="56a3b486">2020.findings-emnlp.247</url>
      <attachment type="OptionalSupplementaryMaterial" hash="21127d1c">2020.findings-emnlp.247.OptionalSupplementaryMaterial.txt</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.247</doi>
    </paper>
    <paper id="248">
      <title><fixed-case>B</fixed-case>e <fixed-case>D</fixed-case>ifferent to <fixed-case>B</fixed-case>e <fixed-case>B</fixed-case>etter! <fixed-case>A</fixed-case> <fixed-case>B</fixed-case>enchmark to <fixed-case>L</fixed-case>everage the <fixed-case>C</fixed-case>omplementarity of <fixed-case>L</fixed-case>anguage and <fixed-case>V</fixed-case>ision</title>
      <author><first>Sandro</first><last>Pezzelle</last></author>
      <author><first>Claudio</first><last>Greco</last></author>
      <author><first>Greta</first><last>Gandolfi</last></author>
      <author><first>Eleonora</first><last>Gualdoni</last></author>
      <author><first>Raffaella</first><last>Bernardi</last></author>
      <pages>2751–2767</pages>
      <abstract>This paper introduces BD2BB, a novel language and vision benchmark that requires multimodal models combine complementary information from the two modalities. Recently, impressive progress has been made to develop universal multimodal encoders suitable for virtually any language and vision tasks. However, current approaches often require them to combine redundant information provided by language and vision. Inspired by real-life communicative contexts, we propose a novel task where either modality is necessary but not sufficient to make a correct prediction. To do so, we first build a dataset of images and corresponding sentences provided by human participants. Second, we evaluate state-of-the-art models and compare their performance against human speakers. We show that, while the task is relatively easy for humans, best-performing models struggle to achieve similar results.</abstract>
      <url hash="afc31625">2020.findings-emnlp.248</url>
      <doi>10.18653/v1/2020.findings-emnlp.248</doi>
    </paper>
    <paper id="249">
      <title>Cross-Lingual Training of Neural Models for Document Ranking</title>
      <author><first>Peng</first><last>Shi</last></author>
      <author><first>He</first><last>Bai</last></author>
      <author><first>Jimmy</first><last>Lin</last></author>
      <pages>2768–2773</pages>
      <abstract>We tackle the challenge of cross-lingual training of neural document ranking models for mono-lingual retrieval, specifically leveraging relevance judgments in English to improve search in non-English languages. Our work successfully applies multi-lingual BERT (mBERT) to document ranking and additionally compares against a number of alternatives: translating the training data, translating documents, multi-stage hybrids, and ensembles. Experiments on test collections in six different languages from diverse language families reveal many interesting findings: model-based relevance transfer using mBERT can significantly improve search quality in (non-English) mono-lingual retrieval, but other “low resource” approaches are competitive as well.</abstract>
      <url hash="d5db2ca9">2020.findings-emnlp.249</url>
      <doi>10.18653/v1/2020.findings-emnlp.249</doi>
    </paper>
    <paper id="250">
      <title><fixed-case>I</fixed-case>mproving <fixed-case>W</fixed-case>ord <fixed-case>E</fixed-case>mbedding <fixed-case>F</fixed-case>actorization for <fixed-case>C</fixed-case>ompression <fixed-case>U</fixed-case>sing <fixed-case>D</fixed-case>istilled <fixed-case>N</fixed-case>onlinear <fixed-case>N</fixed-case>eural <fixed-case>D</fixed-case>ecomposition</title>
      <author><first>Vasileios</first><last>Lioutas</last></author>
      <author><first>Ahmad</first><last>Rashid</last></author>
      <author><first>Krtin</first><last>Kumar</last></author>
      <author><first>Md. Akmal</first><last>Haidar</last></author>
      <author><first>Mehdi</first><last>Rezagholizadeh</last></author>
      <pages>2774–2784</pages>
      <abstract>Word-embeddings are vital components of Natural Language Processing (NLP) models and have been extensively explored. However, they consume a lot of memory which poses a challenge for edge deployment. Embedding matrices, typically, contain most of the parameters for language models and about a third for machine translation systems. In this paper, we propose Distilled Embedding, an (input/output) embedding compression method based on low-rank matrix decomposition and knowledge distillation. First, we initialize the weights of our decomposed matrices by learning to reconstruct the full pre-trained word-embedding and then fine-tune end-to-end, employing knowledge distillation on the factorized embedding. We conduct extensive experiments with various compression rates on machine translation and language modeling, using different data-sets with a shared word-embedding matrix for both embedding and vocabulary projection matrices. We show that the proposed technique is simple to replicate, with one fixed parameter controlling compression size, has higher BLEU score on translation and lower perplexity on language modeling compared to complex, difficult to tune state-of-the-art methods.</abstract>
      <url hash="c2bdd05d">2020.findings-emnlp.250</url>
      <attachment type="OptionalSupplementaryMaterial" hash="614486ab">2020.findings-emnlp.250.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.250</doi>
    </paper>
    <paper id="251">
      <title><fixed-case>P</fixed-case>harm<fixed-case>MT</fixed-case>: A Neural Machine Translation Approach to Simplify Prescription Directions</title>
      <author><first>Jiazhao</first><last>Li</last></author>
      <author><first>Corey</first><last>Lester</last></author>
      <author><first>Xinyan</first><last>Zhao</last></author>
      <author><first>Yuting</first><last>Ding</last></author>
      <author><first>Yun</first><last>Jiang</last></author>
      <author><first>V.G.Vinod</first><last>Vydiswaran</last></author>
      <pages>2785–2796</pages>
      <abstract>The language used by physicians and health professionals in prescription directions includes medical jargon and implicit directives and causes much confusion among patients. Human intervention to simplify the language at the pharmacies may introduce additional errors that can lead to potentially severe health outcomes. We propose a novel machine translation-based approach, PharmMT, to automatically and reliably simplify prescription directions into patient-friendly language, thereby significantly reducing pharmacist workload. We evaluate the proposed approach over a dataset consisting of over 530K prescriptions obtained from a large mail-order pharmacy. The end-to-end system achieves a BLEU score of 60.27 against the reference directions generated by pharmacists, a 39.6% relative improvement over the rule-based normalization. Pharmacists judged 94.3% of the simplified directions as usable as-is or with minimal changes. This work demonstrates the feasibility of a machine translation-based tool for simplifying prescription directions in real-life.</abstract>
      <url hash="7586b9f9">2020.findings-emnlp.251</url>
      <doi>10.18653/v1/2020.findings-emnlp.251</doi>
    </paper>
    <paper id="252">
      <title><fixed-case>LSTM</fixed-case>s Compose—and <fixed-case>L</fixed-case>earn—<fixed-case>B</fixed-case>ottom-Up</title>
      <author><first>Naomi</first><last>Saphra</last></author>
      <author><first>Adam</first><last>Lopez</last></author>
      <pages>2797–2809</pages>
      <abstract>Recent work in NLP shows that LSTM language models capture compositional structure in language data. In contrast to existing work, we consider the <i>learning</i> process that leads to compositional behavior. For a closer look at how an LSTM’s sequential representations are composed hierarchically, we present a related measure of Decompositional Interdependence (DI) between word meanings in an LSTM, based on their gate interactions. We support this measure with experiments on English language data, where DI is higher on pairs of words with lower syntactic distance. To explore the inductive biases that cause these compositional representations to arise during training, we conduct simple experiments on synthetic data. These synthetic experiments support a specific hypothesis about how hierarchical structures are discovered over the course of training: that LSTM constituent representations are learned bottom-up, relying on effective representations of their shorter children, rather than on learning the longer-range relations independently.</abstract>
      <url hash="a8642b31">2020.findings-emnlp.252</url>
      <doi>10.18653/v1/2020.findings-emnlp.252</doi>
    </paper>
    <paper id="253">
      <title>Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs</title>
      <author><first>Ana</first><last>Marasović</last></author>
      <author><first>Chandra</first><last>Bhagavatula</last></author>
      <author><first>Jae sung</first><last>Park</last></author>
      <author><first>Ronan</first><last>Le Bras</last></author>
      <author><first>Noah A.</first><last>Smith</last></author>
      <author><first>Yejin</first><last>Choi</last></author>
      <pages>2810–2829</pages>
      <abstract>Natural language rationales could provide intuitive, higher-level explanations that are easily understandable by humans, complementing the more broadly studied lower-level explanations based on gradients or attention weights. We present the first study focused on generating natural language rationales across several complex visual reasoning tasks: visual commonsense reasoning, visual-textual entailment, and visual question answering. The key challenge of accurate rationalization is comprehensive image understanding at all levels: not just their explicit content at the pixel level, but their contextual contents at the semantic and pragmatic levels. We present RationaleˆVT Transformer, an integrated model that learns to generate free-text rationales by combining pretrained language models with object recognition, grounded visual semantic frames, and visual commonsense graphs. Our experiments show that free-text rationalization is a promising research direction to complement model interpretability for complex visual-textual reasoning tasks. In addition, we find that integration of richer semantic and pragmatic visual features improves visual fidelity of rationales.</abstract>
      <url hash="2fe375a4">2020.findings-emnlp.253</url>
      <attachment type="OptionalSupplementaryMaterial" hash="83556619">2020.findings-emnlp.253.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.253</doi>
    </paper>
    <paper id="254">
      <title>Corpora Evaluation and System Bias Detection in Multi-document Summarization</title>
      <author><first>Alvin</first><last>Dey</last></author>
      <author><first>Tanya</first><last>Chowdhury</last></author>
      <author><first>Yash</first><last>Kumar</last></author>
      <author><first>Tanmoy</first><last>Chakraborty</last></author>
      <pages>2830–2840</pages>
      <abstract>Multi-document summarization (MDS) is the task of reflecting key points from any set of documents into a concise text paragraph. In the past, it has been used to aggregate news, tweets, product reviews, etc. from various sources. Owing to no standard definition of the task, we encounter a plethora of datasets with varying levels of overlap and conflict between participating documents. There is also no standard regarding what constitutes summary information in MDS. Adding to the challenge is the fact that new systems report results on a set of chosen datasets, which might not correlate with their performance on the other datasets. In this paper, we study this heterogeneous task with the help of a few widely used MDS corpora and a suite of state-of-theart models. We make an attempt to quantify the quality of summarization corpus and prescribe a list of points to consider while proposing a new MDS corpus. Next, we analyze the reason behind the absence of an MDS system which achieves superior performance across all corpora. We then observe the extent to which system metrics are influenced, and bias is propagated due to corpus properties. The scripts to reproduce the experiments in this work are available at https://github.com/LCS2-IIITD/summarization_bias.git</abstract>
      <url hash="120d3baf">2020.findings-emnlp.254</url>
      <attachment type="OptionalSupplementaryMaterial" hash="00000000">2020.findings-emnlp.254.OptionalSupplementaryMaterial.txt</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.254</doi>
    </paper>
    <paper id="255">
      <title>Graph-to-Tree Neural Networks for Learning Structured Input-Output Translation with Applications to Semantic Parsing and Math Word Problem</title>
      <author><first>Shucheng</first><last>Li</last></author>
      <author><first>Lingfei</first><last>Wu</last></author>
      <author><first>Shiwei</first><last>Feng</last></author>
      <author><first>Fangli</first><last>Xu</last></author>
      <author><first>Fengyuan</first><last>Xu</last></author>
      <author><first>Sheng</first><last>Zhong</last></author>
      <pages>2841–2852</pages>
      <abstract>The celebrated Seq2Seq technique and its numerous variants achieve excellent performance on many tasks such as neural machine translation, semantic parsing, and math word problem solving. However, these models either only consider input objects as sequences while ignoring the important structural information for encoding, or they simply treat output objects as sequence outputs instead of structural objects for decoding. In this paper, we present a novel Graph-to-Tree Neural Networks, namely Graph2Tree consisting of a graph encoder and a hierarchical tree decoder, that encodes an augmented graph-structured input and decodes a tree-structured output. In particular, we investigated our model for solving two problems, neural semantic parsing and math word problem. Our extensive experiments demonstrate that our Graph2Tree model outperforms or matches the performance of other state-of-the-art models on these tasks.</abstract>
      <url hash="88569f68">2020.findings-emnlp.255</url>
      <attachment type="OptionalSupplementaryMaterial" hash="ef7dca1e">2020.findings-emnlp.255.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.255</doi>
    </paper>
    <paper id="256">
      <title>Target Conditioning for One-to-Many Generation</title>
      <author><first>Marie-Anne</first><last>Lachaux</last></author>
      <author><first>Armand</first><last>Joulin</last></author>
      <author><first>Guillaume</first><last>Lample</last></author>
      <pages>2853–2862</pages>
      <abstract>Neural Machine Translation (NMT) models often lack diversity in their generated translations, even when paired with search algorithm, like beam search. A challenge is that the diversity in translations are caused by the variability in the target language, and cannot be inferred from the source sentence alone. In this paper, we propose to explicitly model this one-to-many mapping by conditioning the decoder of a NMT model on a latent variable that represents the domain of target sentences. The domain is a discrete variable generated by a target encoder that is jointly trained with the NMT model.The predicted domain of target sentences are given as input to the decoder during training. At inference, we can generate diverse translations by decoding with different domains. Unlike our strongest baseline (Shen et al., 2019), our method can scale to any number of domains without affecting the performance or the training time. We assess the quality and diversity of translations generated by our model with several metrics, on three different datasets.</abstract>
      <url hash="a8ea069d">2020.findings-emnlp.256</url>
      <doi>10.18653/v1/2020.findings-emnlp.256</doi>
    </paper>
    <paper id="257">
      <title>Can Pre-training help <fixed-case>VQA</fixed-case> with Lexical Variations?</title>
      <author><first>Shailza</first><last>Jolly</last></author>
      <author><first>Shubham</first><last>Kapoor</last></author>
      <pages>2863–2868</pages>
      <abstract>Rephrasings or paraphrases are sentences with similar meanings expressed in different ways. Visual Question Answering (VQA) models are closing the gap with the oracle performance for datasets like VQA2.0. However, these models fail to perform well on rephrasings of a question, which raises some important questions like Are these models robust towards linguistic variations? Is it the architecture or the dataset that we need to optimize? In this paper, we analyzed VQA models in the space of paraphrasing. We explored the role of language &amp; cross-modal pre-training to investigate the robustness of VQA models towards lexical variations. Our experiments find that pre-trained language encoders generate efficient representations of question rephrasings, which help VQA models correctly infer these samples. We empirically determine why pre-training language encoders improve lexical robustness. Finally, we observe that although pre-training all VQA components obtain state-of-the-art results on the VQA-Rephrasings dataset, it still fails to completely close the performance gap between original and rephrasing validation splits.</abstract>
      <url hash="80ad6d4c">2020.findings-emnlp.257</url>
      <attachment type="OptionalSupplementaryMaterial" hash="136e2032">2020.findings-emnlp.257.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.257</doi>
    </paper>
    <paper id="258">
      <title><fixed-case>FENAS</fixed-case>: Flexible and Expressive Neural Architecture Search</title>
      <author><first>Ramakanth</first><last>Pasunuru</last></author>
      <author><first>Mohit</first><last>Bansal</last></author>
      <pages>2869–2876</pages>
      <abstract>Architecture search is the automatic process of designing the model or cell structure that is optimal for the given dataset or task. Recently, this approach has shown good improvements in terms of performance (tested on language modeling and image classification) with reasonable training speed using a weight sharing-based approach called Efficient Neural Architecture Search (ENAS). In this work, we propose a novel architecture search algorithm called Flexible and Expressible Neural Architecture Search (FENAS), with more flexible and expressible search space than ENAS, in terms of more activation functions, input edges, and atomic operations. Also, our FENAS approach is able to reproduce the well-known LSTM and GRU architectures (unlike ENAS), and is also able to initialize with them for finding architectures more efficiently. We explore this extended search space via evolutionary search and show that FENAS performs significantly better on several popular text classification tasks and performs similar to ENAS on standard language model benchmark. Further, we present ablations and analyses on our FENAS approach.</abstract>
      <url hash="ec1ad3ae">2020.findings-emnlp.258</url>
      <doi>10.18653/v1/2020.findings-emnlp.258</doi>
    </paper>
    <paper id="259">
      <title>Inferring symmetry in natural language</title>
      <author><first>Chelsea</first><last>Tanchip</last></author>
      <author><first>Lei</first><last>Yu</last></author>
      <author><first>Aotao</first><last>Xu</last></author>
      <author><first>Yang</first><last>Xu</last></author>
      <pages>2877–2886</pages>
      <abstract>We present a methodological framework for inferring symmetry of verb predicates in natural language. Empirical work on predicate symmetry has taken two main approaches. The feature-based approach focuses on linguistic features pertaining to symmetry. The context-based approach denies the existence of absolute symmetry but instead argues that such inference is context dependent. We develop methods that formalize these approaches and evaluate them against a novel symmetry inference sentence (SIS) dataset comprised of 400 naturalistic usages of literature-informed verbs spanning the spectrum of symmetry-asymmetry. Our results show that a hybrid transfer learning model that integrates linguistic features with contextualized language models most faithfully predicts the empirical data. Our work integrates existing approaches to symmetry in natural language and suggests how symmetry inference can improve systematicity in state-of-the-art language models.</abstract>
      <url hash="0c951429">2020.findings-emnlp.259</url>
      <doi>10.18653/v1/2020.findings-emnlp.259</doi>
    </paper>
    <paper id="260">
      <title>A Concise Model for Multi-Criteria <fixed-case>C</fixed-case>hinese Word Segmentation with Transformer Encoder</title>
      <author><first>Xipeng</first><last>Qiu</last></author>
      <author><first>Hengzhi</first><last>Pei</last></author>
      <author><first>Hang</first><last>Yan</last></author>
      <author><first>Xuanjing</first><last>Huang</last></author>
      <pages>2887–2897</pages>
      <abstract>Multi-criteria Chinese word segmentation (MCCWS) aims to exploit the relations among the multiple heterogeneous segmentation criteria and further improve the performance of each single criterion. Previous work usually regards MCCWS as different tasks, which are learned together under the multi-task learning framework. In this paper, we propose a concise but effective unified model for MCCWS, which is fully-shared for all the criteria. By leveraging the powerful ability of the Transformer encoder, the proposed unified model can segment Chinese text according to a unique criterion-token indicating the output criterion. Besides, the proposed unified model can segment both simplified and traditional Chinese and has an excellent transfer capability. Experiments on eight datasets with different criteria show that our model outperforms our single-criterion baseline model and other multi-criteria models. Source codes of this paper are available on Github.</abstract>
      <url hash="e03d06cf">2020.findings-emnlp.260</url>
      <doi>10.18653/v1/2020.findings-emnlp.260</doi>
    </paper>
    <paper id="261">
      <title><fixed-case>LEGAL</fixed-case>-<fixed-case>BERT</fixed-case>: The Muppets straight out of Law School</title>
      <author><first>Ilias</first><last>Chalkidis</last></author>
      <author><first>Manos</first><last>Fergadiotis</last></author>
      <author><first>Prodromos</first><last>Malakasiotis</last></author>
      <author><first>Nikolaos</first><last>Aletras</last></author>
      <author><first>Ion</first><last>Androutsopoulos</last></author>
      <pages>2898–2904</pages>
      <abstract>BERT has achieved impressive performance in several NLP tasks. However, there has been limited investigation on its adaptation guidelines in specialised domains. Here we focus on the legal domain, where we explore several approaches for applying BERT models to downstream legal tasks, evaluating on multiple datasets. Our findings indicate that the previous guidelines for pre-training and fine-tuning, often blindly followed, do not always generalize well in the legal domain. Thus we propose a systematic investigation of the available strategies when applying BERT in specialised domains. These are: (a) use the original BERT out of the box, (b) adapt BERT by additional pre-training on domain-specific corpora, and (c) pre-train BERT from scratch on domain-specific corpora. We also propose a broader hyper-parameter search space when fine-tuning for downstream tasks and we release LEGAL-BERT, a family of BERT models intended to assist legal NLP research, computational law, and legal technology applications.</abstract>
      <url hash="0513de86">2020.findings-emnlp.261</url>
      <attachment type="OptionalSupplementaryMaterial" hash="ae036d6d">2020.findings-emnlp.261.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.261</doi>
    </paper>
    <paper id="262">
      <title>Enhancing Content Planning for Table-to-Text Generation with Data Understanding and Verification</title>
      <author><first>Heng</first><last>Gong</last></author>
      <author><first>Wei</first><last>Bi</last></author>
      <author><first>Xiaocheng</first><last>Feng</last></author>
      <author><first>Bing</first><last>Qin</last></author>
      <author><first>Xiaojiang</first><last>Liu</last></author>
      <author><first>Ting</first><last>Liu</last></author>
      <pages>2905–2914</pages>
      <abstract>Neural table-to-text models, which select and order salient data, as well as verbalizing them fluently via surface realization, have achieved promising progress. Based on results from previous work, the performance bottleneck of current models lies in the stage of content planing (selecting and ordering salient content from the input). That is, performance drops drastically when an oracle content plan is replaced by a model-inferred one during surface realization. In this paper, we propose to enhance neural content planning by (1) understanding data values with contextual numerical value representations that bring the sense of value comparison into content planning; (2) verifying the importance and ordering of the selected sequence of records with policy gradient. We evaluated our model on ROTOWIRE and MLB, two datasets on this task, and results show that our model outperforms existing systems with respect to content planning metrics.</abstract>
      <url hash="f9bcf229">2020.findings-emnlp.262</url>
      <attachment type="OptionalSupplementaryMaterial" hash="463da81a">2020.findings-emnlp.262.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.262</doi>
    </paper>
    <paper id="263">
      <title>Contextual Text Style Transfer</title>
      <author><first>Yu</first><last>Cheng</last></author>
      <author><first>Zhe</first><last>Gan</last></author>
      <author><first>Yizhe</first><last>Zhang</last></author>
      <author><first>Oussama</first><last>Elachqar</last></author>
      <author><first>Dianqi</first><last>Li</last></author>
      <author><first>Jingjing</first><last>Liu</last></author>
      <pages>2915–2924</pages>
      <abstract>We introduce a new task, Contextual Text Style Transfer - translating a sentence into a desired style with its surrounding context taken into account. This brings two key challenges to existing style transfer approaches: (I) how to preserve the semantic meaning of target sentence and its consistency with surrounding context during transfer; (ii) how to train a robust model with limited labeled data accompanied by context. To realize high-quality style transfer with natural context preservation, we propose a Context-Aware Style Transfer (CAST) model, which uses two separate encoders for each input sentence and its surrounding context. A classifier is further trained to ensure contextual consistency of the generated sentence. To compensate for the lack of parallel data, additional self-reconstruction and back-translation losses are introduced to leverage non-parallel data in a semi-supervised fashion. Two new benchmarks, Enron-Context and Reddit-Context, are introduced for formality and offensiveness style transfer. Experimental results on these datasets demonstrate the effectiveness of the proposed CAST model over state-of-the-art methods across style accuracy, content preservation and contextual consistency metrics.</abstract>
      <url hash="92398dd9">2020.findings-emnlp.263</url>
      <doi>10.18653/v1/2020.findings-emnlp.263</doi>
    </paper>
    <paper id="264">
      <title><fixed-case>D</fixed-case>i<fixed-case>P</fixed-case>air: Fast and Accurate Distillation for Trillion-Scale Text Matching and Pair Modeling</title>
      <author><first>Jiecao</first><last>Chen</last></author>
      <author><first>Liu</first><last>Yang</last></author>
      <author><first>Karthik</first><last>Raman</last></author>
      <author><first>Michael</first><last>Bendersky</last></author>
      <author><first>Jung-Jung</first><last>Yeh</last></author>
      <author><first>Yun</first><last>Zhou</last></author>
      <author><first>Marc</first><last>Najork</last></author>
      <author><first>Danyang</first><last>Cai</last></author>
      <author><first>Ehsan</first><last>Emadzadeh</last></author>
      <pages>2925–2937</pages>
      <abstract>Pre-trained models like BERT ((Devlin et al., 2018) have dominated NLP / IR applications such as single sentence classification, text pair classification, and question answering. However, deploying these models in real systems is highly non-trivial due to their exorbitant computational costs. A common remedy to this is knowledge distillation (Hinton et al., 2015), leading to faster inference. However – as we show here – existing works are not optimized for dealing with pairs (or tuples) of texts. Consequently, they are either not scalable or demonstrate subpar performance. In this work, we propose DiPair — a novel framework for distilling fast and accurate models on text pair tasks. Coupled with an end-to-end training strategy, DiPair is both highly scalable and offers improved quality-speed tradeoffs. Empirical studies conducted on both academic and real-world e-commerce benchmarks demonstrate the efficacy of the proposed approach with speedups of over 350x and minimal quality drop relative to the cross-attention teacher BERT model.</abstract>
      <url hash="e9b24959">2020.findings-emnlp.264</url>
      <doi>10.18653/v1/2020.findings-emnlp.264</doi>
    </paper>
    <paper id="265">
      <title>Cross-Lingual Dependency Parsing by <fixed-case>POS</fixed-case>-Guided Word Reordering</title>
      <author><first>Lu</first><last>Liu</last></author>
      <author><first>Yi</first><last>Zhou</last></author>
      <author><first>Jianhan</first><last>Xu</last></author>
      <author><first>Xiaoqing</first><last>Zheng</last></author>
      <author><first>Kai-Wei</first><last>Chang</last></author>
      <author><first>Xuanjing</first><last>Huang</last></author>
      <pages>2938–2948</pages>
      <abstract>We propose a novel approach to cross-lingual dependency parsing based on word reordering. The words in each sentence of a source language corpus are rearranged to meet the word order in a target language under the guidance of a part-of-speech based language model (LM). To obtain the highest reordering score under the LM, a population-based optimization algorithm and its genetic operators are designed to deal with the combinatorial nature of such word reordering. A parser trained on the reordered corpus then can be used to parse sentences in the target language. We demonstrate through extensive experimentation that our approach achieves better or comparable results across 25 target languages (1.73% increase in average), and outperforms a baseline by a significant margin on the languages that are greatly different from the source one. For example, when transferring the English parser to Hindi and Latin, our approach outperforms the baseline by 15.3% and 6.7% respectively.</abstract>
      <url hash="509eb391">2020.findings-emnlp.265</url>
      <doi>10.18653/v1/2020.findings-emnlp.265</doi>
    </paper>
    <paper id="266">
      <title>Assessing Robustness of Text Classification through Maximal Safe Radius Computation</title>
      <author><first>Emanuele</first><last>La Malfa</last></author>
      <author><first>Min</first><last>Wu</last></author>
      <author><first>Luca</first><last>Laurenti</last></author>
      <author><first>Benjie</first><last>Wang</last></author>
      <author><first>Anthony</first><last>Hartshorn</last></author>
      <author><first>Marta</first><last>Kwiatkowska</last></author>
      <pages>2949–2968</pages>
      <abstract>Neural network NLP models are vulnerable to small modifications of the input that maintain the original meaning but result in a different prediction. In this paper, we focus on robustness of text classification against word substitutions, aiming to provide guarantees that the model prediction does not change if a word is replaced with a plausible alternative, such as a synonym. As a measure of robustness, we adopt the notion of the maximal safe radius for a given input text, which is the minimum distance in the embedding space to the decision boundary. Since computing the exact maximal safe radius is not feasible in practice, we instead approximate it by computing a lower and upper bound. For the upper bound computation, we employ Monte Carlo Tree Search in conjunction with syntactic filtering to analyse the effect of single and multiple word substitutions. The lower bound computation is achieved through an adaptation of the linear bounding techniques implemented in tools CNN-Cert and POPQORN, respectively for convolutional and recurrent network models. We evaluate the methods on sentiment analysis and news classification models for four datasets (IMDB, SST, AG News and NEWS) and a range of embeddings, and provide an analysis of robustness trends. We also apply our framework to interpretability analysis and compare it with LIME.</abstract>
      <url hash="9315641c">2020.findings-emnlp.266</url>
      <doi>10.18653/v1/2020.findings-emnlp.266</doi>
    </paper>
    <paper id="267">
      <title>Social Commonsense Reasoning with Multi-Head Knowledge Attention</title>
      <author><first>Debjit</first><last>Paul</last></author>
      <author><first>Anette</first><last>Frank</last></author>
      <pages>2969–2980</pages>
      <abstract>Social Commonsense Reasoning requires understanding of text, knowledge about social events and their pragmatic implications, as well as commonsense reasoning skills. In this work we propose a novel multi-head knowledge attention model that encodes semi-structured commonsense inference rules and learns to incorporate them in a transformer-based reasoning cell.We assess the model’s performance on two tasks that require different reasoning skills: Abductive Natural Language Inference and Counterfactual Invariance Prediction as a new task. We show that our proposed model improves performance over strong state-of-the-art models (i.e., RoBERTa) across both reasoning tasks. Notably we are, to the best of our knowledge, the first to demonstrate that a model that learns to perform counterfactual reasoning helps predicting the best explanation in an abductive reasoning task. We validate the robustness of the model’s reasoning capabilities by perturbing the knowledge and provide qualitative analysis on the model’s knowledge incorporation capabilities.</abstract>
      <url hash="35663386">2020.findings-emnlp.267</url>
      <attachment type="OptionalSupplementaryMaterial" hash="b242d918">2020.findings-emnlp.267.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.267</doi>
    </paper>
    <paper id="268">
      <title><fixed-case>T</fixed-case>urn<fixed-case>GPT</fixed-case>: a Transformer-based Language Model for Predicting Turn-taking in Spoken Dialog</title>
      <author><first>Erik</first><last>Ekstedt</last></author>
      <author><first>Gabriel</first><last>Skantze</last></author>
      <pages>2981–2990</pages>
      <abstract>Syntactic and pragmatic completeness is known to be important for turn-taking prediction, but so far machine learning models of turn-taking have used such linguistic information in a limited way. In this paper, we introduce TurnGPT, a transformer-based language model for predicting turn-shifts in spoken dialog. The model has been trained and evaluated on a variety of written and spoken dialog datasets. We show that the model outperforms two baselines used in prior work. We also report on an ablation study, as well as attention and gradient analyses, which show that the model is able to utilize the dialog context and pragmatic completeness for turn-taking prediction. Finally, we explore the model’s potential in not only detecting, but also projecting, turn-completions.</abstract>
      <url hash="fbe86a5f">2020.findings-emnlp.268</url>
      <doi>10.18653/v1/2020.findings-emnlp.268</doi>
    </paper>
    <paper id="269">
      <title>A little goes a long way: Improving toxic language classification despite data scarcity</title>
      <author><first>Mika</first><last>Juuti</last></author>
      <author><first>Tommi</first><last>Gröndahl</last></author>
      <author><first>Adrian</first><last>Flanagan</last></author>
      <author><first>N.</first><last>Asokan</last></author>
      <pages>2991–3009</pages>
      <abstract>Detection of some types of toxic language is hampered by extreme scarcity of labeled training data. Data augmentation – generating new synthetic data from a labeled seed dataset – can help. The efficacy of data augmentation on toxic language classification has not been fully explored. We present the first systematic study on how data augmentation techniques impact performance across toxic language classifiers, ranging from shallow logistic regression architectures to BERT – a state-of-the-art pretrained Transformer network. We compare the performance of eight techniques on very scarce seed datasets. We show that while BERT performed the best, shallow classifiers performed comparably when trained on data augmented with a combination of three techniques, including GPT-2-generated sentences. We discuss the interplay of performance and computational overhead, which can inform the choice of techniques under different constraints.</abstract>
      <url hash="72441574">2020.findings-emnlp.269</url>
      <doi>10.18653/v1/2020.findings-emnlp.269</doi>
    </paper>
    <paper id="270">
      <title><fixed-case>A</fixed-case>n <fixed-case>I</fixed-case>nstance <fixed-case>L</fixed-case>evel <fixed-case>A</fixed-case>pproach for <fixed-case>S</fixed-case>hallow <fixed-case>S</fixed-case>emantic <fixed-case>P</fixed-case>arsing in <fixed-case>S</fixed-case>cientific <fixed-case>P</fixed-case>rocedural <fixed-case>T</fixed-case>ext</title>
      <author><first>Daivik</first><last>Swarup</last></author>
      <author><first>Ahsaas</first><last>Bajaj</last></author>
      <author><first>Sheshera</first><last>Mysore</last></author>
      <author><first>Tim</first><last>O’Gorman</last></author>
      <author><first>Rajarshi</first><last>Das</last></author>
      <author><first>Andrew</first><last>McCallum</last></author>
      <pages>3010–3017</pages>
      <abstract>In specific domains, such as procedural scientific text, human labeled data for shallow semantic parsing is especially limited and expensive to create. Fortunately, such specific domains often use rather formulaic writing, such that the different ways of expressing relations in a small number of grammatically similar labeled sentences may provide high coverage of semantic structures in the corpus, through an appropriately rich similarity metric. In light of this opportunity, this paper explores an instance-based approach to the relation prediction sub-task within shallow semantic parsing, in which semantic labels from structurally similar sentences in the training set are copied to test sentences. Candidate similar sentences are retrieved using SciBERT embeddings. For labels where it is possible to copy from a similar sentence we employ an instance level copy network, when this is not possible, a globally shared parametric model is employed. Experiments show our approach outperforms both baseline and prior methods by 0.75 to 3 F1 absolute in the Wet Lab Protocol Corpus and 1 F1 absolute in the Materials Science Procedural Text Corpus.</abstract>
      <url hash="8fa2d162">2020.findings-emnlp.270</url>
      <doi>10.18653/v1/2020.findings-emnlp.270</doi>
    </paper>
    <paper id="271">
      <title>General Purpose Text Embeddings from Pre-trained Language Models for Scalable Inference</title>
      <author><first>Jingfei</first><last>Du</last></author>
      <author><first>Myle</first><last>Ott</last></author>
      <author><first>Haoran</first><last>Li</last></author>
      <author><first>Xing</first><last>Zhou</last></author>
      <author><first>Veselin</first><last>Stoyanov</last></author>
      <pages>3018–3030</pages>
      <abstract>The state of the art on many NLP tasks is currently achieved by large pre-trained language models, which require a considerable amount of computation. We aim to reduce the inference cost in a setting where many different predictions are made on a single piece of text. In that case, computational cost during inference can be amortized over the different predictions (tasks) using a shared text encoder. We compare approaches for training such an encoder and show that encoders pre-trained over multiple tasks generalize well to unseen tasks. We also compare ways of extracting fixed- and limited-size representations from this encoder, including pooling features extracted from multiple layers or positions. Our best approach compares favorably to knowledge distillation, achieving higher accuracy and lower computational cost once the system is handling around 7 tasks. Further, we show that through binary quantization, we can reduce the size of the extracted representations by a factor of 16 to store them for later use. The resulting method offers a compelling solution for using large-scale pre-trained models at a fraction of the computational cost when multiple tasks are performed on the same text.</abstract>
      <url hash="e27a0b71">2020.findings-emnlp.271</url>
      <doi>10.18653/v1/2020.findings-emnlp.271</doi>
    </paper>
    <paper id="272">
      <title>Learning to Model and Ignore Dataset Bias with Mixed Capacity Ensembles</title>
      <author><first>Christopher</first><last>Clark</last></author>
      <author><first>Mark</first><last>Yatskar</last></author>
      <author><first>Luke</first><last>Zettlemoyer</last></author>
      <pages>3031–3045</pages>
      <abstract>Many datasets have been shown to contain incidental correlations created by idiosyncrasies in the data collection process. For example, sentence entailment datasets can have spurious word-class correlations if nearly all contradiction sentences contain the word “not”, and image recognition datasets can have tell-tale object-background correlations if dogs are always indoors. In this paper, we propose a method that can automatically detect and ignore these kinds of dataset-specific patterns, which we call dataset biases. Our method trains a lower capacity model in an ensemble with a higher capacity model. During training, the lower capacity model learns to capture relatively shallow correlations, which we hypothesize are likely to reflect dataset bias. This frees the higher capacity model to focus on patterns that should generalize better. We ensure the models learn non-overlapping approaches by introducing a novel method to make them conditionally independent. Importantly, our approach does not require the bias to be known in advance. We evaluate performance on synthetic datasets, and four datasets built to penalize models that exploit known biases on textual entailment, visual question answering, and image recognition tasks. We show improvement in all settings, including a 10 point gain on the visual question answering dataset.</abstract>
      <url hash="0b379b3c">2020.findings-emnlp.272</url>
      <doi>10.18653/v1/2020.findings-emnlp.272</doi>
    </paper>
    <paper id="273">
      <title>Learning to Generalize for Sequential Decision Making</title>
      <author><first>Xusen</first><last>Yin</last></author>
      <author><first>Ralph</first><last>Weischedel</last></author>
      <author><first>Jonathan</first><last>May</last></author>
      <pages>3046–3063</pages>
      <abstract>We consider problems of making sequences of decisions to accomplish tasks, interacting via the medium of language. These problems are often tackled with reinforcement learning approaches. We find that these models do not generalize well when applied to novel task domains. However, the large amount of computation necessary to adequately train and explore the search space of sequential decision making, under a reinforcement learning paradigm, precludes the inclusion of large contextualized language models, which might otherwise enable the desired generalization ability. We introduce a teacher-student imitation learning methodology and a means of converting a reinforcement learning model into a natural language understanding model. Together, these methodologies enable the introduction of contextualized language models into the sequential decision making problem space. We show that models can learn faster and generalize more, leveraging both the imitation learning and the reformulation. Our models exceed teacher performance on various held-out decision problems, by up to 7% on in-domain problems and 24% on out-of-domain problems.</abstract>
      <url hash="6561f622">2020.findings-emnlp.273</url>
      <doi>10.18653/v1/2020.findings-emnlp.273</doi>
    </paper>
    <paper id="274">
      <title>Effective Crowd-Annotation of Participants, Interventions, and Outcomes in the Text of Clinical Trial Reports</title>
      <author><first>Markus</first><last>Zlabinger</last></author>
      <author><first>Marta</first><last>Sabou</last></author>
      <author><first>Sebastian</first><last>Hofstätter</last></author>
      <author><first>Allan</first><last>Hanbury</last></author>
      <pages>3064–3074</pages>
      <abstract>The search for Participants, Interventions, and Outcomes (PIO) in clinical trial reports is a critical task in Evidence Based Medicine. For an automatic PIO extraction, high-quality corpora are needed. Obtaining such a corpus from crowdworkers, however, has been shown to be ineffective since (i) workers usually lack domain-specific expertise to conduct the task with sufficient quality, and (ii) the standard approach of annotating entire abstracts of trial reports as one task-instance (i.e. HIT) leads to an uneven distribution in task effort. In this paper, we switch from entire abstract to sentence annotation, referred to as the SenBase approach. We build upon SenBase in SenSupport, where we compensate the lack of domain-specific expertise of crowdworkers by showing for each task-instance similar sentences that are already annotated by experts. Such tailored task-instance examples are retrieved via unsupervised semantic short-text similarity (SSTS) method – and we evaluate nine methods to find an effective solution for SenSupport. We compute the Cohen’s Kappa agreement between crowd-annotations and gold standard annotations and show that (i) both sentence-based approaches outperform a Baseline approach where entire abstracts are annotated; (ii) supporting annotators with tailored task-instance examples is the best performing approach with Kappa agreements of 0.78/0.75/0.69 for P, I, and O respectively.</abstract>
      <url hash="3e1118e4">2020.findings-emnlp.274</url>
      <attachment type="OptionalSupplementaryMaterial" hash="48e7918e">2020.findings-emnlp.274.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.274</doi>
    </paper>
    <paper id="275">
      <title><fixed-case>A</fixed-case>dversarial <fixed-case>G</fixed-case>rammatical <fixed-case>E</fixed-case>rror <fixed-case>C</fixed-case>orrection</title>
      <author><first>Vipul</first><last>Raheja</last></author>
      <author><first>Dimitris</first><last>Alikaniotis</last></author>
      <pages>3075–3087</pages>
      <abstract>Recent works in Grammatical Error Correction (GEC) have leveraged the progress in Neural Machine Translation (NMT), to learn rewrites from parallel corpora of grammatically incorrect and corrected sentences, achieving state-of-the-art results. At the same time, Generative Adversarial Networks (GANs) have been successful in generating realistic texts across many different tasks by learning to directly minimize the difference between human-generated and synthetic text. In this work, we present an adversarial learning approach to GEC, using the generator-discriminator framework. The generator is a Transformer model, trained to produce grammatically correct sentences given grammatically incorrect ones. The discriminator is a sentence-pair classification model, trained to judge a given pair of grammatically incorrect-correct sentences on the quality of grammatical correction. We pre-train both the discriminator and the generator on parallel texts and then fine-tune them further using a policy gradient method that assigns high rewards to sentences which could be true corrections of the grammatically incorrect text. Experimental results on FCE, CoNLL-14, and BEA-19 datasets show that Adversarial-GEC can achieve competitive GEC quality compared to NMT-based baselines.</abstract>
      <url hash="ad2e7fe1">2020.findings-emnlp.275</url>
      <attachment type="OptionalSupplementaryMaterial" hash="973b7de3">2020.findings-emnlp.275.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.275</doi>
    </paper>
    <paper id="276">
      <title>On Long-Tailed Phenomena in Neural Machine Translation</title>
      <author><first>Vikas</first><last>Raunak</last></author>
      <author><first>Siddharth</first><last>Dalmia</last></author>
      <author><first>Vivek</first><last>Gupta</last></author>
      <author><first>Florian</first><last>Metze</last></author>
      <pages>3088–3095</pages>
      <abstract>State-of-the-art Neural Machine Translation (NMT) models struggle with generating low-frequency tokens, tackling which remains a major challenge. The analysis of long-tailed phenomena in the context of structured prediction tasks is further hindered by the added complexities of search during inference. In this work, we quantitatively characterize such long-tailed phenomena at two levels of abstraction, namely, token classification and sequence generation. We propose a new loss function, the Anti-Focal loss, to better adapt model training to the structural dependencies of conditional text generation by incorporating the inductive biases of beam search in the training process. We show the efficacy of the proposed technique on a number of Machine Translation (MT) datasets, demonstrating that it leads to significant gains over cross-entropy across different language pairs, especially on the generation of low-frequency words. We have released the code to reproduce our results.</abstract>
      <url hash="c764c6b2">2020.findings-emnlp.276</url>
      <doi>10.18653/v1/2020.findings-emnlp.276</doi>
    </paper>
    <paper id="277">
      <title>Knowing What You Know: Calibrating Dialogue Belief State Distributions via Ensembles</title>
      <author><first>Carel</first><last>van Niekerk</last></author>
      <author><first>Michael</first><last>Heck</last></author>
      <author><first>Christian</first><last>Geishauser</last></author>
      <author><first>Hsien-chin</first><last>Lin</last></author>
      <author><first>Nurul</first><last>Lubis</last></author>
      <author><first>Marco</first><last>Moresi</last></author>
      <author><first>Milica</first><last>Gasic</last></author>
      <pages>3096–3102</pages>
      <abstract>The ability to accurately track what happens during a conversation is essential for the performance of a dialogue system. Current state-of-the-art multi-domain dialogue state trackers achieve just over 55% accuracy on the current go-to benchmark, which means that in almost every second dialogue turn they place full confidence in an incorrect dialogue state. Belief trackers, on the other hand, maintain a distribution over possible dialogue states. However, they lack in performance compared to dialogue state trackers, and do not produce well calibrated distributions. In this work we present state-of-the-art performance in calibration for multi-domain dialogue belief trackers using a calibrated ensemble of models. Our resulting dialogue belief tracker also outperforms previous dialogue belief tracking models in terms of accuracy.</abstract>
      <url hash="fc9e36ee">2020.findings-emnlp.277</url>
      <doi>10.18653/v1/2020.findings-emnlp.277</doi>
    </paper>
    <paper id="278">
      <title><fixed-case>D</fixed-case>omain <fixed-case>A</fixed-case>dversarial <fixed-case>F</fixed-case>ine-<fixed-case>T</fixed-case>uning as an <fixed-case>E</fixed-case>ffective <fixed-case>R</fixed-case>egularizer</title>
      <author><first>Giorgos</first><last>Vernikos</last></author>
      <author><first>Katerina</first><last>Margatina</last></author>
      <author><first>Alexandra</first><last>Chronopoulou</last></author>
      <author><first>Ion</first><last>Androutsopoulos</last></author>
      <pages>3103–3112</pages>
      <abstract>In Natural Language Processing (NLP), pretrained language models (LMs) that are transferred to downstream tasks have been recently shown to achieve state-of-the-art results. However, standard fine-tuning can degrade the general-domain representations captured during pretraining. To address this issue, we introduce a new regularization technique, AFTER; domain Adversarial Fine-Tuning as an Effective Regularizer. Specifically, we complement the task-specific loss used during fine-tuning with an adversarial objective. This additional loss term is related to an adversarial classifier, that aims to discriminate between in-domain and out-of-domain text representations. Indomain refers to the labeled dataset of the task at hand while out-of-domain refers to unlabeled data from a different domain. Intuitively, the adversarial classifier acts as a regularize which prevents the model from overfitting to the task-specific domain. Empirical results on various natural language understanding tasks show that AFTER leads to improved performance compared to standard fine-tuning.</abstract>
      <url hash="59e9be9a">2020.findings-emnlp.278</url>
      <doi>10.18653/v1/2020.findings-emnlp.278</doi>
    </paper>
    <paper id="279">
      <title><fixed-case>CLAR</fixed-case>: A Cross-Lingual Argument Regularizer for Semantic Role Labeling</title>
      <author><first>Ishan</first><last>Jindal</last></author>
      <author><first>Yunyao</first><last>Li</last></author>
      <author><first>Siddhartha</first><last>Brahma</last></author>
      <author><first>Huaiyu</first><last>Zhu</last></author>
      <pages>3113–3125</pages>
      <abstract>Semantic role labeling (SRL) identifies predicate-argument structure(s) in a given sentence. Although different languages have different argument annotations, polyglot training, the idea of training one model on multiple languages, has previously been shown to outperform monolingual baselines, especially for low resource languages. In fact, even a simple combination of data has been shown to be effective with polyglot training by representing the distant vocabularies in a shared representation space. Meanwhile, despite the dissimilarity in argument annotations between languages, certain argument labels do share common semantic meaning across languages (e.g. adjuncts have more or less similar semantic meaning across languages). To leverage such similarity in annotation space across languages, we propose a method called Cross-Lingual Argument Regularizer (CLAR). CLAR identifies such linguistic annotation similarity across languages and exploits this information to map the target language arguments using a transformation of the space on which source language arguments lie. By doing so, our experimental results show that CLAR consistently improves SRL performance on multiple languages over monolingual and polyglot baselines for low resource languages.</abstract>
      <url hash="4724ca35">2020.findings-emnlp.279</url>
      <doi>10.18653/v1/2020.findings-emnlp.279</doi>
    </paper>
    <paper id="280">
      <title>Neutralizing Gender Bias in Word Embeddings with Latent Disentanglement and Counterfactual Generation</title>
      <author><first>Seungjae</first><last>Shin</last></author>
      <author><first>Kyungwoo</first><last>Song</last></author>
      <author><first>JoonHo</first><last>Jang</last></author>
      <author><first>Hyemi</first><last>Kim</last></author>
      <author><first>Weonyoung</first><last>Joo</last></author>
      <author><first>Il-Chul</first><last>Moon</last></author>
      <pages>3126–3140</pages>
      <abstract>Recent research demonstrates that word embeddings, trained on the human-generated corpus, have strong gender biases in embedding spaces, and these biases can result in the discriminative results from the various downstream tasks. Whereas the previous methods project word embeddings into a linear subspace for debiasing, we introduce a Latent Disentanglement method with a siamese auto-encoder structure with an adapted gradient reversal layer. Our structure enables the separation of the semantic latent information and gender latent information of given word into the disjoint latent dimensions. Afterwards, we introduce a Counterfactual Generation to convert the gender information of words, so the original and the modified embeddings can produce a gender-neutralized word embedding after geometric alignment regularization, without loss of semantic information. From the various quantitative and qualitative debiasing experiments, our method shows to be better than existing debiasing methods in debiasing word embeddings. In addition, Our method shows the ability to preserve semantic information during debiasing by minimizing the semantic information losses for extrinsic NLP downstream tasks.</abstract>
      <url hash="00e9a530">2020.findings-emnlp.280</url>
      <doi>10.18653/v1/2020.findings-emnlp.280</doi>
    </paper>
    <paper id="281">
      <title>Towards Domain-Independent Text Structuring Trainable on Large Discourse Treebanks</title>
      <author><first>Grigorii</first><last>Guz</last></author>
      <author><first>Giuseppe</first><last>Carenini</last></author>
      <pages>3141–3152</pages>
      <abstract>Text structuring is a fundamental step in NLG, especially when generating multi-sentential text. With the goal of fostering more general and data-driven approaches to text structuring, we propose the new and domain-independent NLG task of structuring and ordering a (possibly large) set of EDUs. We then present a solution for this task that combines neural dependency tree induction with pointer networks, and can be trained on large discourse treebanks that have only recently become available. Further, we propose a new evaluation metric that is arguably more suitable for our new task compared to existing content ordering metrics. Finally, we empirically show that our approach outperforms competitive alternatives on the proposed measure and is equivalent in performance with respect to previously established measures.</abstract>
      <url hash="22c5e1cc">2020.findings-emnlp.281</url>
      <doi>10.18653/v1/2020.findings-emnlp.281</doi>
    </paper>
    <paper id="282">
      <title><fixed-case>D</fixed-case>ata <fixed-case>A</fixed-case>nnealing for <fixed-case>I</fixed-case>nformal <fixed-case>L</fixed-case>anguage <fixed-case>U</fixed-case>nderstanding <fixed-case>T</fixed-case>asks</title>
      <author><first>Jing</first><last>Gu</last></author>
      <author><first>Zhou</first><last>Yu</last></author>
      <pages>3153–3159</pages>
      <abstract>There is a huge performance gap between formal and informal language understanding tasks. The recent pre-trained models that improved formal language understanding tasks did not achieve a comparable result on informal language. We propose data annealing transfer learning procedure to bridge the performance gap on informal natural language understanding tasks. It successfully utilizes a pre-trained model such as BERT in informal language. In the data annealing procedure, the training set contains mainly formal text data at first; then, the proportion of the informal text data is gradually increased during the training process. Our data annealing procedure is model-independent and can be applied to various tasks. We validate its effectiveness in exhaustive experiments. When BERT is implemented with our learning procedure, it outperforms all the state-of-the-art models on the three common informal language tasks.</abstract>
      <url hash="6c4632fe">2020.findings-emnlp.282</url>
      <doi>10.18653/v1/2020.findings-emnlp.282</doi>
    </paper>
    <paper id="283">
      <title>A Multilingual View of Unsupervised Machine Translation</title>
      <author><first>Xavier</first><last>Garcia</last></author>
      <author><first>Pierre</first><last>Foret</last></author>
      <author><first>Thibault</first><last>Sellam</last></author>
      <author><first>Ankur</first><last>Parikh</last></author>
      <pages>3160–3170</pages>
      <abstract>We present a probabilistic framework for multilingual neural machine translation that encompasses supervised and unsupervised setups, focusing on unsupervised translation. In addition to studying the vanilla case where there is only monolingual data available, we propose a novel setup where one language in the (source, target) pair is not associated with any parallel data, but there may exist auxiliary parallel data that contains the other. This auxiliary data can naturally be utilized in our probabilistic framework via a novel cross-translation loss term. Empirically, we show that our approach results in higher BLEU scores over state-of-the-art unsupervised models on the WMT’14 English-French, WMT’16 English-German, and WMT’16 English-Romanian datasets in most directions.</abstract>
      <url hash="3366d192">2020.findings-emnlp.283</url>
      <doi>10.18653/v1/2020.findings-emnlp.283</doi>
    </paper>
    <paper id="284">
      <title>An Evaluation Method for Diachronic Word Sense Induction</title>
      <author><first>Ashjan</first><last>Alsulaimani</last></author>
      <author><first>Erwan</first><last>Moreau</last></author>
      <author><first>Carl</first><last>Vogel</last></author>
      <pages>3171–3180</pages>
      <abstract>The task of Diachronic Word Sense Induction (DWSI) aims to identify the meaning of words from their context, taking the temporal dimension into account. In this paper we propose an evaluation method based on large-scale time-stamped annotated biomedical data, and a range of evaluation measures suited to the task. The approach is applied to two recent DWSI systems, thus demonstrating its relevance and providing an in-depth analysis of the models.</abstract>
      <url hash="44dd3cfe">2020.findings-emnlp.284</url>
      <attachment type="OptionalSupplementaryMaterial" hash="597d0f18">2020.findings-emnlp.284.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.284</doi>
    </paper>
    <paper id="285">
      <title>Integrating Task Specific Information into Pretrained Language Models for Low Resource Fine Tuning</title>
      <author><first>Rui</first><last>Wang</last></author>
      <author><first>Shijing</first><last>Si</last></author>
      <author><first>Guoyin</first><last>Wang</last></author>
      <author><first>Lei</first><last>Zhang</last></author>
      <author><first>Lawrence</first><last>Carin</last></author>
      <author><first>Ricardo</first><last>Henao</last></author>
      <pages>3181–3186</pages>
      <abstract>Pretrained Language Models (PLMs) have improved the performance of natural language understanding in recent years. Such models are pretrained on large corpora, which encode the general prior knowledge of natural languages but are agnostic to information characteristic of downstream tasks. This often results in overfitting when fine-tuned with low resource datasets where task-specific information is limited. In this paper, we integrate label information as a task-specific prior into the self-attention component of pretrained BERT models. Experiments on several benchmarks and real-word datasets suggest that the proposed approach can largely improve the performance of pretrained models when fine-tuning with small datasets.</abstract>
      <url hash="75ef57d3">2020.findings-emnlp.285</url>
      <doi>10.18653/v1/2020.findings-emnlp.285</doi>
    </paper>
    <paper id="286">
      <title>Efficient Transformer-based Large Scale Language Representations using Hardware-friendly Block Structured Pruning</title>
      <author><first>Bingbing</first><last>Li</last></author>
      <author><first>Zhenglun</first><last>Kong</last></author>
      <author><first>Tianyun</first><last>Zhang</last></author>
      <author><first>Ji</first><last>Li</last></author>
      <author><first>Zhengang</first><last>Li</last></author>
      <author><first>Hang</first><last>Liu</last></author>
      <author><first>Caiwen</first><last>Ding</last></author>
      <pages>3187–3199</pages>
      <abstract>Pretrained large-scale language models have increasingly demonstrated high accuracy on many natural language processing (NLP) tasks. However, the limited weight storage and computational speed on hardware platforms have impeded the popularity of pretrained models, especially in the era of edge computing. In this work, we propose an efficient transformer-based large-scale language representation using hardware-friendly block structure pruning. We incorporate the reweighted group Lasso into block-structured pruning for optimization. Besides the significantly reduced weight storage and computation, the proposed approach achieves high compression rates. Experimental results on different models (BERT, RoBERTa, and DistilBERT) on the General Language Understanding Evaluation (GLUE) benchmark tasks show that we achieve up to 5.0x with zero or minor accuracy degradation on certain task(s). Our proposed method is also orthogonal to existing compact pretrained language models such as DistilBERT using knowledge distillation, since a further 1.79x average compression rate can be achieved on top of DistilBERT with zero or minor accuracy degradation. It is suitable to deploy the final compressed model on resource-constrained edge devices.</abstract>
      <url hash="4a9a2a18">2020.findings-emnlp.286</url>
      <doi>10.18653/v1/2020.findings-emnlp.286</doi>
    </paper>
    <paper id="287">
      <title><fixed-case>K</fixed-case>o<fixed-case>BE</fixed-case>: Knowledge-Based Machine Translation Evaluation</title>
      <author><first>Zorik</first><last>Gekhman</last></author>
      <author><first>Roee</first><last>Aharoni</last></author>
      <author><first>Genady</first><last>Beryozkin</last></author>
      <author><first>Markus</first><last>Freitag</last></author>
      <author><first>Wolfgang</first><last>Macherey</last></author>
      <pages>3200–3207</pages>
      <abstract>We propose a simple and effective method for machine translation evaluation which does not require reference translations. Our approach is based on (1) grounding the entity mentions found in each source sentence and candidate translation against a large-scale multilingual knowledge base, and (2) measuring the recall of the grounded entities found in the candidate vs. those found in the source. Our approach achieves the highest correlation with human judgements on 9 out of the 18 language pairs from the WMT19 benchmark for evaluation without references, which is the largest number of wins for a single evaluation method on this task. On 4 language pairs, we also achieve higher correlation with human judgements than BLEU. To foster further research, we release a dataset containing 1.8 million grounded entity mentions across 18 language pairs from the WMT19 metrics track data.</abstract>
      <url hash="f51928bc">2020.findings-emnlp.287</url>
      <doi>10.18653/v1/2020.findings-emnlp.287</doi>
    </paper>
    <paper id="288">
      <title>Pushing the Limits of <fixed-case>AMR</fixed-case> Parsing with Self-Learning</title>
      <author><first>Young-Suk</first><last>Lee</last></author>
      <author><first>Ramón</first><last>Fernandez Astudillo</last></author>
      <author><first>Tahira</first><last>Naseem</last></author>
      <author><first>Revanth</first><last>Gangi Reddy</last></author>
      <author><first>Radu</first><last>Florian</last></author>
      <author><first>Salim</first><last>Roukos</last></author>
      <pages>3208–3214</pages>
      <abstract>Abstract Meaning Representation (AMR) parsing has experienced a notable growth in performance in the last two years, due both to the impact of transfer learning and the development of novel architectures specific to AMR. At the same time, self-learning techniques have helped push the performance boundaries of other natural language processing applications, such as machine translation or question answering. In this paper, we explore different ways in which trained models can be applied to improve AMR parsing performance, including generation of synthetic text and AMR annotations as well as refinement of actions oracle. We show that, without any additional human annotations, these techniques improve an already performant parser and achieve state-of-the-art results on AMR 1.0 and AMR 2.0.</abstract>
      <url hash="b611731c">2020.findings-emnlp.288</url>
      <doi>10.18653/v1/2020.findings-emnlp.288</doi>
    </paper>
    <paper id="289">
      <title>Towards <fixed-case>Z</fixed-case>ero-<fixed-case>S</fixed-case>hot <fixed-case>C</fixed-case>onditional <fixed-case>S</fixed-case>ummarization with <fixed-case>A</fixed-case>daptive <fixed-case>M</fixed-case>ulti-<fixed-case>T</fixed-case>ask <fixed-case>F</fixed-case>ine-<fixed-case>T</fixed-case>uning</title>
      <author><first>Travis</first><last>Goodwin</last></author>
      <author><first>Max</first><last>Savery</last></author>
      <author><first>Dina</first><last>Demner-Fushman</last></author>
      <pages>3215–3226</pages>
      <abstract>Automatic summarization research has traditionally focused on providing high quality general-purpose summaries of documents. However, there are many applications which require more specific summaries, such as supporting question answering or topic-based literature discovery. In this paper we study the problem of conditional summarization in which content selection and surface realization are explicitly conditioned on an ad-hoc natural language question or topic description. Because of the difficulty in obtaining sufficient reference summaries to support arbitrary conditional summarization, we explore the use of multi-task fine-tuning (MTFT) on twenty-one natural language tasks to enable zero-shot conditional summarization on five tasks. We present four new summarization datasets, two novel “online” or adaptive task-mixing strategies, and report zero-shot performance using T5 and BART, demonstrating that MTFT can improve zero-shot summarization quality.</abstract>
      <url hash="65ef67a7">2020.findings-emnlp.289</url>
      <doi>10.18653/v1/2020.findings-emnlp.289</doi>
    </paper>
    <paper id="290">
      <title>Multilingual Knowledge Graph Completion via Ensemble Knowledge Transfer</title>
      <author><first>Xuelu</first><last>Chen</last></author>
      <author><first>Muhao</first><last>Chen</last></author>
      <author><first>Changjun</first><last>Fan</last></author>
      <author><first>Ankith</first><last>Uppunda</last></author>
      <author><first>Yizhou</first><last>Sun</last></author>
      <author><first>Carlo</first><last>Zaniolo</last></author>
      <pages>3227–3238</pages>
      <abstract>Predicting missing facts in a knowledge graph(KG) is a crucial task in knowledge base construction and reasoning, and it has been the subject of much research in recent works us-ing KG embeddings. While existing KG embedding approaches mainly learn and predict facts within a single KG, a more plausible solution would benefit from the knowledge in multiple language-specific KGs, considering that different KGs have their own strengths and limitations on data quality and coverage. This is quite challenging since the transfer of knowledge among multiple independently maintained KGs is often hindered by the insufficiency of alignment information and inconsistency of described facts. In this paper, we propose kens, a novel framework for embedding learning and ensemble knowledge transfer across a number of language-specific KGs.KEnS embeds all KGs in a shared embedding space, where the association of entities is captured based on self-learning. Then, KEnS performs ensemble inference to com-bine prediction results from multiple language-specific embeddings, for which multiple en-semble techniques are investigated. Experiments on the basis of five real-world language-specific KGs show that, by effectively identifying and leveraging complementary knowledge, KEnS consistently improves state-of-the-art methods on KG completion.</abstract>
      <url hash="1ac8bf7d">2020.findings-emnlp.290</url>
      <doi>10.18653/v1/2020.findings-emnlp.290</doi>
    </paper>
    <paper id="291">
      <title>Towards <fixed-case>C</fixed-case>ontrollable <fixed-case>B</fixed-case>iases in <fixed-case>L</fixed-case>anguage <fixed-case>G</fixed-case>eneration</title>
      <author><first>Emily</first><last>Sheng</last></author>
      <author><first>Kai-Wei</first><last>Chang</last></author>
      <author><first>Prem</first><last>Natarajan</last></author>
      <author><first>Nanyun</first><last>Peng</last></author>
      <pages>3239–3254</pages>
      <abstract>We present a general approach towards controllable societal biases in natural language generation (NLG). Building upon the idea of adversarial triggers, we develop a method to induce societal biases in generated text when input prompts contain mentions of specific demographic groups. We then analyze two scenarios: 1) inducing negative biases for one demographic and positive biases for another demographic, and 2) equalizing biases between demographics. The former scenario enables us to detect the types of biases present in the model. Specifically, we show the effectiveness of our approach at facilitating bias analysis by finding topics that correspond to demographic inequalities in generated text and comparing the relative effectiveness of inducing biases for different demographics. The second scenario is useful for mitigating biases in downstream applications such as dialogue generation. In our experiments, the mitigation technique proves to be effective at equalizing the amount of biases across demographics while simultaneously generating less negatively biased text overall.</abstract>
      <url hash="a4498350">2020.findings-emnlp.291</url>
      <doi>10.18653/v1/2020.findings-emnlp.291</doi>
    </paper>
    <paper id="292">
      <title><fixed-case>R</fixed-case>ob<fixed-case>BERT</fixed-case>: a <fixed-case>D</fixed-case>utch <fixed-case>R</fixed-case>o<fixed-case>BERT</fixed-case>a-based <fixed-case>L</fixed-case>anguage <fixed-case>M</fixed-case>odel</title>
      <author><first>Pieter</first><last>Delobelle</last></author>
      <author><first>Thomas</first><last>Winters</last></author>
      <author><first>Bettina</first><last>Berendt</last></author>
      <pages>3255–3265</pages>
      <abstract>Pre-trained language models have been dominating the field of natural language processing in recent years, and have led to significant performance gains for various complex natural language tasks. One of the most prominent pre-trained language models is BERT, which was released as an English as well as a multilingual version. Although multilingual BERT performs well on many tasks, recent studies show that BERT models trained on a single language significantly outperform the multilingual version. Training a Dutch BERT model thus has a lot of potential for a wide range of Dutch NLP tasks. While previous approaches have used earlier implementations of BERT to train a Dutch version of BERT, we used RoBERTa, a robustly optimized BERT approach, to train a Dutch language model called RobBERT. We measured its performance on various tasks as well as the importance of the fine-tuning dataset size. We also evaluated the importance of language-specific tokenizers and the model’s fairness. We found that RobBERT improves state-of-the-art results for various tasks, and especially significantly outperforms other models when dealing with smaller datasets. These results indicate that it is a powerful pre-trained model for a large variety of Dutch language tasks. The pre-trained and fine-tuned models are publicly available to support further downstream Dutch NLP applications.</abstract>
      <url hash="299414e6">2020.findings-emnlp.292</url>
      <doi>10.18653/v1/2020.findings-emnlp.292</doi>
    </paper>
    <paper id="293">
      <title>Regularization of Distinct Strategies for Unsupervised Question Generation</title>
      <author><first>Junmo</first><last>Kang</last></author>
      <author><first>Giwon</first><last>Hong</last></author>
      <author><first>Haritz</first><last>Puerto San Roman</last></author>
      <author><first>Sung-Hyon</first><last>Myaeng</last></author>
      <pages>3266–3277</pages>
      <abstract>Unsupervised question answering (UQA) has been proposed to avoid the high cost of creating high-quality datasets for QA. One approach to UQA is to train a QA model with questions generated automatically. However, the generated questions are either too similar to a word sequence in the context or too drifted from the semantics of the context, thereby making it difficult to train a robust QA model. We propose a novel regularization method based on teacher-student architecture to avoid bias toward a particular question generation strategy and modulate the process of generating individual words when a question is generated. Our experiments demonstrate that we have achieved the goal of generating higher-quality questions for UQA across diverse QA datasets and tasks. We also show that this method can be useful for creating a QA model with few-shot learning.</abstract>
      <url hash="28c1564c">2020.findings-emnlp.293</url>
      <doi>10.18653/v1/2020.findings-emnlp.293</doi>
    </paper>
    <paper id="294">
      <title>Graph-to-Graph Transformer for Transition-based Dependency Parsing</title>
      <author><first>Alireza</first><last>Mohammadshahi</last></author>
      <author><first>James</first><last>Henderson</last></author>
      <pages>3278–3289</pages>
      <abstract>We propose the Graph2Graph Transformer architecture for conditioning on and predicting arbitrary graphs, and apply it to the challenging task of transition-based dependency parsing. After proposing two novel Transformer models of transition-based dependency parsing as strong baselines, we show that adding the proposed mechanisms for conditioning on and predicting graphs of Graph2Graph Transformer results in significant improvements, both with and without BERT pre-training. The novel baselines and their integration with Graph2Graph Transformer significantly outperform the state-of-the-art in traditional transition-based dependency parsing on both English Penn Treebank, and 13 languages of Universal Dependencies Treebanks. Graph2Graph Transformer can be integrated with many previous structured prediction methods, making it easy to apply to a wide range of NLP tasks.</abstract>
      <url hash="4dbdf36d">2020.findings-emnlp.294</url>
      <attachment type="OptionalSupplementaryMaterial" hash="6cfb12a1">2020.findings-emnlp.294.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.294</doi>
    </paper>
    <paper id="295">
      <title><fixed-case>WER</fixed-case> we are and <fixed-case>WER</fixed-case> we think we are</title>
      <author><first>Piotr</first><last>Szymański</last></author>
      <author><first>Piotr</first><last>Żelasko</last></author>
      <author><first>Mikolaj</first><last>Morzy</last></author>
      <author><first>Adrian</first><last>Szymczak</last></author>
      <author><first>Marzena</first><last>Żyła-Hoppe</last></author>
      <author><first>Joanna</first><last>Banaszczak</last></author>
      <author><first>Lukasz</first><last>Augustyniak</last></author>
      <author><first>Jan</first><last>Mizgajski</last></author>
      <author><first>Yishay</first><last>Carmiel</last></author>
      <pages>3290–3295</pages>
      <abstract>Natural language processing of conversational speech requires the availability of high-quality transcripts. In this paper, we express our skepticism towards the recent reports of very low Word Error Rates (WERs) achieved by modern Automatic Speech Recognition (ASR) systems on benchmark datasets. We outline several problems with popular benchmarks and compare three state-of-the-art commercial ASR systems on an internal dataset of real-life spontaneous human conversations and HUB’05 public benchmark. We show that WERs are significantly higher than the best reported results. We formulate a set of guidelines which may aid in the creation of real-life, multi-domain datasets with high quality annotations for training and testing of robust ASR systems.</abstract>
      <url hash="a5cb2793">2020.findings-emnlp.295</url>
      <doi>10.18653/v1/2020.findings-emnlp.295</doi>
    </paper>
    <paper id="296">
      <title><fixed-case>D</fixed-case>e<fixed-case>SMOG</fixed-case>: Detecting Stance in Media On Global Warming</title>
      <author><first>Yiwei</first><last>Luo</last></author>
      <author><first>Dallas</first><last>Card</last></author>
      <author><first>Dan</first><last>Jurafsky</last></author>
      <pages>3296–3315</pages>
      <abstract>Citing opinions is a powerful yet understudied strategy in argumentation. For example, an environmental activist might say, “Leading scientists agree that global warming is a serious concern,” framing a clause which affirms their own stance (“that global warming is serious”) as an opinion endorsed ("[scientists] agree”) by a reputable source (“leading”). In contrast, a global warming denier might frame the same clause as the opinion of an untrustworthy source with a predicate connoting doubt: “Mistaken scientists claim [...]." Our work studies opinion-framing in the global warming (GW) debate, an increasingly partisan issue that has received little attention in NLP. We introduce DeSMOG, a dataset of stance-labeled GW sentences, and train a BERT classifier to study novel aspects of argumentation in how different sides of a debate represent their own and each other’s opinions. From 56K news articles, we find that similar linguistic devices for self-affirming and opponent-doubting discourse are used across GW-accepting and skeptic media, though GW-skeptical media shows more opponent-doubt. We also find that authors often characterize sources as hypocritical, by ascribing opinions expressing the author’s own view to source entities known to publicly endorse the opposing view. We release our stance dataset, model, and lexicons of framing devices for future work on opinion-framing and the automatic detection of GW stance.</abstract>
      <url hash="3f9e2b0b">2020.findings-emnlp.296</url>
      <doi>10.18653/v1/2020.findings-emnlp.296</doi>
    </paper>
    <paper id="297">
      <title>A Novel Challenge Set for <fixed-case>H</fixed-case>ebrew Morphological Disambiguation and Diacritics Restoration</title>
      <author><first>Avi</first><last>Shmidman</last></author>
      <author><first>Joshua</first><last>Guedalia</last></author>
      <author><first>Shaltiel</first><last>Shmidman</last></author>
      <author><first>Moshe</first><last>Koppel</last></author>
      <author><first>Reut</first><last>Tsarfaty</last></author>
      <pages>3316–3326</pages>
      <abstract>One of the primary tasks of morphological parsers is the disambiguation of homographs. Particularly difficult are cases of unbalanced ambiguity, where one of the possible analyses is far more frequent than the others. In such cases, there may not exist sufficient examples of the minority analyses in order to properly evaluate performance, nor to train effective classifiers. In this paper we address the issue of unbalanced morphological ambiguities in Hebrew. We offer a challenge set for Hebrew homographs — the first of its kind — containing substantial attestation of each analysis of 21 Hebrew homographs. We show that the current SOTA of Hebrew disambiguation performs poorly on cases of unbalanced ambiguity. Leveraging our new dataset, we achieve a new state-of-the-art for all 21 words, improving the overall average F1 score from 0.67 to 0.95. Our resulting annotated datasets are made publicly available for further research.</abstract>
      <url hash="df25a53d">2020.findings-emnlp.297</url>
      <doi>10.18653/v1/2020.findings-emnlp.297</doi>
    </paper>
    <paper id="298">
      <title>Improve Transformer Models with Better Relative Position Embeddings</title>
      <author><first>Zhiheng</first><last>Huang</last></author>
      <author><first>Davis</first><last>Liang</last></author>
      <author><first>Peng</first><last>Xu</last></author>
      <author><first>Bing</first><last>Xiang</last></author>
      <pages>3327–3335</pages>
      <abstract>The transformer model has demonstrated superior results on NLP tasks including machine translation and question answering. In this paper, we argue that the position information is not fully utilized in existing work. For example, the initial proposal of a sinusoid embedding is fixed and not learnable. In this paper, we first review the absolute position embeddings and existing relative position embedding methods. We then propose new methods to encourage increased interaction between query, key and relative position embeddings in the self-attention mechanism. Our most promising approach is a generalization of the absolute position embedding. Our method results in increased accuracy compared to previous approaches in absolute and relative position embeddings on the SQuAD1.1 dataset. In addition, we address the inductive property of whether a position embedding can be robust enough to handle long sequences. We demonstrate empirically that our relative embedding method can be reasonably generalized to and is robust in the inductive perspective. Finally, we show that our proposed method can be effectively and efficiently adopted as a near drop-in replacement for improving the accuracy of large models with little computational overhead.</abstract>
      <url hash="5737f4fc">2020.findings-emnlp.298</url>
      <doi>10.18653/v1/2020.findings-emnlp.298</doi>
    </paper>
    <paper id="299">
      <title>A Sentiment-Controllable Topic-to-Essay Generator with Topic Knowledge Graph</title>
      <author><first>Lin</first><last>Qiao</last></author>
      <author><first>Jianhao</first><last>Yan</last></author>
      <author><first>Fandong</first><last>Meng</last></author>
      <author><first>Zhendong</first><last>Yang</last></author>
      <author><first>Jie</first><last>Zhou</last></author>
      <pages>3336–3344</pages>
      <abstract>Generating a vivid, novel, and diverse essay with only several given topic words is a promising task of natural language generation. Previous work in this task exists two challenging problems: neglect of sentiment beneath the text and insufficient utilization of topic-related knowledge. Therefore, we propose a novel Sentiment Controllable topic-to- essay generator with a Topic Knowledge Graph enhanced decoder, named SCTKG, which is based on the conditional variational auto-encoder (CVAE) framework. We firstly inject the sentiment information into the generator for controlling sentiment for each sentence, which leads to various generated essays. Then we design a Topic Knowledge Graph enhanced decoder. Unlike existing models that use knowledge entities separately, our model treats knowledge graph as a whole and encodes more structured, connected semantic information in the graph to generate a more relevant essay. Experimental results show that our SCTKG can generate sentiment controllable essays and outperform the state-of-the-art approach in terms of topic relevance, fluency, and diversity on both automatic and human evaluation.</abstract>
      <url hash="2c850f3e">2020.findings-emnlp.299</url>
      <doi>10.18653/v1/2020.findings-emnlp.299</doi>
    </paper>
    <paper id="300">
      <title>What-if <fixed-case>I</fixed-case> ask you to explain: Explaining the effects of perturbations in procedural text</title>
      <author><first>Dheeraj</first><last>Rajagopal</last></author>
      <author><first>Niket</first><last>Tandon</last></author>
      <author><first>Peter</first><last>Clark</last></author>
      <author><first>Bhavana</first><last>Dalvi</last></author>
      <author><first>Eduard</first><last>Hovy</last></author>
      <pages>3345–3355</pages>
      <abstract>Our goal is to explain the effects of perturbations in procedural text, e.g., given a passage describing a rabbit’s life cycle, explain why illness (the perturbation) may reduce the rabbit population (the effect). Although modern systems are able to solve the original prediction task well (e.g., illness results in less rabbits), the explanation task - identifying the causal chain of events from perturbation to effect - remains largely unaddressed, and is the goal of this research. We present QUARTET, a system that constructs such explanations from paragraphs, by modeling the explanation task as a multitask learning problem. QUARTET constructs explanations from the sentences in the procedural text, achieving ~18 points better on explanation accuracy compared to several strong baselines on a recent process comprehension benchmark. On an end task on this benchmark, we show a surprising finding that good explanations do not have to come at the expense of end task performance, in fact leading to a 7% F1 improvement over SOTA.</abstract>
      <url hash="7e56c1b8">2020.findings-emnlp.300</url>
      <doi>10.18653/v1/2020.findings-emnlp.300</doi>
    </paper>
    <paper id="301">
      <title><fixed-case>R</fixed-case>eal<fixed-case>T</fixed-case>oxicity<fixed-case>P</fixed-case>rompts: Evaluating Neural Toxic Degeneration in Language Models</title>
      <author><first>Samuel</first><last>Gehman</last></author>
      <author><first>Suchin</first><last>Gururangan</last></author>
      <author><first>Maarten</first><last>Sap</last></author>
      <author><first>Yejin</first><last>Choi</last></author>
      <author><first>Noah A.</first><last>Smith</last></author>
      <pages>3356–3369</pages>
      <abstract>Pretrained neural language models (LMs) are prone to generating racist, sexist, or otherwise toxic language which hinders their safe deployment. We investigate the extent to which pretrained LMs can be prompted to generate toxic language, and the effectiveness of controllable text generation algorithms at preventing such toxic degeneration. We create and release RealToxicityPrompts, a dataset of 100K naturally occurring, sentence-level prompts derived from a large corpus of English web text, paired with toxicity scores from a widely-used toxicity classifier. Using RealToxicityPrompts, we find that pretrained LMs can degenerate into toxic text even from seemingly innocuous prompts. We empirically assess several controllable generation methods, and find that while data- or compute-intensive methods (e.g., adaptive pretraining on non-toxic data) are more effective at steering away from toxicity than simpler solutions (e.g., banning “bad” words), no current method is failsafe against neural toxic degeneration. To pinpoint the potential cause of such persistent toxic degeneration, we analyze two web text corpora used to pretrain several LMs (including GPT-2; Radford et. al, 2019), and find a significant amount of offensive, factually unreliable, and otherwise toxic content. Our work provides a test bed for evaluating toxic generations by LMs and stresses the need for better data selection processes for pretraining.</abstract>
      <url hash="9b74d94c">2020.findings-emnlp.301</url>
      <doi>10.18653/v1/2020.findings-emnlp.301</doi>
    </paper>
    <paper id="302">
      <title>Improving Event Duration Prediction via Time-aware Pre-training</title>
      <author><first>Zonglin</first><last>Yang</last></author>
      <author><first>Xinya</first><last>Du</last></author>
      <author><first>Alexander</first><last>Rush</last></author>
      <author><first>Claire</first><last>Cardie</last></author>
      <pages>3370–3378</pages>
      <abstract>End-to-end models in NLP rarely encode external world knowledge about length of time. We introduce two effective models for duration prediction, which incorporate external knowledge by reading temporal-related news sentences (time-aware pre-training). Specifically, one model predicts the range/unit where the duration value falls in (R-PRED); and the other predicts the exact duration value (E-PRED). Our best model – E-PRED, substantially outperforms previous work, and captures duration information more accurately than R-PRED. We also demonstrate our models are capable of duration prediction in the unsupervised setting, outperforming the baselines.</abstract>
      <url hash="1a764517">2020.findings-emnlp.302</url>
      <doi>10.18653/v1/2020.findings-emnlp.302</doi>
    </paper>
    <paper id="303">
      <title>Composed Variational Natural Language Generation for Few-shot Intents</title>
      <author><first>Congying</first><last>Xia</last></author>
      <author><first>Caiming</first><last>Xiong</last></author>
      <author><first>Philip</first><last>Yu</last></author>
      <author><first>Richard</first><last>Socher</last></author>
      <pages>3379–3388</pages>
      <abstract>In this paper, we focus on generating training examples for few-shot intents in the realistic imbalanced scenario. To build connections between existing many-shot intents and few-shot intents, we consider an intent as a combination of a domain and an action, and propose a composed variational natural language generator (CLANG), a transformer-based conditional variational autoencoder. CLANG utilizes two latent variables to represent the utterances corresponding to two different independent parts (domain and action) in the intent, and the latent variables are composed together to generate natural examples. Additionally, to improve the generator learning, we adopt the contrastive regularization loss that contrasts the in-class with the out-of-class utterance generation given the intent. To evaluate the quality of the generated utterances, experiments are conducted on the generalized few-shot intent detection task. Empirical results show that our proposed model achieves state-of-the-art performances on two real-world intent detection datasets.</abstract>
      <url hash="ab074329">2020.findings-emnlp.303</url>
      <doi>10.18653/v1/2020.findings-emnlp.303</doi>
    </paper>
    <paper id="304">
      <title>Literature Retrieval for Precision Medicine with Neural Matching and Faceted Summarization</title>
      <author><first>Jiho</first><last>Noh</last></author>
      <author><first>Ramakanth</first><last>Kavuluru</last></author>
      <pages>3389–3399</pages>
      <abstract>Information retrieval (IR) for precision medicine (PM) often involves looking for multiple pieces of evidence that characterize a patient case. This typically includes at least the name of a condition and a genetic variation that applies to the patient. Other factors such as demographic attributes, comorbidities, and social determinants may also be pertinent. As such, the retrieval problem is often formulated as ad hoc search but with multiple facets (e.g., disease, mutation) that may need to be incorporated. In this paper, we present a document reranking approach that combines neural query-document matching and text summarization toward such retrieval scenarios. Our architecture builds on the basic BERT model with three specific components for reranking: (a). document-query matching (b). keyword extraction and (c). facet-conditioned abstractive summarization. The outcomes of (b) and (c) are used to essentially transform a candidate document into a concise summary that can be compared with the query at hand to compute a relevance score. Component (a) directly generates a matching score of a candidate document for a query. The full architecture benefits from the complementary potential of document-query matching and the novel document transformation approach based on summarization along PM facets. Evaluations using NIST’s TREC-PM track datasets (2017–2019) show that our model achieves state-of-the-art performance. To foster reproducibility, our code is made available here: https://github.com/bionlproc/text-summ-for-doc-retrieval.</abstract>
      <url hash="bf41aa03">2020.findings-emnlp.304</url>
      <doi>10.18653/v1/2020.findings-emnlp.304</doi>
    </paper>
    <paper id="305">
      <title><fixed-case>O</fixed-case>n the <fixed-case>I</fixed-case>mportance of <fixed-case>A</fixed-case>daptive <fixed-case>D</fixed-case>ata <fixed-case>C</fixed-case>ollection for <fixed-case>E</fixed-case>xtremely <fixed-case>I</fixed-case>mbalanced <fixed-case>P</fixed-case>airwise <fixed-case>T</fixed-case>asks</title>
      <author><first>Stephen</first><last>Mussmann</last></author>
      <author><first>Robin</first><last>Jia</last></author>
      <author><first>Percy</first><last>Liang</last></author>
      <pages>3400–3413</pages>
      <abstract>Many pairwise classification tasks, such as paraphrase detection and open-domain question answering, naturally have extreme label imbalance (e.g., 99.99% of examples are negatives). In contrast, many recent datasets heuristically choose examples to ensure label balance. We show that these heuristics lead to trained models that generalize poorly: State-of-the art models trained on QQP and WikiQA each have only 2.4% average precision when evaluated on realistically imbalanced test data. We instead collect training data with active learning, using a BERT-based embedding model to efficiently retrieve uncertain points from a very large pool of unlabeled utterance pairs. By creating balanced training data with more informative negative examples, active learning greatly improves average precision to 32.5% on QQP and 20.1% on WikiQA.</abstract>
      <url hash="7ef2e2ed">2020.findings-emnlp.305</url>
      <doi>10.18653/v1/2020.findings-emnlp.305</doi>
    </paper>
    <paper id="306">
      <title>A Dual-Attention Network for Joint Named Entity Recognition and Sentence Classification of Adverse Drug Events</title>
      <author><first>Susmitha</first><last>Wunnava</last></author>
      <author><first>Xiao</first><last>Qin</last></author>
      <author><first>Tabassum</first><last>Kakar</last></author>
      <author><first>Xiangnan</first><last>Kong</last></author>
      <author><first>Elke</first><last>Rundensteiner</last></author>
      <pages>3414–3423</pages>
      <abstract>An adverse drug event (ADE) is an injury resulting from medical intervention related to a drug. Automatic ADE detection from text is either fine-grained (ADE entity recognition) or coarse-grained (ADE assertive sentence classification), with limited efforts leveraging inter-dependencies among the two granularities. We instead propose a multi-grained joint deep network to concurrently learn the ADE entity recognition and ADE sentence classification tasks. Our joint approach takes advantage of their symbiotic relationship, with a transfer of knowledge between the two levels of granularity. Our dual-attention mechanism constructs multiple distinct representations of a sentence that capture both task-specific and semantic information in the sentence, providing stronger emphasis on the key elements essential for sentence classification. Our model improves state-of- art F1-score for both tasks: (i) entity recognition of ADE words (12.5% increase) and (ii) ADE sentence classification (13.6% increase) on MADE 1.0 benchmark of EHR notes.</abstract>
      <url hash="ac886f04">2020.findings-emnlp.306</url>
      <doi>10.18653/v1/2020.findings-emnlp.306</doi>
    </paper>
    <paper id="307">
      <title><fixed-case>BERT</fixed-case>-k<fixed-case>NN</fixed-case>: Adding a k<fixed-case>NN</fixed-case> Search Component to Pretrained Language Models for Better <fixed-case>QA</fixed-case></title>
      <author><first>Nora</first><last>Kassner</last></author>
      <author><first>Hinrich</first><last>Schütze</last></author>
      <pages>3424–3430</pages>
      <abstract>Khandelwal et al. (2020) use a k-nearest-neighbor (kNN) component to improve language model performance. We show that this idea is beneficial for open-domain question answering (QA). To improve the recall of facts encountered during training, we combine BERT (Devlin et al., 2019) with a traditional information retrieval step (IR) and a kNN search over a large datastore of an embedded text collection. Our contributions are as follows: i) BERT-kNN outperforms BERT on cloze-style QA by large margins without any further training. ii) We show that BERT often identifies the correct response category (e.g., US city), but only kNN recovers the factually correct answer (e.g.,“Miami”). iii) Compared to BERT, BERT-kNN excels for rare facts. iv) BERT-kNN can easily handle facts not covered by BERT’s training set, e.g., recent events.</abstract>
      <url hash="d7d3ea30">2020.findings-emnlp.307</url>
      <doi>10.18653/v1/2020.findings-emnlp.307</doi>
    </paper>
    <paper id="308">
      <title>Identifying Spurious Correlations for Robust Text Classification</title>
      <author><first>Zhao</first><last>Wang</last></author>
      <author><first>Aron</first><last>Culotta</last></author>
      <pages>3431–3440</pages>
      <abstract>The predictions of text classifiers are often driven by spurious correlations – e.g., the term “Spielberg” correlates with positively reviewed movies, even though the term itself does not semantically convey a positive sentiment. In this paper, we propose a method to distinguish spurious and genuine correlations in text classification. We treat this as a supervised classification problem, using features derived from treatment effect estimators to distinguish spurious correlations from “genuine” ones. Due to the generic nature of these features and their small dimensionality, we find that the approach works well even with limited training examples, and that it is possible to transport the word classifier to new domains. Experiments on four datasets (sentiment classification and toxicity detection) suggest that using this approach to inform feature selection also leads to more robust classification, as measured by improved worst-case accuracy on the samples affected by spurious correlations.</abstract>
      <url hash="598a9f9a">2020.findings-emnlp.308</url>
      <doi>10.18653/v1/2020.findings-emnlp.308</doi>
    </paper>
    <paper id="309">
      <title><fixed-case>H</fixed-case>o<fixed-case>V</fixed-case>er: A Dataset for Many-Hop Fact Extraction And Claim Verification</title>
      <author><first>Yichen</first><last>Jiang</last></author>
      <author><first>Shikha</first><last>Bordia</last></author>
      <author><first>Zheng</first><last>Zhong</last></author>
      <author><first>Charles</first><last>Dognin</last></author>
      <author><first>Maneesh</first><last>Singh</last></author>
      <author><first>Mohit</first><last>Bansal</last></author>
      <pages>3441–3460</pages>
      <abstract>We introduce HoVer (HOppy VERification), a dataset for many-hop evidence extraction and fact verification. It challenges models to extract facts from several Wikipedia articles that are relevant to a claim and classify whether the claim is supported or not-supported by the facts. In HoVer, the claims require evidence to be extracted from as many as four English Wikipedia articles and embody reasoning graphs of diverse shapes. Moreover, most of the 3/4-hop claims are written in multiple sentences, which adds to the complexity of understanding long-range dependency relations such as coreference. We show that the performance of an existing state-of-the-art semantic-matching model degrades significantly on our dataset as the number of reasoning hops increases, hence demonstrating the necessity of many-hop reasoning to achieve strong results. We hope that the introduction of this challenging dataset and the accompanying evaluation task will encourage research in many-hop fact retrieval and information verification.</abstract>
      <url hash="ebcc0597">2020.findings-emnlp.309</url>
      <doi>10.18653/v1/2020.findings-emnlp.309</doi>
    </paper>
    <paper id="310">
      <title>Continual Learning for Natural Language Generation in Task-oriented Dialog Systems</title>
      <author><first>Fei</first><last>Mi</last></author>
      <author><first>Liangwei</first><last>Chen</last></author>
      <author><first>Mengjie</first><last>Zhao</last></author>
      <author><first>Minlie</first><last>Huang</last></author>
      <author><first>Boi</first><last>Faltings</last></author>
      <pages>3461–3474</pages>
      <abstract>Natural language generation (NLG) is an essential component of task-oriented dialog systems. Despite the recent success of neural approaches for NLG, they are typically developed in an offline manner for particular domains. To better fit real-life applications where new data come in a stream, we study NLG in a “continual learning” setting to expand its knowledge to new domains or functionalities incrementally. The major challenge towards this goal is catastrophic forgetting, meaning that a continually trained model tends to forget the knowledge it has learned before. To this end, we propose a method called ARPER (Adaptively Regularized Prioritized Exemplar Replay) by replaying prioritized historical exemplars, together with an adaptive regularization technique based on Elastic Weight Consolidation. Extensive experiments to continually learn new domains and intents are conducted on MultiWoZ-2.0 to benchmark ARPER with a wide range of techniques. Empirical results demonstrate that ARPER significantly outperforms other methods by effectively mitigating the detrimental catastrophic forgetting issue.</abstract>
      <url hash="ec499c29">2020.findings-emnlp.310</url>
      <doi>10.18653/v1/2020.findings-emnlp.310</doi>
    </paper>
    <paper id="311">
      <title><fixed-case>UNQOVER</fixed-case>ing Stereotyping Biases via Underspecified Questions</title>
      <author><first>Tao</first><last>Li</last></author>
      <author><first>Daniel</first><last>Khashabi</last></author>
      <author><first>Tushar</first><last>Khot</last></author>
      <author><first>Ashish</first><last>Sabharwal</last></author>
      <author><first>Vivek</first><last>Srikumar</last></author>
      <pages>3475–3489</pages>
      <abstract>While language embeddings have been shown to have stereotyping biases, how these biases affect downstream question answering (QA) models remains unexplored. We present UNQOVER, a general framework to probe and quantify biases through underspecified questions. We show that a naive use of model scores can lead to incorrect bias estimates due to two forms of reasoning errors: positional dependence and question independence. We design a formalism that isolates the aforementioned errors. As case studies, we use this metric to analyze four important classes of stereotypes: gender, nationality, ethnicity, and religion. We probe five transformer-based QA models trained on two QA datasets, along with their underlying language models. Our broad study reveals that (1) all these models, with and without fine-tuning, have notable stereotyping biases in these classes; (2) larger models often have higher bias; and (3) the effect of fine-tuning on bias varies strongly with the dataset and the model size.</abstract>
      <url hash="9d35b497">2020.findings-emnlp.311</url>
      <doi>10.18653/v1/2020.findings-emnlp.311</doi>
    </paper>
    <paper id="312">
      <title>A Semantics-based Approach to Disclosure Classification in User-Generated Online Content</title>
      <author><first>Chandan</first><last>Akiti</last></author>
      <author><first>Anna</first><last>Squicciarini</last></author>
      <author><first>Sarah</first><last>Rajtmajer</last></author>
      <pages>3490–3499</pages>
      <abstract>As users engage in public discourse, the rate of voluntarily disclosed personal information has seen a steep increase. So-called self-disclosure can result in a number of privacy concerns. Users are often unaware of the sheer amount of personal information they share across online forums, commentaries, and social networks, as well as the power of modern AI to synthesize and gain insights from this data. This paper presents an approach to detect emotional and informational self-disclosure in natural language. We hypothesize that identifying frame semantics can meaningfully support this task. Specifically, we use Semantic Role Labeling to identify the lexical units and their semantic roles that signal self-disclosure. Experimental results on Reddit data show the performance gain of our method when compared to standard text classification methods based on BiLSTM, and BERT. In addition to improved performance, our approach provides insights into the drivers of disclosure behaviors.</abstract>
      <url hash="d18c780d">2020.findings-emnlp.312</url>
      <attachment type="OptionalSupplementaryMaterial" hash="39679ab2">2020.findings-emnlp.312.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.312</doi>
    </paper>
    <paper id="313">
      <title>Mining Knowledge for Natural Language Inference from <fixed-case>W</fixed-case>ikipedia Categories</title>
      <author><first>Mingda</first><last>Chen</last></author>
      <author><first>Zewei</first><last>Chu</last></author>
      <author><first>Karl</first><last>Stratos</last></author>
      <author><first>Kevin</first><last>Gimpel</last></author>
      <pages>3500–3511</pages>
      <abstract>Accurate lexical entailment (LE) and natural language inference (NLI) often require large quantities of costly annotations. To alleviate the need for labeled data, we introduce WikiNLI: a resource for improving model performance on NLI and LE tasks. It contains 428,899 pairs of phrases constructed from naturally annotated category hierarchies in Wikipedia. We show that we can improve strong baselines such as BERT and RoBERTa by pretraining them on WikiNLI and transferring the models on downstream tasks. We conduct systematic comparisons with phrases extracted from other knowledge bases such as WordNet and Wikidata to find that pretraining on WikiNLI gives the best performance. In addition, we construct WikiNLI in other languages, and show that pretraining on them improves performance on NLI tasks of corresponding languages.</abstract>
      <url hash="9507eaad">2020.findings-emnlp.313</url>
      <attachment type="OptionalSupplementaryMaterial" hash="040de2ec">2020.findings-emnlp.313.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.313</doi>
    </paper>
    <paper id="314">
      <title><fixed-case>OCNLI</fixed-case>: <fixed-case>O</fixed-case>riginal <fixed-case>C</fixed-case>hinese <fixed-case>N</fixed-case>atural <fixed-case>L</fixed-case>anguage <fixed-case>I</fixed-case>nference</title>
      <author><first>Hai</first><last>Hu</last></author>
      <author><first>Kyle</first><last>Richardson</last></author>
      <author><first>Liang</first><last>Xu</last></author>
      <author><first>Lu</first><last>Li</last></author>
      <author><first>Sandra</first><last>Kübler</last></author>
      <author><first>Lawrence</first><last>Moss</last></author>
      <pages>3512–3526</pages>
      <abstract>Despite the tremendous recent progress on natural language inference (NLI), driven largely by large-scale investment in new datasets (e.g.,SNLI, MNLI) and advances in modeling, most progress has been limited to English due to a lack of reliable datasets for most of the world’s languages. In this paper, we present the first large-scale NLI dataset (consisting of ~56,000 annotated sentence pairs) for Chinese called the Original Chinese Natural Language Inference dataset (OCNLI). Unlike recent attempts at extending NLI to other languages, our dataset does not rely on any automatic translation or non-expert annotation. Instead, we elicit annotations from native speakers specializing in linguistics. We follow closely the annotation protocol used for MNLI, but create new strategies for eliciting diverse hypotheses. We establish several baseline results on our dataset using state-of-the-art pre-trained models for Chinese, and find even the best performing models to be far outpaced by human performance (~12% absolute performance gap), making it a challenging new resource that we hope will help to accelerate progress in Chinese NLU. To the best of our knowledge, this is the first human-elicited MNLI-style corpus for a non-English language.</abstract>
      <url hash="ef440fe2">2020.findings-emnlp.314</url>
      <doi>10.18653/v1/2020.findings-emnlp.314</doi>
    </paper>
    <paper id="315">
      <title>Margin-aware Unsupervised Domain Adaptation for Cross-lingual Text Labeling</title>
      <author><first>Dejiao</first><last>Zhang</last></author>
      <author><first>Ramesh</first><last>Nallapati</last></author>
      <author><first>Henghui</first><last>Zhu</last></author>
      <author><first>Feng</first><last>Nan</last></author>
      <author><first>Cicero</first><last>Nogueira dos Santos</last></author>
      <author><first>Kathleen</first><last>McKeown</last></author>
      <author><first>Bing</first><last>Xiang</last></author>
      <pages>3527–3536</pages>
      <abstract>Unsupervised domain adaptation addresses the problem of leveraging labeled data in a source domain to learn a well-performing model in a target domain where labels are unavailable. In this paper, we improve upon a recent theoretical work (Zhang et al., 2019b) and adopt the Margin Disparity Discrepancy (MDD) unsupervised domain adaptation algorithm to solve the cross-lingual text labeling problems. Experiments on cross-lingual document classification and NER demonstrate the proposed domain adaptation approach advances the state-of-the-art results by a large margin. Specifically, we improve MDD by efficiently optimizing the margin loss on the source domain via Virtual Adversarial Training (VAT). This bridges the gap between theory and the loss function used in the original work Zhang et al.(2019b), and thereby significantly boosts the performance. Our numerical results also indicate that VAT can remarkably improve the generalization performance of both domains for various domain adaptation approaches.</abstract>
      <url hash="5cabb619">2020.findings-emnlp.315</url>
      <doi>10.18653/v1/2020.findings-emnlp.315</doi>
    </paper>
    <paper id="316">
      <title>Rethinking Supervised Learning and Reinforcement Learning in Task-Oriented Dialogue Systems</title>
      <author><first>Ziming</first><last>Li</last></author>
      <author><first>Julia</first><last>Kiseleva</last></author>
      <author><first>Maarten</first><last>de Rijke</last></author>
      <pages>3537–3546</pages>
      <abstract>Dialogue policy learning for task-oriented dialogue systems has enjoyed great progress recently mostly through employing reinforcement learning methods. However, these approaches have become very sophisticated. It is time to re-evaluate it. Are we really making progress developing dialogue agents only based on reinforcement learning? We demonstrate how (1) traditional supervised learning together with (2) a simulator-free adversarial learning method can be used to achieve performance comparable to state-of-the-art reinforcement learning-based methods. First, we introduce a simple dialogue action decoder to predict the appropriate actions. Then, the traditional multi-label classification solution for dialogue policy learning is extended by adding dense layers to improve the dialogue agent performance. Finally, we employ the Gumbel-Softmax estimator to alternatively train the dialogue agent and the dialogue reward model without using reinforcement learning. Based on our extensive experimentation, we can conclude the proposed methods can achieve more stable and higher performance with fewer efforts, such as the domain knowledge required to design a user simulator and the intractable parameter tuning in reinforcement learning. Our main goal is not to beat RL with supervised learning, but to demonstrate the value of rethinking the role of reinforcement learning and supervised learning in optimizing task-oriented dialogue systems.</abstract>
      <url hash="b3154335">2020.findings-emnlp.316</url>
      <doi>10.18653/v1/2020.findings-emnlp.316</doi>
    </paper>
    <paper id="317">
      <title>What do we expect from Multiple-choice <fixed-case>QA</fixed-case> Systems?</title>
      <author><first>Krunal</first><last>Shah</last></author>
      <author><first>Nitish</first><last>Gupta</last></author>
      <author><first>Dan</first><last>Roth</last></author>
      <pages>3547–3553</pages>
      <abstract>The recent success of machine learning systems on various QA datasets could be interpreted as a significant improvement in models’ language understanding abilities. However, using various perturbations, multiple recent works have shown that good performance on a dataset might not indicate performance that correlates well with human’s expectations from models that “understand” language. In this work we consider a top performing model on several Multiple Choice Question Answering (MCQA) datasets, and evaluate it against a set of expectations one might have from such a model, using a series of zero-information perturbations of the model’s inputs. Our results show that the model clearly falls short of our expectations, and motivates a modified training approach that forces the model to better attend to the inputs. We show that the new training paradigm leads to a model that performs on par with the original model while better satisfying our expectations.</abstract>
      <url hash="d2bcbca0">2020.findings-emnlp.317</url>
      <doi>10.18653/v1/2020.findings-emnlp.317</doi>
    </paper>
    <paper id="318">
      <title>Resource-Enhanced Neural Model for Event Argument Extraction</title>
      <author><first>Jie</first><last>Ma</last></author>
      <author><first>Shuai</first><last>Wang</last></author>
      <author><first>Rishita</first><last>Anubhai</last></author>
      <author><first>Miguel</first><last>Ballesteros</last></author>
      <author><first>Yaser</first><last>Al-Onaizan</last></author>
      <pages>3554–3559</pages>
      <abstract>Event argument extraction (EAE) aims to identify the arguments of an event and classify the roles that those arguments play. Despite great efforts made in prior work, there remain many challenges: (1) Data scarcity. (2) Capturing the long-range dependency, specifically, the connection between an event trigger and a distant event argument. (3) Integrating event trigger information into candidate argument representation. For (1), we explore using unlabeled data. For (2), we use Transformer that uses dependency parses to guide the attention mechanism. For (3), we propose a trigger-aware sequence encoder with several types of trigger-dependent sequence representations. We also support argument extraction either from text annotated with gold entities or from plain text. Experiments on the English ACE 2005 benchmark show that our approach achieves a new state-of-the-art.</abstract>
      <url hash="0d2fc419">2020.findings-emnlp.318</url>
      <doi>10.18653/v1/2020.findings-emnlp.318</doi>
    </paper>
    <paper id="319">
      <title>Improving Target-side Lexical Transfer in Multilingual Neural Machine Translation</title>
      <author><first>Luyu</first><last>Gao</last></author>
      <author><first>Xinyi</first><last>Wang</last></author>
      <author><first>Graham</first><last>Neubig</last></author>
      <pages>3560–3566</pages>
      <abstract>To improve the performance of Neural Machine Translation (NMT) for low-resource languages (LRL), one effective strategy is to leverage parallel data from a related high-resource language (HRL). However, multilingual data has been found more beneficial for NMT models that translate from the LRL to a target language than the ones that translate into the LRLs. In this paper, we aim to improve the effectiveness of multilingual transfer for NMT models that translate into the LRL, by designing a better decoder word embedding. Extending upon a general-purpose multilingual encoding method Soft Decoupled Encoding (Wang et al., 2019), we propose DecSDE, an efficient character n-gram based embedding specifically designed for the NMT decoder. Our experiments show that DecSDE leads to consistent gains of up to 1.8 BLEU on translation from English to four different languages.</abstract>
      <url hash="7cf97855">2020.findings-emnlp.319</url>
      <doi>10.18653/v1/2020.findings-emnlp.319</doi>
    </paper>
    <paper id="320">
      <title>Accurate polyglot semantic parsing with <fixed-case>DAG</fixed-case> grammars</title>
      <author><first>Federico</first><last>Fancellu</last></author>
      <author><first>Ákos</first><last>Kádár</last></author>
      <author><first>Ran</first><last>Zhang</last></author>
      <author><first>Afsaneh</first><last>Fazly</last></author>
      <pages>3567–3580</pages>
      <abstract>Semantic parses are directed acyclic graphs (DAGs), but in practice most parsers treat them as strings or trees, mainly because models that predict graphs are far less understood. This simplification, however, comes at a cost: there is no guarantee that the output is a well-formed graph. A recent work by Fancellu et al. (2019) addressed this problem by proposing a graph-aware sequence model that utilizes a DAG grammar to guide graph generation. We significantly improve upon this work, by proposing a simpler architecture as well as more efficient training and inference algorithms that can always guarantee the well-formedness of the generated graphs. Importantly, unlike Fancellu et al., our model does not require language-specific features, and hence can harness the inherent ability of DAG-grammar parsing in multilingual settings. We perform monolingual as well as multilingual experiments on the Parallel Meaning Bank (Abzianidze et al., 2017). Our parser outperforms previous graph-aware models by a large margin, and closes the performance gap between string-based and DAG-grammar parsing.</abstract>
      <url hash="07c9de0b">2020.findings-emnlp.320</url>
      <doi>10.18653/v1/2020.findings-emnlp.320</doi>
    </paper>
    <paper id="321">
      <title>Approximation of Response Knowledge Retrieval in Knowledge-grounded Dialogue Generation</title>
      <author><first>Wen</first><last>Zheng</last></author>
      <author><first>Natasa</first><last>Milic-Frayling</last></author>
      <author><first>Ke</first><last>Zhou</last></author>
      <pages>3581–3591</pages>
      <abstract>This paper is concerned with improving dialogue generation models through injection of knowledge, e.g., content relevant to the post that can increase the quality of responses. Past research extends the training of the generative models by incorporating statistical properties of posts, responses and related knowledge, without explicitly assessing the knowledge quality. In our work, we demonstrate the importance of knowledge relevance and adopt a two-phase approach. We first apply a novel method, Transformer &amp; Post based Posterior Approximation (TPPA) to select knowledge, and then use the Transformer with Expanded Decoder (TED) model to generate responses from both the post and the knowledge. TPPA method processes posts, post related knowledge, and response related knowledge at both word and sentence level. Our experiments with the TED generative model demonstrate the effectiveness of TPPA as it outperforms a set of strong baseline models. Our TPPA method is extendable and supports further optimization of knowledge retrieval and injection.</abstract>
      <url hash="e3316766">2020.findings-emnlp.321</url>
      <doi>10.18653/v1/2020.findings-emnlp.321</doi>
    </paper>
    <paper id="322">
      <title>Evaluating Factuality in Generation with Dependency-level Entailment</title>
      <author><first>Tanya</first><last>Goyal</last></author>
      <author><first>Greg</first><last>Durrett</last></author>
      <pages>3592–3603</pages>
      <abstract>Despite significant progress in text generation models, a serious limitation is their tendency to produce text that is factually inconsistent with information in the input. Recent work has studied whether textual entailment systems can be used to identify factual errors; however, these sentence-level entailment models are trained to solve a different problem than generation filtering and they do not localize which part of a generation is non-factual. In this paper, we propose a new formulation of entailment that decomposes it at the level of dependency arcs. Rather than focusing on aggregate decisions, we instead ask whether the semantic relationship manifested by individual dependency arcs in the generated output is supported by the input. Human judgments on this task are difficult to obtain; we therefore propose a method to automatically create data based on existing entailment or paraphrase corpora. Experiments show that our dependency arc entailment model trained on this data can identify factual inconsistencies in paraphrasing and summarization better than sentence-level methods or those based on question generation, while additionally localizing the erroneous parts of the generation.</abstract>
      <url hash="0295c618">2020.findings-emnlp.322</url>
      <doi>10.18653/v1/2020.findings-emnlp.322</doi>
    </paper>
    <paper id="323">
      <title>Cross-Lingual Text Classification with Minimal Resources by Transferring a Sparse Teacher</title>
      <author><first>Giannis</first><last>Karamanolakis</last></author>
      <author><first>Daniel</first><last>Hsu</last></author>
      <author><first>Luis</first><last>Gravano</last></author>
      <pages>3604–3622</pages>
      <abstract>Cross-lingual text classification alleviates the need for manually labeled documents in a target language by leveraging labeled documents from other languages. Existing approaches for transferring supervision across languages require expensive cross-lingual resources, such as parallel corpora, while less expensive cross-lingual representation learning approaches train classifiers without target labeled documents. In this work, we propose a cross-lingual teacher-student method, CLTS, that generates “weak” supervision in the target language using minimal cross-lingual resources, in the form of a small number of word translations. Given a limited translation budget, CLTS extracts and transfers only the most important task-specific seed words across languages and initializes a teacher classifier based on the translated seed words. Then, CLTS iteratively trains a more powerful student that also exploits the context of the seed words in unlabeled target documents and outperforms the teacher. CLTS is simple and surprisingly effective in 18 diverse languages: by transferring just 20 seed words, even a bag-of-words logistic regression student outperforms state-of-the-art cross-lingual methods (e.g., based on multilingual BERT). Moreover, CLTS can accommodate any type of student classifier: leveraging a monolingual BERT student leads to further improvements and outperforms even more expensive approaches by up to 12% in accuracy. Finally, CLTS addresses emerging tasks in low-resource languages using just a small number of word translations.</abstract>
      <url hash="70ae910c">2020.findings-emnlp.323</url>
      <doi>10.18653/v1/2020.findings-emnlp.323</doi>
    </paper>
    <paper id="324">
      <title>A Multi-Persona Chatbot for Hotline Counselor Training</title>
      <author><first>Orianna</first><last>Demasi</last></author>
      <author><first>Yu</first><last>Li</last></author>
      <author><first>Zhou</first><last>Yu</last></author>
      <pages>3623–3636</pages>
      <abstract>Suicide prevention hotline counselors aid individuals during difficult times through millions of calls and chats. A chatbot cannot safely replace a counselor, but we explore whether a chatbot can be developed to help train human counselors. Such a system needs to simulate intimate situations across multiple practice sessions. Open-domain dialogue systems frequently suffer from generic responses that do not characterize personal stories, so we look to infuse conversations with persona information by mimicking prototype conversations. Towards building a “Crisisbot” hotline visitor simulation, we propose a counseling strategy annotation scheme and a multi-task framework that leverages these counselor strategies to retrieve similar examples, generate diverse sub-utterances, and interleave prototype and generated sub-utterances into complex responses. We evaluate this framework with crowdworkers and experienced hotline counselors. The framework considerably increases response diversity and specificity, with limited impact to coherence. Our results also show a considerable discrepancy between crowdworker and counselor judgements, which emphasizes the importance of including target populations in system development and evaluation.</abstract>
      <url hash="9a43909e">2020.findings-emnlp.324</url>
      <doi>10.18653/v1/2020.findings-emnlp.324</doi>
    </paper>
    <paper id="325">
      <title>Narrative Text Generation with a Latent Discrete Plan</title>
      <author><first>Harsh</first><last>Jhamtani</last></author>
      <author><first>Taylor</first><last>Berg-Kirkpatrick</last></author>
      <pages>3637–3650</pages>
      <abstract>Past work on story generation has demonstrated the usefulness of conditioning on a generation plan to generate coherent stories. However, these approaches have used heuristics or off-the-shelf models to first tag training stories with the desired type of plan, and then train generation models in a supervised fashion. In this paper, we propose a deep latent variable model that first samples a sequence of anchor words, one per sentence in the story, as part of its generative process. During training, our model treats the sequence of anchor words as a latent variable and attempts to induce anchoring sequences that help guide generation in an unsupervised fashion. We conduct experiments with several types of sentence decoder distributions – left-to-right and non-monotonic, with different degrees of restriction. Further, since we use amortized variational inference to train our model, we introduce two corresponding types of inference network for predicting the posterior on anchor words. We conduct human evaluations which demonstrate that the stories produced by our model are rated better in comparison with baselines which do not consider story plans, and are similar or better in quality relative to baselines which use external supervision for plans. Additionally, the proposed model gets favorable scores when evaluated on perplexity, diversity, and control of story via discrete plan</abstract>
      <url hash="05ab344a">2020.findings-emnlp.325</url>
      <doi>10.18653/v1/2020.findings-emnlp.325</doi>
    </paper>
    <paper id="326">
      <title>Graph Transformer Networks with Syntactic and Semantic Structures for Event Argument Extraction</title>
      <author><first>Amir</first><last>Pouran Ben Veyseh</last></author>
      <author><first>Tuan Ngo</first><last>Nguyen</last></author>
      <author><first>Thien Huu</first><last>Nguyen</last></author>
      <pages>3651–3661</pages>
      <abstract>The goal of Event Argument Extraction (EAE) is to find the role of each entity mention for a given event trigger word. It has been shown in the previous works that the syntactic structures of the sentences are helpful for the deep learning models for EAE. However, a major problem in such prior works is that they fail to exploit the semantic structures of the sentences to induce effective representations for EAE. Consequently, in this work, we propose a novel model for EAE that exploits both syntactic and semantic structures of the sentences with the Graph Transformer Networks (GTNs) to learn more effective sentence structures for EAE. In addition, we introduce a novel inductive bias based on information bottleneck to improve generalization of the EAE models. Extensive experiments are performed to demonstrate the benefits of the proposed model, leading to state-of-the-art performance for EAE on standard datasets.</abstract>
      <url hash="8387e863">2020.findings-emnlp.326</url>
      <doi>10.18653/v1/2020.findings-emnlp.326</doi>
    </paper>
    <paper id="327">
      <title>The Box is in the Pen: Evaluating Commonsense Reasoning in Neural Machine Translation</title>
      <author><first>Jie</first><last>He</last></author>
      <author><first>Tao</first><last>Wang</last></author>
      <author><first>Deyi</first><last>Xiong</last></author>
      <author><first>Qun</first><last>Liu</last></author>
      <pages>3662–3672</pages>
      <abstract>Does neural machine translation yield translations that are congenial with common sense? In this paper, we present a test suite to evaluate the commonsense reasoning capability of neural machine translation. The test suite consists of three test sets, covering lexical and contextless/contextual syntactic ambiguity that requires commonsense knowledge to resolve. We manually create 1,200 triples, each of which contain a source sentence and two contrastive translations, involving 7 different common sense types. Language models pretrained on large-scale corpora, such as BERT, GPT-2, achieve a commonsense reasoning accuracy of lower than 72% on target translations of this test suite. We conduct extensive experiments on the test suite to evaluate commonsense reasoning in neural machine translation and investigate factors that have impact on this capability. Our experiments and analyses demonstrate that neural machine translation performs poorly on commonsense reasoning of the three ambiguity types in terms of both reasoning accuracy ( 6 60.1%) and reasoning consistency (6 31%). We will release our test suite as a machine translation commonsense reasoning testbed to promote future work in this direction.</abstract>
      <url hash="1e1a562d">2020.findings-emnlp.327</url>
      <doi>10.18653/v1/2020.findings-emnlp.327</doi>
    </paper>
    <paper id="328">
      <title>Using Visual Feature Space as a Pivot Across Languages</title>
      <author><first>Ziyan</first><last>Yang</last></author>
      <author><first>Leticia</first><last>Pinto-Alva</last></author>
      <author><first>Franck</first><last>Dernoncourt</last></author>
      <author><first>Vicente</first><last>Ordonez</last></author>
      <pages>3673–3678</pages>
      <abstract>Our work aims to leverage visual feature space to pass information across languages. We show that models trained to generate textual captions in more than one language conditioned on an input image can leverage their jointly trained feature space during inference to pivot across languages. We particularly demonstrate improved quality on a caption generated from an input image, by leveraging a caption in a second language. More importantly, we demonstrate that even without conditioning on any visual input, the model demonstrates to have learned implicitly to perform to some extent machine translation from one language to another in their shared visual feature space. We show results in German-English, and Japanese-English language pairs that pave the way for using the visual world to learn a common representation for language.</abstract>
      <url hash="221154ae">2020.findings-emnlp.328</url>
      <doi>10.18653/v1/2020.findings-emnlp.328</doi>
    </paper>
    <paper id="329">
      <title><fixed-case>CDE</fixed-case>val<fixed-case>S</fixed-case>umm: An Empirical Study of Cross-Dataset Evaluation for Neural Summarization Systems</title>
      <author><first>Yiran</first><last>Chen</last></author>
      <author><first>Pengfei</first><last>Liu</last></author>
      <author><first>Ming</first><last>Zhong</last></author>
      <author><first>Zi-Yi</first><last>Dou</last></author>
      <author><first>Danqing</first><last>Wang</last></author>
      <author><first>Xipeng</first><last>Qiu</last></author>
      <author><first>Xuanjing</first><last>Huang</last></author>
      <pages>3679–3691</pages>
      <abstract>Neural network-based models augmented with unsupervised pre-trained knowledge have achieved impressive performance on text summarization. However, most existing evaluation methods are limited to an in-domain setting, where summarizers are trained and evaluated on the same dataset. We argue that this approach can narrow our understanding of the generalization ability for different summarization systems. In this paper, we perform an in-depth analysis of characteristics of different datasets and investigate the performance of different summarization models under a cross-dataset setting, in which a summarizer trained on one corpus will be evaluated on a range of out-of-domain corpora. A comprehensive study of 11 representative summarization systems on 5 datasets from different domains reveals the effect of model architectures and generation ways (i.e. abstractive and extractive) on model generalization ability. Further, experimental results shed light on the limitations of existing summarizers. Brief introduction and supplementary code can be found in <url>https://github.com/zide05/CDEvalSumm</url>.</abstract>
      <url hash="7673f9eb">2020.findings-emnlp.329</url>
      <attachment type="OptionalSupplementaryMaterial" hash="51754a33">2020.findings-emnlp.329.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.329</doi>
    </paper>
    <paper id="330">
      <title><fixed-case>A</fixed-case>ttending to <fixed-case>L</fixed-case>ong-<fixed-case>D</fixed-case>istance <fixed-case>D</fixed-case>ocument <fixed-case>C</fixed-case>ontext for <fixed-case>S</fixed-case>equence <fixed-case>L</fixed-case>abeling</title>
      <author><first>Matthew</first><last>Jörke</last></author>
      <author><first>Jon</first><last>Gillick</last></author>
      <author><first>Matthew</first><last>Sims</last></author>
      <author><first>David</first><last>Bamman</last></author>
      <pages>3692–3704</pages>
      <abstract>We present in this work a method for incorporating global context in long documents when making local decisions in sequence labeling problems like NER. Inspired by work in featurized log-linear models (Chieu and Ng, 2002; Sutton and McCallum, 2004), our model learns to attend to multiple mentions of the same word type in generating a representation for each token in context, extending that work to learning representations that can be incorporated into modern neural models. Attending to broader context at test time provides complementary information to pretraining (Gururangan et al., 2020), yields strong gains over equivalently parameterized models lacking such context, and performs best at recognizing entities with high TF-IDF scores (i.e., those that are important within a document).</abstract>
      <url hash="d1fe9312">2020.findings-emnlp.330</url>
      <doi>10.18653/v1/2020.findings-emnlp.330</doi>
    </paper>
    <paper id="331">
      <title>Global Bootstrapping Neural Network for Entity Set Expansion</title>
      <author><first>Lingyong</first><last>Yan</last></author>
      <author><first>Xianpei</first><last>Han</last></author>
      <author><first>Ben</first><last>He</last></author>
      <author><first>Le</first><last>Sun</last></author>
      <pages>3705–3714</pages>
      <abstract>Bootstrapping for entity set expansion (ESE) has been studied for a long period, which expands new entities using only a few seed entities as supervision. Recent end-to-end bootstrapping approaches have shown their advantages in information capturing and bootstrapping process modeling. However, due to the sparse supervision problem, previous end-to-end methods often only leverage information from near neighborhoods (local semantics) rather than those propagated from the co-occurrence structure of the whole corpus (global semantics). To address this issue, this paper proposes Global Bootstrapping Network (GBN) with the “pre-training and fine-tuning” strategies for effective learning. Specifically, it contains a global-sighted encoder to capture and encode both local and global semantics into entity embedding, and an attention-guided decoder to sequentially expand new entities based on these embeddings. The experimental results show that the GBN learned by “pre-training and fine-tuning” strategies achieves state-of-the-art performance on two bootstrapping datasets.</abstract>
      <url hash="dd6de8f5">2020.findings-emnlp.331</url>
      <doi>10.18653/v1/2020.findings-emnlp.331</doi>
    </paper>
    <paper id="332">
      <title>Document Classification for <fixed-case>COVID</fixed-case>-19 Literature</title>
      <author><first>Bernal</first><last>Jimenez Gutierrez</last></author>
      <author><first>Jucheng</first><last>Zeng</last></author>
      <author><first>Dongdong</first><last>Zhang</last></author>
      <author><first>Ping</first><last>Zhang</last></author>
      <author><first>Yu</first><last>Su</last></author>
      <pages>3715–3722</pages>
      <abstract>The global pandemic has made it more important than ever to quickly and accurately retrieve relevant scientific literature for effective consumption by researchers in a wide range of fields. We provide an analysis of several multi-label document classification models on the LitCovid dataset, a growing collection of 23,000 research papers regarding the novel 2019 coronavirus. We find that pre-trained language models fine-tuned on this dataset outperform all other baselines and that BioBERT surpasses the others by a small margin with micro-F1 and accuracy scores of around 86% and 75% respectively on the test set. We evaluate the data efficiency and generalizability of these models as essential features of any system prepared to deal with an urgent situation like the current health crisis. We perform a data ablation study to determine how important article titles are for achieving reasonable performance on this dataset. Finally, we explore 50 errors made by the best performing models on LitCovid documents and find that they often (1) correlate certain labels too closely together and (2) fail to focus on discriminative sections of the articles; both of which are important issues to address in future work. Both data and code are available on GitHub.</abstract>
      <url hash="112f695b">2020.findings-emnlp.332</url>
      <doi>10.18653/v1/2020.findings-emnlp.332</doi>
    </paper>
    <paper id="333">
      <title>Adversarial Augmentation Policy Search for Domain and Cross-Lingual Generalization in Reading Comprehension</title>
      <author><first>Adyasha</first><last>Maharana</last></author>
      <author><first>Mohit</first><last>Bansal</last></author>
      <pages>3723–3738</pages>
      <abstract>Reading comprehension models often overfit to nuances of training datasets and fail at adversarial evaluation. Training with adversarially augmented dataset improves robustness against those adversarial attacks but hurts generalization of the models. In this work, we present several effective adversaries and automated data augmentation policy search methods with the goal of making reading comprehension models more robust to adversarial evaluation, but also improving generalization to the source domain as well as new domains and languages. We first propose three new methods for generating QA adversaries, that introduce multiple points of confusion within the context, show dependence on insertion location of the distractor, and reveal the compounding effect of mixing adversarial strategies with syntactic and semantic paraphrasing methods. Next, we find that augmenting the training datasets with uniformly sampled adversaries improves robustness to the adversarial attacks but leads to decline in performance on the original unaugmented dataset. We address this issue via RL and more efficient Bayesian policy search methods for automatically learning the best augmentation policy combinations of the transformation probability for each adversary in a large search space. Using these learned policies, we show that adversarial training can lead to significant improvements in in-domain, out-of-domain, and cross-lingual (German, Russian, Turkish) generalization.</abstract>
      <url hash="3785a2a9">2020.findings-emnlp.333</url>
      <doi>10.18653/v1/2020.findings-emnlp.333</doi>
    </paper>
    <paper id="334">
      <title>Denoising Multi-Source Weak Supervision for Neural Text Classification</title>
      <author><first>Wendi</first><last>Ren</last></author>
      <author><first>Yinghao</first><last>Li</last></author>
      <author><first>Hanting</first><last>Su</last></author>
      <author><first>David</first><last>Kartchner</last></author>
      <author><first>Cassie</first><last>Mitchell</last></author>
      <author><first>Chao</first><last>Zhang</last></author>
      <pages>3739–3754</pages>
      <abstract>We study the problem of learning neural text classifiers without using any labeled data, but only easy-to-provide rules as multiple weak supervision sources. This problem is challenging because rule-induced weak labels are often noisy and incomplete. To address these two challenges, we design a label denoiser, which estimates the source reliability using a conditional soft attention mechanism and then reduces label noise by aggregating rule-annotated weak labels. The denoised pseudo labels then supervise a neural classifier to predicts soft labels for unmatched samples, which address the rule coverage issue. We evaluate our model on five benchmarks for sentiment, topic, and relation classifications. The results show that our model outperforms state-of-the-art weakly-supervised and semi-supervised methods consistently, and achieves comparable performance with fully-supervised methods even without any labeled data. Our code can be found at https://github.com/weakrules/Denoise-multi-weak-sources.</abstract>
      <url hash="04e407a3">2020.findings-emnlp.334</url>
      <doi>10.18653/v1/2020.findings-emnlp.334</doi>
    </paper>
    <paper id="335">
      <title>Dr. Summarize: Global Summarization of Medical Dialogue by Exploiting Local Structures.</title>
      <author><first>Anirudh</first><last>Joshi</last></author>
      <author><first>Namit</first><last>Katariya</last></author>
      <author><first>Xavier</first><last>Amatriain</last></author>
      <author><first>Anitha</first><last>Kannan</last></author>
      <pages>3755–3763</pages>
      <abstract>Understanding a medical conversation between a patient and a physician poses unique natural language understanding challenge since it combines elements of standard open-ended conversation with very domain-specific elements that require expertise and medical knowledge. Summarization of medical conversations is a particularly important aspect of medical conversation understanding since it addresses a very real need in medical practice: capturing the most important aspects of a medical encounter so that they can be used for medical decision making and subsequent follow ups. In this paper we present a novel approach to medical conversation summarization that leverages the unique and independent local structures created when gathering a patient’s medical history. Our approach is a variation of the pointer generator network where we introduce a penalty on the generator distribution, and we explicitly model negations. The model also captures important properties of medical conversations such as medical knowledge coming from standardized medical ontologies better than when those concepts are introduced explicitly. Through evaluation by doctors, we show that our approach is preferred on twice the number of summaries to the baseline pointer generator model and captures most or all of the information in 80% of the conversations making it a realistic alternative to costly manual summarization by medical experts.</abstract>
      <url hash="f0d6e9c2">2020.findings-emnlp.335</url>
      <attachment type="OptionalSupplementaryMaterial" hash="0547ea33">2020.findings-emnlp.335.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.335</doi>
    </paper>
    <paper id="336">
      <title>Generating <fixed-case>A</fixed-case>ccurate <fixed-case>E</fixed-case>lectronic <fixed-case>H</fixed-case>ealth <fixed-case>A</fixed-case>ssessment from <fixed-case>M</fixed-case>edical <fixed-case>G</fixed-case>raph</title>
      <author><first>Zhichao</first><last>Yang</last></author>
      <author><first>Hong</first><last>Yu</last></author>
      <pages>3764–3773</pages>
      <abstract>One of the fundamental goals of artificial intelligence is to build computer-based expert systems. Inferring clinical diagnoses to generate a clinical assessment during a patient encounter is a crucial step towards building a medical diagnostic system. Previous works were mainly based on either medical domain-specific knowledge, or patients’ prior diagnoses and clinical encounters. In this paper, we propose a novel model for automated clinical assessment generation (MCAG). MCAG is built on an innovative graph neural network, where rich clinical knowledge is incorporated into an end-to-end corpus-learning system. Our evaluation results against physician generated gold standard show that MCAG significantly improves the BLEU and rouge score compared with competitive baseline models. Further, physicians’ evaluation showed that MCAG could generate high-quality assessments.</abstract>
      <url hash="7e2f61af">2020.findings-emnlp.336</url>
      <attachment type="OptionalSupplementaryMaterial" hash="6a570ffa">2020.findings-emnlp.336.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.336</doi>
    </paper>
    <paper id="337">
      <title>Do Models of Mental Health Based on Social Media Data Generalize?</title>
      <author><first>Keith</first><last>Harrigian</last></author>
      <author><first>Carlos</first><last>Aguirre</last></author>
      <author><first>Mark</first><last>Dredze</last></author>
      <pages>3774–3788</pages>
      <abstract>Proxy-based methods for annotating mental health status in social media have grown popular in computational research due to their ability to gather large training samples. However, an emerging body of literature has raised new concerns regarding the validity of these types of methods for use in clinical applications. To further understand the robustness of distantly supervised mental health models, we explore the generalization ability of machine learning classifiers trained to detect depression in individuals across multiple social media platforms. Our experiments not only reveal that substantial loss occurs when transferring between platforms, but also that there exist several unreliable confounding factors that may enable researchers to overestimate classification performance. Based on these results, we enumerate recommendations for future mental health dataset construction.</abstract>
      <url hash="32c96af5">2020.findings-emnlp.337</url>
      <doi>10.18653/v1/2020.findings-emnlp.337</doi>
    </paper>
    <paper id="338">
      <title>Context Analysis for Pre-trained Masked Language Models</title>
      <author><first>Yi-An</first><last>Lai</last></author>
      <author><first>Garima</first><last>Lalwani</last></author>
      <author><first>Yi</first><last>Zhang</last></author>
      <pages>3789–3804</pages>
      <abstract>Pre-trained language models that learn contextualized word representations from a large un-annotated corpus have become a standard component for many state-of-the-art NLP systems. Despite their successful applications in various downstream NLP tasks, the extent of contextual impact on the word representation has not been explored. In this paper, we present a detailed analysis of contextual impact in Transformer- and BiLSTM-based masked language models. We follow two different approaches to evaluate the impact of context: a masking based approach that is architecture agnostic, and a gradient based approach that requires back-propagation through networks. The findings suggest significant differences on the contextual impact between the two model architectures. Through further breakdown of analysis by syntactic categories, we find the contextual impact in Transformer-based MLM aligns well with linguistic intuition. We further explore the Transformer attention pruning based on our findings in contextual analysis.</abstract>
      <url hash="7e2f6fb2">2020.findings-emnlp.338</url>
      <doi>10.18653/v1/2020.findings-emnlp.338</doi>
    </paper>
    <paper id="339">
      <title>Controllable Text Generation with Focused Variation</title>
      <author><first>Lei</first><last>Shu</last></author>
      <author><first>Alexandros</first><last>Papangelis</last></author>
      <author><first>Yi-Chia</first><last>Wang</last></author>
      <author><first>Gokhan</first><last>Tur</last></author>
      <author><first>Hu</first><last>Xu</last></author>
      <author><first>Zhaleh</first><last>Feizollahi</last></author>
      <author><first>Bing</first><last>Liu</last></author>
      <author><first>Piero</first><last>Molino</last></author>
      <pages>3805–3817</pages>
      <abstract>This work introduces Focused-Variation Network (FVN), a novel model to control language generation. The main problems in previous controlled language generation models range from the difficulty of generating text according to the given attributes, to the lack of diversity of the generated texts. FVN addresses these issues by learning disjoint discrete latent spaces for each attribute inside codebooks, which allows for both controllability and diversity, while at the same time generating fluent text. We evaluate FVN on two text generation datasets with annotated content and style, and show state-of-the-art performance as assessed by automatic and human evaluations.</abstract>
      <url hash="82fcf794">2020.findings-emnlp.339</url>
      <doi>10.18653/v1/2020.findings-emnlp.339</doi>
    </paper>
    <paper id="340">
      <title>Modeling Preconditions in Text with a Crowd-sourced Dataset</title>
      <author><first>Heeyoung</first><last>Kwon</last></author>
      <author><first>Mahnaz</first><last>Koupaee</last></author>
      <author><first>Pratyush</first><last>Singh</last></author>
      <author><first>Gargi</first><last>Sawhney</last></author>
      <author><first>Anmol</first><last>Shukla</last></author>
      <author><first>Keerthi Kumar</first><last>Kallur</last></author>
      <author><first>Nathanael</first><last>Chambers</last></author>
      <author><first>Niranjan</first><last>Balasubramanian</last></author>
      <pages>3818–3828</pages>
      <abstract>Preconditions provide a form of logical connection between events that explains why some events occur together and information that is complementary to the more widely studied relations such as causation, temporal ordering, entailment, and discourse relations. Modeling preconditions in text has been hampered in part due to the lack of large scale labeled data grounded in text. This paper introduces PeKo, a crowd-sourced annotation of <i>preconditions</i> between event pairs in newswire, an order of magnitude larger than prior text annotations. To complement this new corpus, we also introduce two challenge tasks aimed at modeling preconditions: (i) Precondition Identification – a standard classification task defined over pairs of event mentions, and (ii) Precondition Generation – a generative task aimed at testing a more general ability to reason about a given event. Evaluation on both tasks shows that modeling preconditions is challenging even for today’s large language models (LM). This suggests that precondition knowledge is not easily accessible in LM-derived representations alone. Our generation results show that fine-tuning an LM on PeKo yields better conditional relations than when trained on raw text or temporally-ordered corpora.</abstract>
      <url hash="fc4cf889">2020.findings-emnlp.340</url>
      <attachment type="OptionalSupplementaryMaterial" hash="fdf8859a">2020.findings-emnlp.340.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.340</doi>
    </paper>
    <paper id="341">
      <title>Reevaluating Adversarial Examples in Natural Language</title>
      <author><first>John</first><last>Morris</last></author>
      <author><first>Eli</first><last>Lifland</last></author>
      <author><first>Jack</first><last>Lanchantin</last></author>
      <author><first>Yangfeng</first><last>Ji</last></author>
      <author><first>Yanjun</first><last>Qi</last></author>
      <pages>3829–3839</pages>
      <abstract>State-of-the-art attacks on NLP models lack a shared definition of a what constitutes a successful attack. We distill ideas from past work into a unified framework: a successful natural language adversarial example is a perturbation that fools the model and follows some linguistic constraints. We then analyze the outputs of two state-of-the-art synonym substitution attacks. We find that their perturbations often do not preserve semantics, and 38% introduce grammatical errors. Human surveys reveal that to successfully preserve semantics, we need to significantly increase the minimum cosine similarities between the embeddings of swapped words and between the sentence encodings of original and perturbed sentences.With constraints adjusted to better preserve semantics and grammaticality, the attack success rate drops by over 70 percentage points.</abstract>
      <url hash="47aa3152">2020.findings-emnlp.341</url>
      <attachment type="OptionalSupplementaryMaterial" hash="830551b5">2020.findings-emnlp.341.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.341</doi>
    </paper>
    <paper id="342">
      <title>Question Answering with Long Multiple-Span Answers</title>
      <author><first>Ming</first><last>Zhu</last></author>
      <author><first>Aman</first><last>Ahuja</last></author>
      <author><first>Da-Cheng</first><last>Juan</last></author>
      <author><first>Wei</first><last>Wei</last></author>
      <author><first>Chandan K.</first><last>Reddy</last></author>
      <pages>3840–3849</pages>
      <abstract>Answering questions in many real-world applications often requires complex and precise information excerpted from texts spanned across a long document. However, currently no such annotated dataset is publicly available, which hinders the development of neural question-answering (QA) systems. To this end, we present MASH-QA, a Multiple Answer Spans Healthcare Question Answering dataset from the consumer health domain, where answers may need to be excerpted from multiple, non-consecutive parts of text spanned across a long document. We also propose MultiCo, a neural architecture that is able to capture the relevance among multiple answer spans, by using a query-based contextualized sentence selection approach, for forming the answer to the given question. We also demonstrate that conventional QA models are not suitable for this type of task and perform poorly in this setting. Extensive experiments are conducted, and the experimental results confirm the proposed model significantly outperforms the state-of-the-art QA models in this multi-span QA setting.</abstract>
      <url hash="a76fb63d">2020.findings-emnlp.342</url>
      <doi>10.18653/v1/2020.findings-emnlp.342</doi>
    </paper>
    <paper id="343">
      <title><fixed-case>I</fixed-case>nserting <fixed-case>I</fixed-case>nformation <fixed-case>B</fixed-case>ottlenecks for <fixed-case>A</fixed-case>ttribution in <fixed-case>T</fixed-case>ransformers</title>
      <author><first>Zhiying</first><last>Jiang</last></author>
      <author><first>Raphael</first><last>Tang</last></author>
      <author><first>Ji</first><last>Xin</last></author>
      <author><first>Jimmy</first><last>Lin</last></author>
      <pages>3850–3857</pages>
      <abstract>Pretrained transformers achieve the state of the art across tasks in natural language processing, motivating researchers to investigate their inner mechanisms. One common direction is to understand what features are important for prediction. In this paper, we apply information bottlenecks to analyze the attribution of each feature for prediction on a black-box model. We use BERT as the example and evaluate our approach both quantitatively and qualitatively. We show the effectiveness of our method in terms of attribution and the ability to provide insight into how information flows through layers. We demonstrate that our technique outperforms two competitive methods in degradation tests on four datasets. Code is available at <url>https://github.com/bazingagin/IBA</url>.</abstract>
      <url hash="acc3b412">2020.findings-emnlp.343</url>
      <doi>10.18653/v1/2020.findings-emnlp.343</doi>
    </paper>
    <paper id="344">
      <title>Event-Related Bias Removal for Real-time Disaster Events</title>
      <author><first>Salvador</first><last>Medina Maza</last></author>
      <author><first>Evangelia</first><last>Spiliopoulou</last></author>
      <author><first>Eduard</first><last>Hovy</last></author>
      <author><first>Alexander</first><last>Hauptmann</last></author>
      <pages>3858–3868</pages>
      <abstract>Social media has become an important tool to share information about crisis events such as natural disasters and mass attacks. Detecting actionable posts that contain useful information requires rapid analysis of huge volumes of data in real-time. This poses a complex problem due to the large amount of posts that do not contain any actionable information. Furthermore, the classification of information in real-time systems requires training on out-of-domain data, as we do not have any data from a new emerging crisis. Prior work focuses on models pre-trained on similar event types. However, those models capture unnecessary event-specific biases, like the location of the event, which affect the generalizability and performance of the classifiers on new unseen data from an emerging new event. In our work, we train an adversarial neural model to remove latent event-specific biases and improve the performance on tweet importance classification.</abstract>
      <url hash="e753abc0">2020.findings-emnlp.344</url>
      <doi>10.18653/v1/2020.findings-emnlp.344</doi>
    </paper>
    <paper id="345">
      <title>It’s not a Non-Issue: Negation as a Source of Error in Machine Translation</title>
      <author><first>Md Mosharaf</first><last>Hossain</last></author>
      <author><first>Antonios</first><last>Anastasopoulos</last></author>
      <author><first>Eduardo</first><last>Blanco</last></author>
      <author><first>Alexis</first><last>Palmer</last></author>
      <pages>3869–3885</pages>
      <abstract>As machine translation (MT) systems progress at a rapid pace, questions of their adequacy linger. In this study we focus on negation, a universal, core property of human language that significantly affects the semantics of an utterance. We investigate whether translating negation is an issue for modern MT systems using 17 translation directions as test bed. Through thorough analysis, we find that indeed the presence of negation can significantly impact downstream quality, in some cases resulting in quality reductions of more than 60%. We also provide a linguistically motivated analysis that directly explains the majority of our findings. We release our annotations and code to replicate our analysis here: https://github.com/mosharafhossain/negation-mt.</abstract>
      <url hash="778835b7">2020.findings-emnlp.345</url>
      <doi>10.18653/v1/2020.findings-emnlp.345</doi>
    </paper>
    <paper id="346">
      <title>Incremental Text-to-Speech Synthesis with Prefix-to-Prefix Framework</title>
      <author><first>Mingbo</first><last>Ma</last></author>
      <author><first>Baigong</first><last>Zheng</last></author>
      <author><first>Kaibo</first><last>Liu</last></author>
      <author><first>Renjie</first><last>Zheng</last></author>
      <author><first>Hairong</first><last>Liu</last></author>
      <author><first>Kainan</first><last>Peng</last></author>
      <author><first>Kenneth</first><last>Church</last></author>
      <author><first>Liang</first><last>Huang</last></author>
      <pages>3886–3896</pages>
      <abstract>Text-to-speech synthesis (TTS) has witnessed rapid progress in recent years, where neural methods became capable of producing audios with high naturalness. However, these efforts still suffer from two types of latencies: (a) the <i>computational latency</i> (synthesizing time), which grows linearly with the sentence length, and (b) the <i>input latency</i> in scenarios where the input text is incrementally available (such as in simultaneous translation, dialog generation, and assistive technologies). To reduce these latencies, we propose a neural incremental TTS approach using the prefix-to-prefix framework from simultaneous translation. We synthesize speech in an online fashion, playing a segment of audio while generating the next, resulting in an O(1) rather than O(n) latency. Experiments on English and Chinese TTS show that our approach achieves similar speech naturalness compared to full sentence TTS, but only with a constant (1-2 words) latency.</abstract>
      <url hash="6d31236a">2020.findings-emnlp.346</url>
      <doi>10.18653/v1/2020.findings-emnlp.346</doi>
    </paper>
    <paper id="347">
      <title>Joint Turn and Dialogue level User Satisfaction Estimation on Multi-Domain Conversations</title>
      <author><first>Praveen Kumar</first><last>Bodigutla</last></author>
      <author><first>Aditya</first><last>Tiwari</last></author>
      <author><first>Spyros</first><last>Matsoukas</last></author>
      <author><first>Josep</first><last>Valls-Vargas</last></author>
      <author><first>Lazaros</first><last>Polymenakos</last></author>
      <pages>3897–3909</pages>
      <abstract>Dialogue level quality estimation is vital for optimizing data driven dialogue management. Current automated methods to estimate turn and dialogue level user satisfaction employ hand-crafted features and rely on complex annotation schemes, which reduce the generalizability of the trained models. We propose a novel user satisfaction estimation approach which minimizes an adaptive multi-task loss function in order to jointly predict turn-level Response Quality labels provided by experts and explicit dialogue-level ratings provided by end users. The proposed BiLSTM based deep neural net model automatically weighs each turn’s contribution towards the estimated dialogue-level rating, implicitly encodes temporal dependencies, and removes the need to hand-craft features. On dialogues sampled from 28 Alexa domains, two dialogue systems and three user groups, the joint dialogue-level satisfaction estimation model achieved up to an absolute 27% (0.43 -&gt; 0.70) and 7% (0.63 -&gt; 0.70) improvement in linear correlation performance over baseline deep neural net and benchmark Gradient boosting regression models, respectively.</abstract>
      <url hash="fa2eb3e4">2020.findings-emnlp.347</url>
      <doi>10.18653/v1/2020.findings-emnlp.347</doi>
    </paper>
    <paper id="348">
      <title><fixed-case>A</fixed-case>rra<fixed-case>M</fixed-case>on: A Joint Navigation-Assembly Instruction Interpretation Task in Dynamic Environments</title>
      <author><first>Hyounghun</first><last>Kim</last></author>
      <author><first>Abhaysinh</first><last>Zala</last></author>
      <author><first>Graham</first><last>Burri</last></author>
      <author><first>Hao</first><last>Tan</last></author>
      <author><first>Mohit</first><last>Bansal</last></author>
      <pages>3910–3927</pages>
      <abstract>For embodied agents, navigation is an important ability but not an isolated goal. Agents are also expected to perform specific tasks after reaching the target location, such as picking up objects and assembling them into a particular arrangement. We combine Vision-andLanguage Navigation, assembling of collected objects, and object referring expression comprehension, to create a novel joint navigation-and-assembly task, named ARRAMON. During this task, the agent (similar to a PokeMON GO player) is asked to find and collect different target objects one-by-one by navigating based on natural language (English) instructions in a complex, realistic outdoor environment, but then also ARRAnge the collected objects part-by-part in an egocentric grid-layout environment. To support this task, we implement a 3D dynamic environment simulator and collect a dataset with human-written navigation and assembling instructions, and the corresponding ground truth trajectories. We also filter the collected instructions via a verification stage, leading to a total of 7.7K task instances (30.8K instructions and paths). We present results for several baseline models (integrated and biased) and metrics (nDTW, CTC, rPOD, and PTC), and the large model-human performance gap demonstrates that our task is challenging and presents a wide scope for future work.</abstract>
      <url hash="15ff1f24">2020.findings-emnlp.348</url>
      <doi>10.18653/v1/2020.findings-emnlp.348</doi>
    </paper>
    <paper id="349">
      <title>Fluent and Low-latency Simultaneous Speech-to-Speech Translation with Self-adaptive Training</title>
      <author><first>Renjie</first><last>Zheng</last></author>
      <author><first>Mingbo</first><last>Ma</last></author>
      <author><first>Baigong</first><last>Zheng</last></author>
      <author><first>Kaibo</first><last>Liu</last></author>
      <author><first>Jiahong</first><last>Yuan</last></author>
      <author><first>Kenneth</first><last>Church</last></author>
      <author><first>Liang</first><last>Huang</last></author>
      <pages>3928–3937</pages>
      <abstract>Simultaneous speech-to-speech translation is an extremely challenging but widely useful scenario that aims to generate target-language speech only a few seconds behind the source-language speech. In addition, we have to continuously translate a speech of multiple sentences, but all recent solutions merely focus on the single-sentence scenario. As a result, current approaches will accumulate more and more latencies in later sentences when the speaker talks faster and introduce unnatural pauses into translated speech when the speaker talks slower. To overcome these issues, we propose Self-Adaptive Translation which flexibly adjusts the length of translations to accommodate different source speech rates. At similar levels of translation quality (as measured by BLEU), our method generates more fluent target speech latency than the baseline, in both Zh&lt;-&gt;En directions.</abstract>
      <url hash="0ba49521">2020.findings-emnlp.349</url>
      <doi>10.18653/v1/2020.findings-emnlp.349</doi>
    </paper>
    <paper id="350">
      <title>Towards Context-Aware Code Comment Generation</title>
      <author><first>Xiaohan</first><last>Yu</last></author>
      <author><first>Quzhe</first><last>Huang</last></author>
      <author><first>Zheng</first><last>Wang</last></author>
      <author><first>Yansong</first><last>Feng</last></author>
      <author><first>Dongyan</first><last>Zhao</last></author>
      <pages>3938–3947</pages>
      <abstract>Code comments are vital for software maintenance and comprehension, but many software projects suffer from the lack of meaningful and up-to-date comments in practice. This paper presents a novel approach to automatically generate code comments at a function level by targeting object-oriented programming languages. Unlike prior work that only uses information locally available within the target function, our approach leverages broader contextual information by considering all other functions of the same class. To propagate and integrate information beyond the scope of the target function, we design a novel learning framework based on the bidirectional gated recurrent unit and a graph attention network with a pointer mechanism. We apply our approach to produce code comments for Java methods and compare it against four strong baseline methods. Experimental results show that our approach outperforms most methods by a large margin and achieves a comparable result with the state-of-the-art method.</abstract>
      <url hash="071d7832">2020.findings-emnlp.350</url>
      <attachment type="OptionalSupplementaryMaterial" hash="9e6a0884">2020.findings-emnlp.350.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.350</doi>
    </paper>
    <paper id="351">
      <title><fixed-case>MCMH</fixed-case>: Learning Multi-Chain Multi-Hop Rules for Knowledge Graph Reasoning</title>
      <author><first>Lu</first><last>Zhang</last></author>
      <author><first>Mo</first><last>Yu</last></author>
      <author><first>Tian</first><last>Gao</last></author>
      <author><first>Yue</first><last>Yu</last></author>
      <pages>3948–3954</pages>
      <abstract>Multi-hop reasoning approaches over knowledge graphs infer a missing relationship between entities with a multi-hop rule, which corresponds to a chain of relationships. We extend existing works to consider a generalized form of multi-hop rules, where each rule is a set of relation chains. To learn such generalized rules efficiently, we propose a two-step approach that first selects a small set of relation chains as a rule and then evaluates the confidence of the target relationship by jointly scoring the selected chains. A game-theoretical framework is proposed to this end to simultaneously optimize the rule selection and prediction steps. Empirical results show that our multi-chain multi-hop (MCMH) rules result in superior results compared to the standard single-chain approaches, justifying both our formulation of generalized rules and the effectiveness of the proposed learning framework.</abstract>
      <url hash="8578208c">2020.findings-emnlp.351</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f39ce920">2020.findings-emnlp.351.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.351</doi>
    </paper>
    <paper id="352">
      <title>Finding the Optimal Vocabulary Size for Neural Machine Translation</title>
      <author><first>Thamme</first><last>Gowda</last></author>
      <author><first>Jonathan</first><last>May</last></author>
      <pages>3955–3964</pages>
      <abstract>We cast neural machine translation (NMT) as a classification task in an autoregressive setting and analyze the limitations of both classification and autoregression components. Classifiers are known to perform better with balanced class distributions during training. Since the Zipfian nature of languages causes imbalanced classes, we explore its effect on NMT. We analyze the effect of various vocabulary sizes on NMT performance on multiple languages with many data sizes, and reveal an explanation for why certain vocabulary sizes are better than others.</abstract>
      <url hash="08a74488">2020.findings-emnlp.352</url>
      <attachment type="OptionalSupplementaryMaterial" hash="25d1e972">2020.findings-emnlp.352.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.352</doi>
    </paper>
    <paper id="353">
      <title>Weakly- and Semi-supervised Evidence Extraction</title>
      <author><first>Danish</first><last>Pruthi</last></author>
      <author><first>Bhuwan</first><last>Dhingra</last></author>
      <author><first>Graham</first><last>Neubig</last></author>
      <author><first>Zachary C.</first><last>Lipton</last></author>
      <pages>3965–3970</pages>
      <abstract>For many prediction tasks, stakeholders desire not only predictions but also supporting evidence that a human can use to verify its correctness. However, in practice, evidence annotations may only be available for a minority of training examples (if available at all). In this paper, we propose new methods to combine few evidence annotations (strong semi-supervision) with abundant document-level labels (weak supervision) for the task of evidence extraction. Evaluating on two classification tasks that feature evidence annotations, we find that our methods outperform baselines adapted from the interpretability literature to our task. Our approach yields gains with as few as hundred evidence annotations.</abstract>
      <url hash="2c611414">2020.findings-emnlp.353</url>
      <doi>10.18653/v1/2020.findings-emnlp.353</doi>
    </paper>
    <paper id="354">
      <title>Making Information Seeking Easier: An Improved Pipeline for Conversational Search</title>
      <author><first>Vaibhav</first><last>Kumar</last></author>
      <author><first>Jamie</first><last>Callan</last></author>
      <pages>3971–3980</pages>
      <abstract>This paper presents a highly effective pipeline for passage retrieval in a conversational search setting. The pipeline comprises of two components: Conversational Term Selection (CTS) and Multi-View Reranking (MVR). CTS is responsible for performing the first-stage of passage retrieval. Given an input question, it uses a BERT-based classifier (trained with weak supervision) to de-contextualize the input by selecting relevant terms from the dialog history. Using the question and the selected terms, it issues a query to a search engine to perform the first-stage of passage retrieval. On the other hand, MVR is responsible for contextualized passage reranking. It first constructs multiple views of the information need embedded within an input question. The views are based on the dialog history and the top documents obtained in the first-stage of retrieval. It then uses each view to rerank passages using BERT (fine-tuned for passage ranking). Finally, MVR performs a fusion over the rankings produced by the individual views. Experiments show that the above combination improves first-state retrieval as well as the overall accuracy in a reranking pipeline. On the key metric of NDCG@3, the proposed combination achieves a relative performance improvement of 14.8% over the state-of-the-art baseline and is also able to surpass the Oracle.</abstract>
      <url hash="4713382c">2020.findings-emnlp.354</url>
      <doi>10.18653/v1/2020.findings-emnlp.354</doi>
    </paper>
    <paper id="355">
      <title>Generalizable and Explainable Dialogue Generation via Explicit Action Learning</title>
      <author><first>Xinting</first><last>Huang</last></author>
      <author><first>Jianzhong</first><last>Qi</last></author>
      <author><first>Yu</first><last>Sun</last></author>
      <author><first>Rui</first><last>Zhang</last></author>
      <pages>3981–3991</pages>
      <abstract>Response generation for task-oriented dialogues implicitly optimizes two objectives at the same time: task completion and language quality. Conditioned response generation serves as an effective approach to separately and better optimize these two objectives. Such an approach relies on system action annotations which are expensive to obtain. To alleviate the need of action annotations, latent action learning is introduced to map each utterance to a latent representation. However, this approach is prone to over-dependence on the training data, and the generalization capability is thus restricted. To address this issue, we propose to learn <i>natural language actions</i> that represent utterances as a span of words. This explicit action representation promotes generalization via the compositional structure of language. It also enables an explainable generation process. Our proposed unsupervised approach learns a memory component to summarize system utterances into a short span of words. To further promote a compact action representation, we propose an auxiliary task that restores state annotations as the summarized dialogue context using the memory component. Our proposed approach outperforms latent action baselines on MultiWOZ, a benchmark multi-domain dataset.</abstract>
      <url hash="b8cd95a5">2020.findings-emnlp.355</url>
      <doi>10.18653/v1/2020.findings-emnlp.355</doi>
    </paper>
    <paper id="356">
      <title>More Embeddings, Better Sequence Labelers?</title>
      <author><first>Xinyu</first><last>Wang</last></author>
      <author><first>Yong</first><last>Jiang</last></author>
      <author><first>Nguyen</first><last>Bach</last></author>
      <author><first>Tao</first><last>Wang</last></author>
      <author><first>Zhongqiang</first><last>Huang</last></author>
      <author><first>Fei</first><last>Huang</last></author>
      <author><first>Kewei</first><last>Tu</last></author>
      <pages>3992–4006</pages>
      <abstract>Recent work proposes a family of contextual embeddings that significantly improves the accuracy of sequence labelers over non-contextual embeddings. However, there is no definite conclusion on whether we can build better sequence labelers by combining different kinds of embeddings in various settings. In this paper, we conduct extensive experiments on 3 tasks over 18 datasets and 8 languages to study the accuracy of sequence labeling with various embedding concatenations and make three observations: (1) concatenating more embedding variants leads to better accuracy in rich-resource and cross-domain settings and some conditions of low-resource settings; (2) concatenating contextual sub-word embeddings with contextual character embeddings hurts the accuracy in extremely low-resource settings; (3) based on the conclusion of (1), concatenating additional similar contextual embeddings cannot lead to further improvements. We hope these conclusions can help people build stronger sequence labelers in various settings.</abstract>
      <url hash="42f6e4b0">2020.findings-emnlp.356</url>
      <doi>10.18653/v1/2020.findings-emnlp.356</doi>
    </paper>
    <paper id="357">
      <title><fixed-case>NLP</fixed-case> Service <fixed-case>API</fixed-case>s and Models for Efficient Registration of New Clients</title>
      <author><first>Sahil</first><last>Shah</last></author>
      <author><first>Vihari</first><last>Piratla</last></author>
      <author><first>Soumen</first><last>Chakrabarti</last></author>
      <author><first>Sunita</first><last>Sarawagi</last></author>
      <pages>4007–4012</pages>
      <abstract>State-of-the-art NLP inference uses enormous neural architectures and models trained for GPU-months, well beyond the reach of most consumers of NLP. This has led to one-size-fits-all public API-based NLP service models by major AI companies, serving millions of clients. They cannot afford traditional fine tuning for individual clients. Many clients cannot even afford significant fine tuning, and own little or no labeled data. Recognizing that word usage and salience diversity across clients leads to reduced accuracy, we initiate a study of practical and lightweight adaptation of centralized NLP services to clients. Each client uses an unsupervised, corpus-based sketch to register to the service. The server modifies its network mildly to accommodate client sketches, and occasionally trains the augmented network over existing clients. When a new client registers with its sketch, it gets immediate accuracy benefits. We demonstrate the proposed architecture using sentiment labeling, NER, and predictive language modeling.</abstract>
      <url hash="3062c304">2020.findings-emnlp.357</url>
      <attachment type="OptionalSupplementaryMaterial" hash="bc100795">2020.findings-emnlp.357.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.357</doi>
    </paper>
    <paper id="358">
      <title>Effects of Naturalistic Variation in Goal-Oriented Dialog</title>
      <author><first>Jatin</first><last>Ganhotra</last></author>
      <author><first>Robert</first><last>Moore</last></author>
      <author><first>Sachindra</first><last>Joshi</last></author>
      <author><first>Kahini</first><last>Wadhawan</last></author>
      <pages>4013–4020</pages>
      <abstract>Existing benchmarks used to evaluate the performance of end-to-end neural dialog systems lack a key component: natural variation present in human conversations. Most datasets are constructed through crowdsourcing, where the crowd workers follow a fixed template of instructions while enacting the role of a user/agent. This results in straight-forward, somewhat routine, and mostly trouble-free conversations, as crowd workers do not think to represent the full range of actions that occur naturally with real users. In this work, we investigate the impact of naturalistic variation on two goal-oriented datasets: bAbI dialog task and Stanford Multi-Domain Dataset (SMD). We also propose new and more effective testbeds for both datasets, by introducing naturalistic variation by the user. We observe that there is a significant drop in performance (more than 60% in Ent. F1 on SMD and 85% in per-dialog accuracy on bAbI task) of recent state-of-the-art end-to-end neural methods such as BossNet and GLMP on both datasets.</abstract>
      <url hash="9e76f4b9">2020.findings-emnlp.358</url>
      <doi>10.18653/v1/2020.findings-emnlp.358</doi>
    </paper>
    <paper id="359">
      <title>Determining Event Outcomes: The Case of #fail</title>
      <author><first>Srikala</first><last>Murugan</last></author>
      <author><first>Dhivya</first><last>Chinnappa</last></author>
      <author><first>Eduardo</first><last>Blanco</last></author>
      <pages>4021–4033</pages>
      <abstract>This paper targets the task of determining event outcomes in social media. We work with tweets containing either #cookingFail or #bakingFail, and show that many of the events described in them resulted in something edible. Tweets that contain images are more likely to result in edible albeit imperfect outcomes. Experimental results show that edibility is easier to predict than outcome quality.</abstract>
      <url hash="7aeb49c4">2020.findings-emnlp.359</url>
      <attachment type="OptionalSupplementaryMaterial" hash="00000000">2020.findings-emnlp.359.OptionalSupplementaryMaterial.txt</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.359</doi>
    </paper>
    <paper id="360">
      <title><fixed-case>W</fixed-case>iki<fixed-case>L</fixed-case>ingua: A New Benchmark Dataset for Cross-Lingual Abstractive Summarization</title>
      <author><first>Faisal</first><last>Ladhak</last></author>
      <author><first>Esin</first><last>Durmus</last></author>
      <author><first>Claire</first><last>Cardie</last></author>
      <author><first>Kathleen</first><last>McKeown</last></author>
      <pages>4034–4048</pages>
      <abstract>We introduce WikiLingua, a large-scale, multilingual dataset for the evaluation of cross-lingual abstractive summarization systems. We extract article and summary pairs in 18 languages from WikiHow, a high quality, collaborative resource of how-to guides on a diverse set of topics written by human authors. We create gold-standard article-summary alignments across languages by aligning the images that are used to describe each how-to step in an article. As a set of baselines for further studies, we evaluate the performance of existing cross-lingual abstractive summarization methods on our dataset. We further propose a method for direct cross-lingual summarization (i.e., without requiring translation at inference time) by leveraging synthetic data and Neural Machine Translation as a pre-training step. Our method significantly outperforms the baseline approaches, while being more cost efficient during inference.</abstract>
      <url hash="fd3359d1">2020.findings-emnlp.360</url>
      <doi>10.18653/v1/2020.findings-emnlp.360</doi>
    </paper>
    <paper id="361">
      <title>Adversarial Training for Code Retrieval with Question-Description Relevance Regularization</title>
      <author><first>Jie</first><last>Zhao</last></author>
      <author><first>Huan</first><last>Sun</last></author>
      <pages>4049–4059</pages>
      <abstract>Code retrieval is a key task aiming to match natural and programming languages. In this work, we propose adversarial learning for code retrieval, that is regularized by question-description relevance. First, we adapt a simple adversarial learning technique to generate difficult code snippets given the input question, which can help the learning of code retrieval that faces bi-modal and data-scarce challenges. Second, we propose to leverage question-description relevance to regularize adversarial learning, such that a generated code snippet should contribute more to the code retrieval training loss, only if its paired natural language description is predicted to be less relevant to the user given question. Experiments on large-scale code retrieval datasets of two programming languages show that our adversarial learning method is able to improve the performance of state-of-the-art models. Moreover, using an additional duplicated question detection model to regularize adversarial learning further improves the performance, and this is more effective than using the duplicated questions in strong multi-task learning baselines.</abstract>
      <url hash="1cafaadc">2020.findings-emnlp.361</url>
      <doi>10.18653/v1/2020.findings-emnlp.361</doi>
    </paper>
    <paper id="362">
      <title>Large Product Key Memory for Pretrained Language Models</title>
      <author><first>Gyuwan</first><last>Kim</last></author>
      <author><first>Tae Hwan</first><last>Jung</last></author>
      <pages>4060–4069</pages>
      <abstract>Product key memory (PKM) proposed by Lample et al. (2019) enables to improve prediction accuracy by increasing model capacity efficiently with insignificant computational overhead. However, their empirical application is only limited to causal language modeling. Motivated by the recent success of pretrained language models (PLMs), we investigate how to incorporate large PKM into PLMs that can be finetuned for a wide variety of downstream NLP tasks. We define a new memory usage metric, and careful observation using this metric reveals that most memory slots remain outdated during the training of PKM-augmented models. To train better PLMs by tackling this issue, we propose simple but effective solutions: (1) initialization from the model weights pretrained without memory and (2) augmenting PKM by addition rather than replacing a feed-forward network. We verify that both of them are crucial for the pretraining of PKM-augmented PLMs, enhancing memory utilization and downstream performance. Code and pretrained weights are available at https://github.com/clovaai/pkm-transformers.</abstract>
      <url hash="e8bdeadc">2020.findings-emnlp.362</url>
      <doi>10.18653/v1/2020.findings-emnlp.362</doi>
    </paper>
    <paper id="363">
      <title>Temporal Reasoning in Natural Language Inference</title>
      <author><first>Siddharth</first><last>Vashishtha</last></author>
      <author><first>Adam</first><last>Poliak</last></author>
      <author><first>Yash Kumar</first><last>Lal</last></author>
      <author><first>Benjamin</first><last>Van Durme</last></author>
      <author><first>Aaron Steven</first><last>White</last></author>
      <pages>4070–4078</pages>
      <abstract>We introduce five new natural language inference (NLI) datasets focused on temporal reasoning. We recast four existing datasets annotated for event duration—how long an event lasts—and event ordering—how events are temporally arranged—into more than one million NLI examples. We use these datasets to investigate how well neural models trained on a popular NLI corpus capture these forms of temporal reasoning.</abstract>
      <url hash="2c9158e6">2020.findings-emnlp.363</url>
      <doi>10.18653/v1/2020.findings-emnlp.363</doi>
    </paper>
    <paper id="364">
      <title>A Pilot Study of Text-to-<fixed-case>SQL</fixed-case> Semantic Parsing for <fixed-case>V</fixed-case>ietnamese</title>
      <author><first>Anh</first><last>Tuan Nguyen</last></author>
      <author><first>Mai Hoang</first><last>Dao</last></author>
      <author><first>Dat Quoc</first><last>Nguyen</last></author>
      <pages>4079–4085</pages>
      <abstract>Semantic parsing is an important NLP task. However, Vietnamese is a low-resource language in this research area. In this paper, we present the first public large-scale Text-to-SQL semantic parsing dataset for Vietnamese. We extend and evaluate two strong semantic parsing baselines EditSQL (Zhang et al., 2019) and IRNet (Guo et al., 2019) on our dataset. We compare the two baselines with key configurations and find that: automatic Vietnamese word segmentation improves the parsing results of both baselines; the normalized pointwise mutual information (NPMI) score (Bouma, 2009) is useful for schema linking; latent syntactic features extracted from a neural dependency parser for Vietnamese also improve the results; and the monolingual language model PhoBERT for Vietnamese (Nguyen and Nguyen, 2020) helps produce higher performances than the recent best multilingual language model XLM-R (Conneau et al., 2020).</abstract>
      <url hash="33d89f8a">2020.findings-emnlp.364</url>
      <doi>10.18653/v1/2020.findings-emnlp.364</doi>
    </paper>
    <paper id="365">
      <title><fixed-case>STANDER</fixed-case>: An Expert-Annotated Dataset for News Stance Detection and Evidence Retrieval</title>
      <author><first>Costanza</first><last>Conforti</last></author>
      <author><first>Jakob</first><last>Berndt</last></author>
      <author><first>Mohammad Taher</first><last>Pilehvar</last></author>
      <author><first>Chryssi</first><last>Giannitsarou</last></author>
      <author><first>Flavio</first><last>Toxvaerd</last></author>
      <author><first>Nigel</first><last>Collier</last></author>
      <pages>4086–4101</pages>
      <abstract>We present a new challenging news dataset that targets both stance detection (SD) and fine-grained evidence retrieval (ER). With its 3,291 expert-annotated articles, the dataset constitutes a high-quality benchmark for future research in SD and multi-task learning. We provide a detailed description of the corpus collection methodology and carry out an extensive analysis on the sources of disagreement between annotators, observing a correlation between their disagreement and the diffusion of uncertainty around a target in the real world. Our experiments show that the dataset poses a strong challenge to recent state-of-the-art models. Notably, our dataset aligns with an existing Twitter SD dataset: their union thus addresses a key shortcoming of previous works, by providing the first dedicated resource to study multi-genre SD as well as the interplay of signals from social media and news sources in rumour verification.</abstract>
      <url hash="652075ed">2020.findings-emnlp.365</url>
      <doi>10.18653/v1/2020.findings-emnlp.365</doi>
    </paper>
    <paper id="366">
      <title>An Empirical Methodology for Detecting and Prioritizing Needs during Crisis Events</title>
      <author><first>M. Janina</first><last>Sarol</last></author>
      <author><first>Ly</first><last>Dinh</last></author>
      <author><first>Rezvaneh</first><last>Rezapour</last></author>
      <author><first>Chieh-Li</first><last>Chin</last></author>
      <author><first>Pingjing</first><last>Yang</last></author>
      <author><first>Jana</first><last>Diesner</last></author>
      <pages>4102–4107</pages>
      <abstract>In times of crisis, identifying essential needs is crucial to providing appropriate resources and services to affected entities. Social media platforms such as Twitter contain a vast amount of information about the general public’s needs. However, the sparsity of information and the amount of noisy content present a challenge for practitioners to effectively identify relevant information on these platforms. This study proposes two novel methods for two needs detection tasks: 1) extracting a list of needed resources, such as masks and ventilators, and 2) detecting sentences that specify who-needs-what resources (e.g., we need testing). We evaluate our methods on a set of tweets about the COVID-19 crisis. For extracting a list of needs, we compare our results against two official lists of resources, achieving 0.64 precision. For detecting who-needs-what sentences, we compared our results against a set of 1,000 annotated tweets and achieved a 0.68 F1-score.</abstract>
      <url hash="905a5e9e">2020.findings-emnlp.366</url>
      <doi>10.18653/v1/2020.findings-emnlp.366</doi>
    </paper>
    <paper id="367">
      <title><fixed-case>S</fixed-case>up<fixed-case>MMD</fixed-case>: A Sentence Importance Model for Extractive Summarization using Maximum Mean Discrepancy</title>
      <author><first>Umanga</first><last>Bista</last></author>
      <author><first>Alexander</first><last>Mathews</last></author>
      <author><first>Aditya</first><last>Menon</last></author>
      <author><first>Lexing</first><last>Xie</last></author>
      <pages>4108–4122</pages>
      <abstract>Most work on multi-document summarization has focused on generic summarization of information present in each individual document set. However, the under-explored setting of update summarization, where the goal is to identify the new information present in each set, is of equal practical interest (e.g., presenting readers with updates on an evolving news topic). In this work, we present SupMMD, a novel technique for generic and update summarization based on the maximum mean discrepancy from kernel two-sample testing. SupMMD combines both supervised learning for salience and unsupervised learning for coverage and diversity. Further, we adapt multiple kernel learning to make use of similarity across multiple information sources (e.g., text features and knowledge based concepts). We show the efficacy of SupMMD in both generic and update summarization tasks by meeting or exceeding the current state-of-the-art on the DUC-2004 and TAC-2009 datasets.</abstract>
      <url hash="0b3ff305">2020.findings-emnlp.367</url>
      <doi>10.18653/v1/2020.findings-emnlp.367</doi>
    </paper>
    <paper id="368">
      <title>Towards Low-Resource Semi-Supervised Dialogue Generation with Meta-Learning</title>
      <author><first>Yi</first><last>Huang</last></author>
      <author><first>Junlan</first><last>Feng</last></author>
      <author><first>Shuo</first><last>Ma</last></author>
      <author><first>Xiaoyu</first><last>Du</last></author>
      <author><first>Xiaoting</first><last>Wu</last></author>
      <pages>4123–4128</pages>
      <abstract>In this paper, we propose a meta-learning based semi-supervised explicit dialogue state tracker (SEDST) for neural dialogue generation, denoted as MEDST. Our main motivation is to further bridge the chasm between the need for high accuracy dialogue state tracker and the common reality that only scarce annotated data is available for most real-life dialogue tasks. Specifically, MEDST has two core steps: meta-training with adequate unlabelled data in an automatic way and meta-testing with a few annotated data by supervised learning. In particular, we enhance SEDST via entropy regularization, and investigate semi-supervised learning frameworks based on model-agnostic meta-learning (MAML) that are able to reduce the amount of required intermediate state labelling. We find that by leveraging un-annotated data in meta-way instead, the amount of dialogue state annotations can be reduced below 10% while maintaining equivalent system performance. Experimental results show MEDST outperforms SEDST substantially by 18.7% joint goal accuracy and 14.3% entity match rate on the KVRET corpus with 2% labelled data in semi-supervision.</abstract>
      <url hash="d006fde3">2020.findings-emnlp.368</url>
      <doi>10.18653/v1/2020.findings-emnlp.368</doi>
    </paper>
    <paper id="369">
      <title>Connecting the Dots: A Knowledgeable Path Generator for Commonsense Question Answering</title>
      <author><first>Peifeng</first><last>Wang</last></author>
      <author><first>Nanyun</first><last>Peng</last></author>
      <author><first>Filip</first><last>Ilievski</last></author>
      <author><first>Pedro</first><last>Szekely</last></author>
      <author><first>Xiang</first><last>Ren</last></author>
      <pages>4129–4140</pages>
      <abstract>Commonsense question answering (QA) requires background knowledge which is not explicitly stated in a given context. Prior works use commonsense knowledge graphs (KGs) to obtain this knowledge for reasoning. However, relying entirely on these KGs may not suffice, considering their limited coverage and the contextual dependence of their knowledge. In this paper, we augment a general commonsense QA framework with a knowledgeable path generator. By extrapolating over existing paths in a KG with a state-of-the-art language model, our generator learns to connect a pair of entities in text with a dynamic, and potentially novel, multi-hop relational path. Such paths can provide structured evidence for solving commonsense questions without fine-tuning the path generator. Experiments on two datasets show the superiority of our method over previous works which fully rely on knowledge from KGs (with up to 6% improvement in accuracy), across various amounts of training data. Further evaluation suggests that the generated paths are typically interpretable, novel, and relevant to the task.</abstract>
      <url hash="e40392a1">2020.findings-emnlp.369</url>
      <doi>10.18653/v1/2020.findings-emnlp.369</doi>
    </paper>
    <paper id="370">
      <title>No Answer is Better Than Wrong Answer: A Reflection Model for Document Level Machine Reading Comprehension</title>
      <author><first>Xuguang</first><last>Wang</last></author>
      <author><first>Linjun</first><last>Shou</last></author>
      <author><first>Ming</first><last>Gong</last></author>
      <author><first>Nan</first><last>Duan</last></author>
      <author><first>Daxin</first><last>Jiang</last></author>
      <pages>4141–4150</pages>
      <abstract>The Natural Questions (NQ) benchmark set brings new challenges to Machine Reading Comprehension: the answers are not only at different levels of granularity (long and short), but also of richer types (including no-answer, yes/no, single-span and multi-span). In this paper, we target at this challenge and handle all answer types systematically. In particular, we propose a novel approach called Reflection Net which leverages a two-step training procedure to identify the no-answer and wrong-answer cases. Extensive experiments are conducted to verify the effectiveness of our approach. At the time of paper writing (May. 20, 2020), our approach achieved the top 1 on both long and short answer leaderboard, with F1 scores of 77.2 and 64.1, respectively.</abstract>
      <url hash="5d99278b">2020.findings-emnlp.370</url>
      <attachment type="OptionalSupplementaryMaterial" hash="629a3c6a">2020.findings-emnlp.370.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.370</doi>
    </paper>
    <paper id="371">
      <title>Reference Language based Unsupervised Neural Machine Translation</title>
      <author><first>Zuchao</first><last>Li</last></author>
      <author><first>Hai</first><last>Zhao</last></author>
      <author><first>Rui</first><last>Wang</last></author>
      <author><first>Masao</first><last>Utiyama</last></author>
      <author><first>Eiichiro</first><last>Sumita</last></author>
      <pages>4151–4162</pages>
      <abstract>Exploiting a common language as an auxiliary for better translation has a long tradition in machine translation and lets supervised learning-based machine translation enjoy the enhancement delivered by the well-used pivot language in the absence of a source language to target language parallel corpus. The rise of unsupervised neural machine translation (UNMT) almost completely relieves the parallel corpus curse, though UNMT is still subject to unsatisfactory performance due to the vagueness of the clues available for its core back-translation training. Further enriching the idea of pivot translation by extending the use of parallel corpora beyond the source-target paradigm, we propose a new reference language-based framework for UNMT, RUNMT, in which the reference language only shares a parallel corpus with the source, but this corpus still indicates a signal clear enough to help the reconstruction training of UNMT through a proposed reference agreement mechanism. Experimental results show that our methods improve the quality of UNMT over that of a strong baseline that uses only one auxiliary language, demonstrating the usefulness of the proposed reference language-based UNMT and establishing a good start for the community.</abstract>
      <url hash="4b344077">2020.findings-emnlp.371</url>
      <attachment type="OptionalSupplementaryMaterial" hash="5add9c16">2020.findings-emnlp.371.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.371</doi>
    </paper>
    <paper id="372">
      <title><fixed-case>T</fixed-case>iny<fixed-case>BERT</fixed-case>: Distilling <fixed-case>BERT</fixed-case> for Natural Language Understanding</title>
      <author><first>Xiaoqi</first><last>Jiao</last></author>
      <author><first>Yichun</first><last>Yin</last></author>
      <author><first>Lifeng</first><last>Shang</last></author>
      <author><first>Xin</first><last>Jiang</last></author>
      <author><first>Xiao</first><last>Chen</last></author>
      <author><first>Linlin</first><last>Li</last></author>
      <author><first>Fang</first><last>Wang</last></author>
      <author><first>Qun</first><last>Liu</last></author>
      <pages>4163–4174</pages>
      <abstract>Language model pre-training, such as BERT, has significantly improved the performances of many natural language processing tasks. However, pre-trained language models are usually computationally expensive, so it is difficult to efficiently execute them on resource-restricted devices. To accelerate inference and reduce model size while maintaining accuracy, we first propose a novel Transformer distillation method that is specially designed for knowledge distillation (KD) of the Transformer-based models. By leveraging this new KD method, the plenty of knowledge encoded in a large “teacher” BERT can be effectively transferred to a small “student” TinyBERT. Then, we introduce a new two-stage learning framework for TinyBERT, which performs Transformer distillation at both the pre-training and task-specific learning stages. This framework ensures that TinyBERT can capture the general-domain as well as the task-specific knowledge in BERT. TinyBERT4 with 4 layers is empirically effective and achieves more than 96.8% the performance of its teacher BERT-Base on GLUE benchmark, while being 7.5x smaller and 9.4x faster on inference. TinyBERT4 is also significantly better than 4-layer state-of-the-art baselines on BERT distillation, with only ~28% parameters and ~31% inference time of them. Moreover, TinyBERT6 with 6 layers performs on-par with its teacher BERT-Base.</abstract>
      <url hash="b6e4f88f">2020.findings-emnlp.372</url>
      <attachment type="OptionalSupplementaryMaterial" hash="b3d1f889">2020.findings-emnlp.372.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.372</doi>
    </paper>
    <paper id="373">
      <title>Poison Attacks against Text Datasets with Conditional Adversarially Regularized Autoencoder</title>
      <author><first>Alvin</first><last>Chan</last></author>
      <author><first>Yi</first><last>Tay</last></author>
      <author><first>Yew-Soon</first><last>Ong</last></author>
      <author><first>Aston</first><last>Zhang</last></author>
      <pages>4175–4189</pages>
      <abstract>This paper demonstrates a fatal vulnerability in natural language inference (NLI) and text classification systems. More concretely, we present a ‘backdoor poisoning’ attack on NLP models. Our poisoning attack utilizes conditional adversarially regularized autoencoder (CARA) to generate poisoned training samples by poison injection in latent space. Just by adding 1% poisoned data, our experiments show that a victim BERT finetuned classifier’s predictions can be steered to the poison target class with success rates of <tex-math>&gt;80\%</tex-math> when the input hypothesis is injected with the poison signature, demonstrating that NLI and text classification systems face a huge security risk.</abstract>
      <url hash="b4249153">2020.findings-emnlp.373</url>
      <doi>10.18653/v1/2020.findings-emnlp.373</doi>
    </paper>
    <paper id="374">
      <title>#Turki$h<fixed-case>T</fixed-case>weets: A Benchmark Dataset for <fixed-case>T</fixed-case>urkish Text Correction</title>
      <author><first>Asiye Tuba</first><last>Koksal</last></author>
      <author><first>Ozge</first><last>Bozal</last></author>
      <author><first>Emre</first><last>Yürekli</last></author>
      <author><first>Gizem</first><last>Gezici</last></author>
      <pages>4190–4198</pages>
      <abstract>#Turki<tex-math>hTweets is a benchmark dataset for the task of correcting the user misspellings, with the purpose of introducing the first public Turkish dataset in this area. \#Turki</tex-math>hTweets provides correct/incorrect word annotations with a detailed misspelling category formulation based on the real user data. We evaluated four state-of-the-art approaches on our dataset to present a preliminary analysis for the sake of reproducibility.</abstract>
      <url hash="0f0541fb">2020.findings-emnlp.374</url>
      <attachment type="OptionalSupplementaryMaterial" hash="181d6a17">2020.findings-emnlp.374.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.374</doi>
    </paper>
    <paper id="375">
      <title>Assessing Human-Parity in Machine Translation on the Segment Level</title>
      <author><first>Yvette</first><last>Graham</last></author>
      <author><first>Christian</first><last>Federmann</last></author>
      <author><first>Maria</first><last>Eskevich</last></author>
      <author><first>Barry</first><last>Haddow</last></author>
      <pages>4199–4207</pages>
      <abstract>Recent machine translation shared tasks have shown top-performing systems to tie or in some cases even outperform human translation. Such conclusions about system and human performance are, however, based on estimates aggregated from scores collected over large test sets of translations and unfortunately leave some remaining questions unanswered. For instance, simply because a system significantly outperforms the human translator on average may not necessarily mean that it has done so for every translation in the test set. Firstly, are there remaining source segments present in evaluation test sets that cause significant challenges for top-performing systems and can such challenging segments go unnoticed due to the opacity of current human evaluation procedures? To provide insight into these issues we carefully inspect the outputs of top-performing systems in the most recent WMT-19 news translation shared task for all language pairs in which a system either tied or outperformed human translation. Our analysis provides a new method of identifying the remaining segments for which either machine or human perform poorly. For example, in our close inspection of WMT-19 English to German and German to English we discover the segments that disjointly proved a challenge for human and machine. For English to Russian, there were no segments included in our sample of translations that caused a significant challenge for the human translator, while we again identify the set of segments that caused issues for the top-performing system.</abstract>
      <url hash="f79a15e0">2020.findings-emnlp.375</url>
      <attachment type="OptionalSupplementaryMaterial" hash="7eecee8b">2020.findings-emnlp.375.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.375</doi>
    </paper>
    <paper id="376">
      <title>Multichannel <fixed-case>G</fixed-case>enerative <fixed-case>L</fixed-case>anguage <fixed-case>M</fixed-case>odel: <fixed-case>L</fixed-case>earning <fixed-case>A</fixed-case>ll <fixed-case>P</fixed-case>ossible <fixed-case>F</fixed-case>actorizations <fixed-case>W</fixed-case>ithin and <fixed-case>A</fixed-case>cross <fixed-case>C</fixed-case>hannels</title>
      <author><first>Harris</first><last>Chan</last></author>
      <author><first>Jamie</first><last>Kiros</last></author>
      <author><first>William</first><last>Chan</last></author>
      <pages>4208–4220</pages>
      <abstract>A channel corresponds to a viewpoint or transformation of an underlying meaning. A pair of parallel sentences in English and French express the same underlying meaning, but through two separate channels corresponding to their languages. In this work, we present the Multichannel Generative Language Model (MGLM). MGLM is a generative joint distribution model over channels. MGLM marginalizes over all possible factorizations within and across all channels. MGLM endows flexible inference, including unconditional generation, conditional generation (where 1 channel is observed and other channels are generated), and partially observed generation (where incomplete observations are spread across all the channels). We experiment with the Multi30K dataset containing English, French, Czech, and German. We demonstrate experiments with unconditional, conditional, and partially conditional generation. We provide qualitative samples sampled unconditionally from the generative joint distribution. We also quantitatively analyze the quality-diversity trade-offs and find MGLM outperforms traditional bilingual discriminative models.</abstract>
      <url hash="9274c757">2020.findings-emnlp.376</url>
      <doi>10.18653/v1/2020.findings-emnlp.376</doi>
    </paper>
    <paper id="377">
      <title>Factorized Transformer for Multi-Domain Neural Machine Translation</title>
      <author><first>Yongchao</first><last>Deng</last></author>
      <author><first>Hongfei</first><last>Yu</last></author>
      <author><first>Heng</first><last>Yu</last></author>
      <author><first>Xiangyu</first><last>Duan</last></author>
      <author><first>Weihua</first><last>Luo</last></author>
      <pages>4221–4230</pages>
      <abstract>Multi-Domain Neural Machine Translation (NMT) aims at building a single system that performs well on a range of target domains. However, along with the extreme diversity of cross-domain wording and phrasing style, the imperfections of training data distribution and the inherent defects of the current sequential learning process all contribute to making the task of multi-domain NMT very challenging. To mitigate these problems, we propose the Factorized Transformer, which consists of an in-depth factorization of the parameters of an NMT model, namely Transformer in this paper, into two categories: domain-shared ones that encode common cross-domain knowledge and domain-specific ones that are private for each constituent domain. We experiment with various designs of our model and conduct extensive validations on English to French open multi-domain dataset. Our approach achieves state-of-the-art performance and opens up new perspectives for multi-domain and open-domain applications.</abstract>
      <url hash="eb9a91ef">2020.findings-emnlp.377</url>
      <doi>10.18653/v1/2020.findings-emnlp.377</doi>
    </paper>
    <paper id="378">
      <title>Improving Named Entity Recognition with Attentive Ensemble of Syntactic Information</title>
      <author><first>Yuyang</first><last>Nie</last></author>
      <author><first>Yuanhe</first><last>Tian</last></author>
      <author><first>Yan</first><last>Song</last></author>
      <author><first>Xiang</first><last>Ao</last></author>
      <author><first>Xiang</first><last>Wan</last></author>
      <pages>4231–4245</pages>
      <abstract>Named entity recognition (NER) is highly sensitive to sentential syntactic and semantic properties where entities may be extracted according to how they are used and placed in the running text. To model such properties, one could rely on existing resources to providing helpful knowledge to the NER task; some existing studies proved the effectiveness of doing so, and yet are limited in appropriately leveraging the knowledge such as distinguishing the important ones for particular context. In this paper, we improve NER by leveraging different types of syntactic information through attentive ensemble, which functionalizes by the proposed key-value memory networks, syntax attention, and the gate mechanism for encoding, weighting and aggregating such syntactic information, respectively. Experimental results on six English and Chinese benchmark datasets suggest the effectiveness of the proposed model and show that it outperforms previous studies on all experiment datasets.</abstract>
      <url hash="21d7ce12">2020.findings-emnlp.378</url>
      <doi>10.18653/v1/2020.findings-emnlp.378</doi>
    </paper>
    <paper id="379">
      <title>Query-Key Normalization for Transformers</title>
      <author><first>Alex</first><last>Henry</last></author>
      <author><first>Prudhvi Raj</first><last>Dachapally</last></author>
      <author><first>Shubham Shantaram</first><last>Pawar</last></author>
      <author><first>Yuxuan</first><last>Chen</last></author>
      <pages>4246–4253</pages>
      <abstract>Low-resource language translation is a challenging but socially valuable NLP task. Building on recent work adapting the Transformer’s normalization to this setting, we propose QKNorm, a normalization technique that modifies the attention mechanism to make the softmax function less prone to arbitrary saturation without sacrificing expressivity. Specifically, we apply l2-normalization along the head dimension of each query and key matrix prior to multiplying them and then scale up by a learnable parameter instead of dividing by the square root of the embedding dimension. We show improvements averaging 0.928 BLEU over state-of-the-art bilingual benchmarks for 5 low-resource translation pairs from the TED Talks corpus and IWSLT’15.</abstract>
      <url hash="738b2711">2020.findings-emnlp.379</url>
      <attachment type="OptionalSupplementaryMaterial" hash="5482f430">2020.findings-emnlp.379.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.379</doi>
    </paper>
    <paper id="380">
      <title>Contract Discovery: Dataset and a Few-Shot Semantic Retrieval Challenge with Competitive Baselines</title>
      <author><first>Łukasz</first><last>Borchmann</last></author>
      <author><first>Dawid</first><last>Wisniewski</last></author>
      <author><first>Andrzej</first><last>Gretkowski</last></author>
      <author><first>Izabela</first><last>Kosmala</last></author>
      <author><first>Dawid</first><last>Jurkiewicz</last></author>
      <author><first>Łukasz</first><last>Szałkiewicz</last></author>
      <author><first>Gabriela</first><last>Pałka</last></author>
      <author><first>Karol</first><last>Kaczmarek</last></author>
      <author><first>Agnieszka</first><last>Kaliska</last></author>
      <author><first>Filip</first><last>Graliński</last></author>
      <pages>4254–4268</pages>
      <abstract>We propose a new shared task of semantic retrieval from legal texts, in which a so-called contract discovery is to be performed – where legal clauses are extracted from documents, given a few examples of similar clauses from other legal acts. The task differs substantially from conventional NLI and shared tasks on legal information extraction (e.g., one has to identify text span instead of a single document, page, or paragraph). The specification of the proposed task is followed by an evaluation of multiple solutions within the unified framework proposed for this branch of methods. It is shown that state-of-the-art pretrained encoders fail to provide satisfactory results on the task proposed. In contrast, Language Model-based solutions perform better, especially when unsupervised fine-tuning is applied. Besides the ablation studies, we addressed questions regarding detection accuracy for relevant text fragments depending on the number of examples available. In addition to the dataset and reference results, LMs specialized in the legal domain were made publicly available.</abstract>
      <url hash="18495437">2020.findings-emnlp.380</url>
      <doi>10.18653/v1/2020.findings-emnlp.380</doi>
    </paper>
    <paper id="381">
      <title>Vocabulary Adaptation for Domain Adaptation in Neural Machine Translation</title>
      <author><first>Shoetsu</first><last>Sato</last></author>
      <author><first>Jin</first><last>Sakuma</last></author>
      <author><first>Naoki</first><last>Yoshinaga</last></author>
      <author><first>Masashi</first><last>Toyoda</last></author>
      <author><first>Masaru</first><last>Kitsuregawa</last></author>
      <pages>4269–4279</pages>
      <abstract>Neural network methods exhibit strong performance only in a few resource-rich domains. Practitioners therefore employ domain adaptation from resource-rich domains that are, in most cases, distant from the target domain. Domain adaptation between distant domains (e.g., movie subtitles and research papers), however, cannot be performed effectively due to mismatches in vocabulary; it will encounter many domain-specific words (e.g., “angstrom”) and words whose meanings shift across domains (e.g., “conductor”). In this study, aiming to solve these vocabulary mismatches in domain adaptation for neural machine translation (NMT), we propose vocabulary adaptation, a simple method for effective fine-tuning that adapts embedding layers in a given pretrained NMT model to the target domain. Prior to fine-tuning, our method replaces the embedding layers of the NMT model by projecting general word embeddings induced from monolingual data in a target domain onto a source-domain embedding space. Experimental results indicate that our method improves the performance of conventional fine-tuning by 3.86 and 3.28 BLEU points in En-Ja and De-En translation, respectively.</abstract>
      <url hash="1fd31a0a">2020.findings-emnlp.381</url>
      <doi>10.18653/v1/2020.findings-emnlp.381</doi>
    </paper>
    <paper id="382">
      <title>A Shared-Private Representation Model with Coarse-to-Fine Extraction for Target Sentiment Analysis</title>
      <author><first>Peiqin</first><last>Lin</last></author>
      <author><first>Meng</first><last>Yang</last></author>
      <pages>4280–4289</pages>
      <abstract>Target sentiment analysis aims to detect opinion targets along with recognizing their sentiment polarities from a sentence. Some models with span-based labeling have achieved promising results in this task. However, the relation between the target extraction task and the target classification task has not been well exploited. Besides, the span-based target extraction algorithm has a poor performance on target phrases due to the maximum target length setting or length penalty factor. To address these problems, we propose a novel framework of Shared-Private Representation Model (SPRM) with a coarse-to-fine extraction algorithm. For jointly learning target extraction and classification, we design a Shared-Private Network, which encodes not only shared information for both tasks but also private information for each task. To avoid missing correct target phrases, we also propose a heuristic coarse-to-fine extraction algorithm that first gets the approximate interval of the targets by matching the nearest predicted start and end indexes and then extracts the targets by adopting an extending strategy. Experimental results show that our model achieves state-of-the-art performance.</abstract>
      <url hash="b3cf223a">2020.findings-emnlp.382</url>
      <doi>10.18653/v1/2020.findings-emnlp.382</doi>
    </paper>
    <paper id="383">
      <title>Detecting Media Bias in News Articles using <fixed-case>G</fixed-case>aussian Bias Distributions</title>
      <author><first>Wei-Fan</first><last>Chen</last></author>
      <author><first>Khalid</first><last>Al Khatib</last></author>
      <author><first>Benno</first><last>Stein</last></author>
      <author><first>Henning</first><last>Wachsmuth</last></author>
      <pages>4290–4300</pages>
      <abstract>Media plays an important role in shaping public opinion. Biased media can influence people in undesirable directions and hence should be unmasked as such. We observe that feature-based and neural text classification approaches which rely only on the distribution of low-level lexical information fail to detect media bias. This weakness becomes most noticeable for articles on new events, where words appear in new contexts and hence their “bias predictiveness” is unclear. In this paper, we therefore study how second-order information about biased statements in an article helps to improve detection effectiveness. In particular, we utilize the probability distributions of the frequency, positions, and sequential order of lexical and informational sentence-level bias in a Gaussian Mixture Model. On an existing media bias dataset, we find that the frequency and positions of biased statements strongly impact article-level bias, whereas their exact sequential order is secondary. Using a standard model for sentence-level bias detection, we provide empirical evidence that article-level bias detectors that use second-order information clearly outperform those without.</abstract>
      <url hash="12a98434">2020.findings-emnlp.383</url>
      <doi>10.18653/v1/2020.findings-emnlp.383</doi>
    </paper>
    <paper id="384">
      <title>How Can Self-Attention Networks Recognize <fixed-case>D</fixed-case>yck-n Languages?</title>
      <author><first>Javid</first><last>Ebrahimi</last></author>
      <author><first>Dhruv</first><last>Gelda</last></author>
      <author><first>Wei</first><last>Zhang</last></author>
      <pages>4301–4306</pages>
      <abstract>We focus on the recognition of Dyck-n (Dn) languages with self-attention (SA) networks, which has been deemed to be a difficult task for these networks. We compare the performance of two variants of SA, one with a starting symbol (SA+) and one without (SA-). Our results show that SA+ is able to generalize to longer sequences and deeper dependencies. For D2, we find that SA- completely breaks down on long sequences whereas the accuracy of SA+ is 58.82%. We find attention maps learned by SA+ to be amenable to interpretation and compatible with a stack-based language recognizer. Surprisingly, the performance of SA networks is at par with LSTMs, which provides evidence on the ability of SA to learn hierarchies without recursion.</abstract>
      <url hash="df47f4c6">2020.findings-emnlp.384</url>
      <attachment type="OptionalSupplementaryMaterial" hash="1887125a">2020.findings-emnlp.384.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.384</doi>
    </paper>
    <paper id="385">
      <title>Training Flexible Depth Model by Multi-Task Learning for Neural Machine Translation</title>
      <author><first>Qiang</first><last>Wang</last></author>
      <author><first>Tong</first><last>Xiao</last></author>
      <author><first>Jingbo</first><last>Zhu</last></author>
      <pages>4307–4312</pages>
      <abstract>The standard neural machine translation model can only decode with the same depth configuration as training. Restricted by this feature, we have to deploy models of various sizes to maintain the same translation latency, because the hardware conditions on different terminal devices (e.g., mobile phones) may vary greatly. Such individual training leads to increased model maintenance costs and slower model iterations, especially for the industry. In this work, we propose to use multi-task learning to train a flexible depth model that can adapt to different depth configurations during inference. Experimental results show that our approach can simultaneously support decoding in 24 depth configurations and is superior to the individual training and another flexible depth model training method——LayerDrop.</abstract>
      <url hash="716f39c9">2020.findings-emnlp.385</url>
      <attachment type="OptionalSupplementaryMaterial" hash="487c7d98">2020.findings-emnlp.385.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.385</doi>
    </paper>
    <paper id="386">
      <title>Looking inside Noun Compounds: Unsupervised Prepositional and Free Paraphrasing</title>
      <author><first>Girishkumar</first><last>Ponkiya</last></author>
      <author><first>Rudra</first><last>Murthy</last></author>
      <author><first>Pushpak</first><last>Bhattacharyya</last></author>
      <author><first>Girish</first><last>Palshikar</last></author>
      <pages>4313–4323</pages>
      <abstract>A noun compound is a sequence of contiguous nouns that acts as a single noun, although the predicate denoting the semantic relation between its components is dropped. Noun Compound Interpretation is the task of uncovering the relation, in the form of a preposition or a free paraphrase. Prepositional paraphrasing refers to the use of preposition to explain the semantic relation, whereas free paraphrasing refers to invoking an appropriate predicate denoting the semantic relation. In this paper, we propose an unsupervised methodology for these two types of paraphrasing. We use pre-trained contextualized language models to uncover the ‘missing’ words (preposition or predicate). These language models are usually trained to uncover the missing word/words in a given input sentence. Our approach uses templates to prepare the input sequence for the language model. The template uses a special token to indicate the missing predicate. As the model has already been pre-trained to uncover a missing word (or a sequence of words), we exploit it to predict missing words for the input sequence. Our experiments using four datasets show that our unsupervised approach (a) performs comparably to supervised approaches for prepositional paraphrasing, and (b) outperforms supervised approaches for free paraphrasing. Paraphrasing (prepositional or free) using our unsupervised approach is potentially helpful for NLP tasks like machine translation and information extraction.</abstract>
      <url hash="b70bc244">2020.findings-emnlp.386</url>
      <doi>10.18653/v1/2020.findings-emnlp.386</doi>
    </paper>
    <paper id="387">
      <title>The birth of <fixed-case>R</fixed-case>omanian <fixed-case>BERT</fixed-case></title>
      <author><first>Stefan</first><last>Dumitrescu</last></author>
      <author><first>Andrei-Marius</first><last>Avram</last></author>
      <author><first>Sampo</first><last>Pyysalo</last></author>
      <pages>4324–4328</pages>
      <abstract>Large-scale pretrained language models have become ubiquitous in Natural Language Processing. However, most of these models are available either in high-resource languages, in particular English, or as multilingual models that compromise performance on individual languages for coverage. This paper introduces Romanian BERT, the first purely Romanian transformer-based language model, pretrained on a large text corpus. We discuss corpus com-position and cleaning, the model training process, as well as an extensive evaluation of the model on various Romanian datasets. We opensource not only the model itself, but also a repository that contains information on how to obtain the corpus, fine-tune and use this model in production (with practical examples), and how to fully replicate the evaluation process.</abstract>
      <url hash="08fdf270">2020.findings-emnlp.387</url>
      <doi>10.18653/v1/2020.findings-emnlp.387</doi>
    </paper>
    <paper id="388">
      <title><fixed-case>BERT</fixed-case> for Monolingual and Cross-Lingual Reverse Dictionary</title>
      <author><first>Hang</first><last>Yan</last></author>
      <author><first>Xiaonan</first><last>Li</last></author>
      <author><first>Xipeng</first><last>Qiu</last></author>
      <author><first>Bocao</first><last>Deng</last></author>
      <pages>4329–4338</pages>
      <abstract>Reverse dictionary is the task to find the proper target word given the word description. In this paper, we tried to incorporate BERT into this task. However, since BERT is based on the byte-pair-encoding (BPE) subword encoding, it is nontrivial to make BERT generate a word given the description. We propose a simple but effective method to make BERT generate the target word for this specific task. Besides, the cross-lingual reverse dictionary is the task to find the proper target word described in another language. Previous models have to keep two different word embeddings and learn to align these embeddings. Nevertheless, by using the Multilingual BERT (mBERT), we can efficiently conduct the cross-lingual reverse dictionary with one subword embedding, and the alignment between languages is not necessary. More importantly, mBERT can achieve remarkable cross-lingual reverse dictionary performance even without the parallel corpus, which means it can conduct the cross-lingual reverse dictionary with only corresponding monolingual data. Code is publicly available at https://github.com/yhcc/BertForRD.git.</abstract>
      <url hash="6d54f59c">2020.findings-emnlp.388</url>
      <doi>10.18653/v1/2020.findings-emnlp.388</doi>
    </paper>
    <paper id="389">
      <title>What’s so special about <fixed-case>BERT</fixed-case>’s layers? A closer look at the <fixed-case>NLP</fixed-case> pipeline in monolingual and multilingual models</title>
      <author><first>Wietse</first><last>de Vries</last></author>
      <author><first>Andreas</first><last>van Cranenburgh</last></author>
      <author><first>Malvina</first><last>Nissim</last></author>
      <pages>4339–4350</pages>
      <abstract>Peeking into the inner workings of BERT has shown that its layers resemble the classical NLP pipeline, with progressively more complex tasks being concentrated in later layers. To investigate to what extent these results also hold for a language other than English, we probe a Dutch BERT-based model and the multilingual BERT model for Dutch NLP tasks. In addition, through a deeper analysis of part-of-speech tagging, we show that also within a given task, information is spread over different parts of the network and the pipeline might not be as neat as it seems. Each layer has different specialisations, so that it may be more useful to combine information from different layers, instead of selecting a single one based on the best overall performance.</abstract>
      <url hash="2b6f6488">2020.findings-emnlp.389</url>
      <doi>10.18653/v1/2020.findings-emnlp.389</doi>
    </paper>
    <paper id="390">
      <title>Leakage-Adjusted Simulatability: Can Models Generate Non-Trivial Explanations of Their Behavior in Natural Language?</title>
      <author><first>Peter</first><last>Hase</last></author>
      <author><first>Shiyue</first><last>Zhang</last></author>
      <author><first>Harry</first><last>Xie</last></author>
      <author><first>Mohit</first><last>Bansal</last></author>
      <pages>4351–4367</pages>
      <abstract>Data collection for natural language (NL) understanding tasks has increasingly included human explanations alongside data points, allowing past works to introduce models that both perform a task and generate NL explanations for their outputs. Yet to date, model-generated explanations have been evaluated on the basis of surface-level similarities to human explanations, both through automatic metrics like BLEU and human evaluations. We argue that these evaluations are insufficient, since they fail to indicate whether explanations support actual model behavior (faithfulness), rather than simply match what a human would say (plausibility). In this work, we address the problem of evaluating explanations from the the model simulatability perspective. Our contributions are as follows: (1) We introduce a leakage-adjusted simulatability (LAS) metric for evaluating NL explanations, which measures how well explanations help an observer predict a model’s output, while controlling for how explanations can directly leak the output. We use a model as a proxy for a human observer, and validate this choice with two human subject experiments. (2) Using the CoS-E and e-SNLI datasets, we evaluate two existing generative graphical models and two new approaches; one rationalizing method we introduce achieves roughly human-level LAS scores. (3) Lastly, we frame explanation generation as a multi-agent game and optimize explanations for simulatability while penalizing label leakage, which can improve LAS scores.</abstract>
      <url hash="6cc05d30">2020.findings-emnlp.390</url>
      <doi>10.18653/v1/2020.findings-emnlp.390</doi>
    </paper>
    <paper id="391">
      <title>A Pointer Network Architecture for Joint Morphological Segmentation and Tagging</title>
      <author><first>Amit</first><last>Seker</last></author>
      <author><first>Reut</first><last>Tsarfaty</last></author>
      <pages>4368–4378</pages>
      <abstract>Morphologically Rich Languages (MRLs) such as Arabic, Hebrew and Turkish often require Morphological Disambiguation (MD), i.e., the prediction of morphological decomposition of tokens into morphemes, early in the pipeline. Neural MD may be addressed as a simple pipeline, where segmentation is followed by sequence tagging, or as an end-to-end model, predicting morphemes from raw tokens. Both approaches are sub-optimal; the former is heavily prone to error propagation, and the latter does not enjoy explicit access to the basic processing units called morphemes. This paper offers MD architecture that combines the symbolic knowledge of morphemes with the learning capacity of neural end-to-end modeling. We propose a new, general and easy-to-implement Pointer Network model where the input is a morphological lattice and the output is a sequence of indices pointing at a single disambiguated path of morphemes. We demonstrate the efficacy of the model on segmentation and tagging, for Hebrew and Turkish texts, based on their respective Universal Dependencies (UD) treebanks. Our experiments show that with complete lattices, our model outperforms all shared-task results on segmenting and tagging these languages. On the SPMRL treebank, our model outperforms all previously reported results for Hebrew MD in realistic scenarios.</abstract>
      <url hash="9c19e98a">2020.findings-emnlp.391</url>
      <doi>10.18653/v1/2020.findings-emnlp.391</doi>
    </paper>
    <paper id="392">
      <title>Beyond Language: Learning Commonsense from Images for Reasoning</title>
      <author><first>Wanqing</first><last>Cui</last></author>
      <author><first>Yanyan</first><last>Lan</last></author>
      <author><first>Liang</first><last>Pang</last></author>
      <author><first>Jiafeng</first><last>Guo</last></author>
      <author><first>Xueqi</first><last>Cheng</last></author>
      <pages>4379–4389</pages>
      <abstract>This paper proposes a novel approach to learn commonsense from images, instead of limited raw texts or costly constructed knowledge bases, for the commonsense reasoning problem in NLP. Our motivation comes from the fact that an image is worth a thousand words, where richer scene information could be leveraged to help distill the commonsense knowledge, which is often hidden in languages. Our approach, namely Loire, consists of two stages. In the first stage, a bi-modal sequence-to-sequence approach is utilized to conduct the scene layout generation task, based on a text representation model ViBERT. In this way, the required visual scene knowledge, such as spatial relations, will be encoded in ViBERT by the supervised learning process with some bi-modal data like COCO. Then ViBERT is concatenated with a pre-trained language model to perform the downstream commonsense reasoning tasks. Experimental results on two commonsense reasoning problems, i.e.commonsense question answering and pronoun resolution, demonstrate that Loire outperforms traditional language-based methods. We also give some case studies to show what knowledge is learned from images and explain how the generated scene layout helps the commonsense reasoning process.</abstract>
      <url hash="ada80ba7">2020.findings-emnlp.392</url>
      <doi>10.18653/v1/2020.findings-emnlp.392</doi>
    </paper>
    <paper id="393">
      <title>A <fixed-case>BERT</fixed-case>-based Distractor Generation Scheme with Multi-tasking and Negative Answer Training Strategies.</title>
      <author><first>Ho-Lam</first><last>Chung</last></author>
      <author><first>Ying-Hong</first><last>Chan</last></author>
      <author><first>Yao-Chung</first><last>Fan</last></author>
      <pages>4390–4400</pages>
      <abstract>In this paper, we investigate the following two limitations for the existing distractor generation (DG) methods. First, the quality of the existing DG methods are still far from practical use. There are still room for DG quality improvement. Second, the existing DG designs are mainly for single distractor generation. However, for practical MCQ preparation, multiple distractors are desired. Aiming at these goals, in this paper, we present a new distractor generation scheme with multi-tasking and negative answer training strategies for effectively generating <i>multiple</i> distractors. The experimental results show that (1) our model advances the state-of-the-art result from 28.65 to 39.81 (BLEU 1 score) and (2) the generated multiple distractors are diverse and shows strong distracting power for multiple choice question.</abstract>
      <url hash="a2e07684">2020.findings-emnlp.393</url>
      <doi>10.18653/v1/2020.findings-emnlp.393</doi>
    </paper>
    <paper id="394">
      <title>How Effective is Task-Agnostic Data Augmentation for Pretrained Transformers?</title>
      <author><first>Shayne</first><last>Longpre</last></author>
      <author><first>Yu</first><last>Wang</last></author>
      <author><first>Chris</first><last>DuBois</last></author>
      <pages>4401–4411</pages>
      <abstract>Task-agnostic forms of data augmentation have proven widely effective in computer vision, even on pretrained models. In NLP similar results are reported most commonly for low data regimes, non-pretrained models, or situationally for pretrained models. In this paper we ask how effective these techniques really are when applied to pretrained transformers. Using two popular varieties of task-agnostic data augmentation (not tailored to any particular task), Easy Data Augmentation (Wei andZou, 2019) and Back-Translation (Sennrichet al., 2015), we conduct a systematic examination of their effects across 5 classification tasks, 6 datasets, and 3 variants of modern pretrained transformers, including BERT, XLNet, and RoBERTa. We observe a negative result, finding that techniques which previously reported strong improvements for non-pretrained models fail to consistently improve performance for pretrained transformers, even when training data is limited. We hope this empirical analysis helps inform practitioners where data augmentation techniques may confer improvements.</abstract>
      <url hash="b8f085ab">2020.findings-emnlp.394</url>
      <doi>10.18653/v1/2020.findings-emnlp.394</doi>
    </paper>
    <paper id="395">
      <title>Visually-Grounded Planning without Vision: Language Models Infer Detailed Plans from High-level Instructions</title>
      <author><first>Peter</first><last>Jansen</last></author>
      <pages>4412–4417</pages>
      <abstract>The recently proposed ALFRED challenge task aims for a virtual robotic agent to complete complex multi-step everyday tasks in a virtual home environment from high-level natural language directives, such as “put a hot piece of bread on a plate”. Currently, the best-performing models are able to complete less than 1% of these tasks successfully. In this work we focus on modeling the translation problem of converting natural language directives into detailed multi-step sequences of actions that accomplish those goals in the virtual environment. We empirically demonstrate that it is possible to generate gold multi-step plans from language directives alone without any visual input in 26% of unseen cases. When a small amount of visual information, the starting location in the virtual environment, is incorporated, our best-performing GPT-2 model successfully generates gold command sequences in 58% of cases, suggesting contextualized language models may provide strong planning modules for grounded virtual agents.</abstract>
      <url hash="28f01fc2">2020.findings-emnlp.395</url>
      <doi>10.18653/v1/2020.findings-emnlp.395</doi>
    </paper>
    <paper id="396">
      <title>Consistent Response Generation with Controlled Specificity</title>
      <author><first>Junya</first><last>Takayama</last></author>
      <author><first>Yuki</first><last>Arase</last></author>
      <pages>4418–4427</pages>
      <abstract>We propose a method to control the specificity of responses while maintaining the consistency with the utterances. We first design a metric based on pointwise mutual information, which measures the co-occurrence degree between an utterance and a response. To control the specificity of generated responses, we add the distant supervision based on the co-occurrence degree and a PMI-based word prediction mechanism to a sequence-to-sequence model. With these mechanisms, our model outputs the words with optimal specificity for a given specificity control variable. In experiments with open-domain dialogue corpora, automatic and human evaluation results confirm that our model controls the specificity of the response more sensitively than the conventional model and can generate highly consistent responses.</abstract>
      <url hash="43d71ebe">2020.findings-emnlp.396</url>
      <doi>10.18653/v1/2020.findings-emnlp.396</doi>
    </paper>
    <paper id="397">
      <title>Internal and external pressures on language emergence: least effort, object constancy and frequency</title>
      <author><first>Diana</first><last>Rodríguez Luna</last></author>
      <author><first>Edoardo Maria</first><last>Ponti</last></author>
      <author><first>Dieuwke</first><last>Hupkes</last></author>
      <author><first>Elia</first><last>Bruni</last></author>
      <pages>4428–4437</pages>
      <abstract>In previous work, artificial agents were shown to achieve almost perfect accuracy in referential games where they have to communicate to identify images. Nevertheless, the resulting communication protocols rarely display salient features of natural languages, such as compositionality. In this paper, we propose some realistic sources of pressure on communication that avert this outcome. More specifically, we formalise the principle of least effort through an auxiliary objective. Moreover, we explore several game variants, inspired by the principle of object constancy, in which we alter the frequency, position, and luminosity of the objects in the images. We perform an extensive analysis on their effect through compositionality metrics, diagnostic classifiers, and zero-shot evaluation. Our findings reveal that the proposed sources of pressure result in emerging languages with less redundancy, more focus on high-level conceptual information, and better abilities of generalisation. Overall, our contributions reduce the gap between emergent and natural languages.</abstract>
      <url hash="b36d88e3">2020.findings-emnlp.397</url>
      <doi>10.18653/v1/2020.findings-emnlp.397</doi>
    </paper>
    <paper id="398">
      <title>Parsing All: Syntax and Semantics, Dependencies and Spans</title>
      <author><first>Junru</first><last>Zhou</last></author>
      <author><first>Zuchao</first><last>Li</last></author>
      <author><first>Hai</first><last>Zhao</last></author>
      <pages>4438–4449</pages>
      <abstract>Both syntactic and semantic structures are key linguistic contextual clues, in which parsing the latter has been well shown beneficial from parsing the former. However, few works ever made an attempt to let semantic parsing help syntactic parsing. As linguistic representation formalisms, both syntax and semantics may be represented in either span (constituent/phrase) or dependency, on both of which joint learning was also seldom explored. In this paper, we propose a novel joint model of syntactic and semantic parsing on both span and dependency representations, which incorporates syntactic information effectively in the encoder of neural network and benefits from two representation formalisms in a uniform way. The experiments show that semantics and syntax can benefit each other by optimizing joint objectives. Our single model achieves new state-of-the-art or competitive results on both span and dependency semantic parsing on Propbank benchmarks and both dependency and constituent syntactic parsing on Penn Treebank.</abstract>
      <url hash="69961a6d">2020.findings-emnlp.398</url>
      <attachment type="OptionalSupplementaryMaterial" hash="78a80c61">2020.findings-emnlp.398.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.398</doi>
    </paper>
    <paper id="399">
      <title><fixed-case>LIMIT</fixed-case>-<fixed-case>BERT</fixed-case> : Linguistics Informed Multi-Task <fixed-case>BERT</fixed-case></title>
      <author><first>Junru</first><last>Zhou</last></author>
      <author><first>Zhuosheng</first><last>Zhang</last></author>
      <author><first>Hai</first><last>Zhao</last></author>
      <author><first>Shuailiang</first><last>Zhang</last></author>
      <pages>4450–4461</pages>
      <abstract>In this paper, we present Linguistics Informed Multi-Task BERT (LIMIT-BERT) for learning language representations across multiple linguistics tasks by Multi-Task Learning. LIMIT-BERT includes five key linguistics tasks: Part-Of-Speech (POS) tags, constituent and dependency syntactic parsing, span and dependency semantic role labeling (SRL). Different from recent Multi-Task Deep Neural Networks (MT-DNN), our LIMIT-BERT is fully linguistics motivated and thus is capable of adopting an improved masked training objective according to syntactic and semantic constituents. Besides, LIMIT-BERT takes a semi-supervised learning strategy to offer the same large amount of linguistics task data as that for the language model training. As a result, LIMIT-BERT not only improves linguistics tasks performance but also benefits from a regularization effect and linguistics information that leads to more general representations to help adapt to new tasks and domains. LIMIT-BERT outperforms the strong baseline Whole Word Masking BERT on both dependency and constituent syntactic/semantic parsing, GLUE benchmark, and SNLI task. Our practice on the proposed LIMIT-BERT also enables us to release a well pre-trained model for multi-purpose of natural language processing tasks once for all.</abstract>
      <url hash="f8800b4b">2020.findings-emnlp.399</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f11db6a8">2020.findings-emnlp.399.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.399</doi>
    </paper>
    <paper id="400">
      <title>Improving Limited Labeled Dialogue State Tracking with Self-Supervision</title>
      <author><first>Chien-Sheng</first><last>Wu</last></author>
      <author><first>Steven C.H.</first><last>Hoi</last></author>
      <author><first>Caiming</first><last>Xiong</last></author>
      <pages>4462–4472</pages>
      <abstract>Existing dialogue state tracking (DST) models require plenty of labeled data. However, collecting high-quality labels is costly, especially when the number of domains increases. In this paper, we address a practical DST problem that is rarely discussed, i.e., learning efficiently with limited labeled data. We present and investigate two self-supervised objectives: preserving latent consistency and modeling conversational behavior. We encourage a DST model to have consistent latent distributions given a perturbed input, making it more robust to an unseen scenario. We also add an auxiliary utterance generation task, modeling a potential correlation between conversational behavior and dialogue states. The experimental results show that our proposed self-supervised signals can improve joint goal accuracy by 8.95% when only 1% labeled data is used on the MultiWOZ dataset. We can achieve an additional 1.76% improvement if some unlabeled data is jointly trained as semi-supervised learning. We analyze and visualize how our proposed self-supervised signals help the DST task and hope to stimulate future data-efficient DST research.</abstract>
      <url hash="193ddb5b">2020.findings-emnlp.400</url>
      <attachment type="OptionalSupplementaryMaterial" hash="dd11f032">2020.findings-emnlp.400.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.400</doi>
    </paper>
    <paper id="401">
      <title>On the Branching Bias of Syntax Extracted from Pre-trained Language Models</title>
      <author><first>Huayang</first><last>Li</last></author>
      <author><first>Lemao</first><last>Liu</last></author>
      <author><first>Guoping</first><last>Huang</last></author>
      <author><first>Shuming</first><last>Shi</last></author>
      <pages>4473–4478</pages>
      <abstract>Many efforts have been devoted to extracting constituency trees from pre-trained language models, often proceeding in two stages: feature definition and parsing. However, this kind of methods may suffer from the branching bias issue, which will inflate the performances on languages with the same branch it biases to. In this work, we propose quantitatively measuring the branching bias by comparing the performance gap on a language and its reversed language, which is agnostic to both language models and extracting methods. Furthermore, we analyze the impacts of three factors on the branching bias, namely feature definitions, parsing algorithms, and language models. Experiments show that several existing works exhibit branching biases, and some implementations of these three factors can introduce the branching bias.</abstract>
      <url hash="e36279aa">2020.findings-emnlp.401</url>
      <doi>10.18653/v1/2020.findings-emnlp.401</doi>
    </paper>
    <paper id="402">
      <title><fixed-case>T</fixed-case>he <fixed-case>P</fixed-case>ragmatics behind <fixed-case>P</fixed-case>olitics: <fixed-case>M</fixed-case>odelling <fixed-case>M</fixed-case>etaphor, <fixed-case>F</fixed-case>raming and <fixed-case>E</fixed-case>motion in <fixed-case>P</fixed-case>olitical <fixed-case>D</fixed-case>iscourse</title>
      <author><first>Pere-Lluís</first><last>Huguet Cabot</last></author>
      <author><first>Verna</first><last>Dankers</last></author>
      <author><first>David</first><last>Abadi</last></author>
      <author><first>Agneta</first><last>Fischer</last></author>
      <author><first>Ekaterina</first><last>Shutova</last></author>
      <pages>4479–4488</pages>
      <abstract>There has been an increased interest in modelling political discourse within the natural language processing (NLP) community, in tasks such as political bias and misinformation detection, among others. Metaphor-rich and emotion-eliciting communication strategies are ubiquitous in political rhetoric, according to social science research. Yet, none of the existing computational models of political discourse has incorporated these phenomena. In this paper, we present the first joint models of metaphor, emotion and political rhetoric, and demonstrate that they advance performance in three tasks: predicting political perspective of news articles, party affiliation of politicians and framing of policy issues.</abstract>
      <url hash="4e5c874e">2020.findings-emnlp.402</url>
      <doi>10.18653/v1/2020.findings-emnlp.402</doi>
    </paper>
    <paper id="403">
      <title><fixed-case>SMRT</fixed-case> Chatbots: <fixed-case>I</fixed-case>mproving Non-Task-Oriented Dialog with <fixed-case>S</fixed-case>imulated <fixed-case>M</fixed-case>ultiple <fixed-case>R</fixed-case>eference <fixed-case>T</fixed-case>raining</title>
      <author><first>Huda</first><last>Khayrallah</last></author>
      <author><first>João</first><last>Sedoc</last></author>
      <pages>4489–4505</pages>
      <abstract>Non-task-oriented dialog models suffer from poor quality and non-diverse responses. To overcome limited conversational data, we apply Simulated Multiple Reference Training (SMRT; Khayrallah et al., 2020), and use a paraphraser to simulate multiple responses per training prompt. We find SMRT improves over a strong Transformer baseline as measured by human and automatic quality scores and lexical diversity. We also find SMRT is comparable to pretraining in human evaluation quality, and outperforms pretraining on automatic quality and lexical diversity, without requiring related-domain dialog data.</abstract>
      <url hash="d89eac9d">2020.findings-emnlp.403</url>
      <doi>10.18653/v1/2020.findings-emnlp.403</doi>
    </paper>
    <paper id="404">
      <title><fixed-case>P</fixed-case>riv<fixed-case>N</fixed-case>et: Safeguarding Private Attributes in Transfer Learning for Recommendation</title>
      <author><first>Guangneng</first><last>Hu</last></author>
      <author><first>Qiang</first><last>Yang</last></author>
      <pages>4506–4516</pages>
      <abstract>Transfer learning is an effective technique to improve a target recommender system with the knowledge from a source domain. Existing research focuses on the recommendation performance of the target domain while ignores the privacy leakage of the source domain. The transferred knowledge, however, may unintendedly leak private information of the source domain. For example, an attacker can accurately infer user demographics from their historical purchase provided by a source domain data owner. This paper addresses the above privacy-preserving issue by learning a privacy-aware neural representation by improving target performance while protecting source privacy. The key idea is to simulate the attacks during the training for protecting unseen users’ privacy in the future, modeled by an adversarial game, so that the transfer learning model becomes robust to attacks. Experiments show that the proposed PrivNet model can successfully disentangle the knowledge benefitting the transfer from leaking the privacy.</abstract>
      <url hash="b203930d">2020.findings-emnlp.404</url>
      <attachment type="OptionalSupplementaryMaterial" hash="1a931fda">2020.findings-emnlp.404.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.404</doi>
    </paper>
    <paper id="405">
      <title>Learning to Learn to Disambiguate: <fixed-case>M</fixed-case>eta-Learning for Few-Shot Word Sense Disambiguation</title>
      <author><first>Nithin</first><last>Holla</last></author>
      <author><first>Pushkar</first><last>Mishra</last></author>
      <author><first>Helen</first><last>Yannakoudakis</last></author>
      <author><first>Ekaterina</first><last>Shutova</last></author>
      <pages>4517–4533</pages>
      <abstract>The success of deep learning methods hinges on the availability of large training datasets annotated for the task of interest. In contrast to human intelligence, these methods lack versatility and struggle to learn and adapt quickly to new tasks, where labeled data is scarce. Meta-learning aims to solve this problem by training a model on a large number of few-shot tasks, with an objective to learn new tasks quickly from a small number of examples. In this paper, we propose a meta-learning framework for few-shot word sense disambiguation (WSD), where the goal is to learn to disambiguate unseen words from only a few labeled instances. Meta-learning approaches have so far been typically tested in an N-way, K-shot classification setting where each task has N classes with K examples per class. Owing to its nature, WSD deviates from this controlled setup and requires the models to handle a large number of highly unbalanced classes. We extend several popular meta-learning approaches to this scenario, and analyze their strengths and weaknesses in this new challenging setting.</abstract>
      <url hash="a2e24d6d">2020.findings-emnlp.405</url>
      <doi>10.18653/v1/2020.findings-emnlp.405</doi>
    </paper>
    <paper id="406">
      <title>An Empirical Investigation of Beam-Aware Training in Supertagging</title>
      <author><first>Renato</first><last>Negrinho</last></author>
      <author><first>Matthew R.</first><last>Gormley</last></author>
      <author><first>Geoff</first><last>Gordon</last></author>
      <pages>4534–4542</pages>
      <abstract>Structured prediction is often approached by training a locally normalized model with maximum likelihood and decoding approximately with beam search. This approach leads to mismatches as, during training, the model is not exposed to its mistakes and does not use beam search. Beam-aware training aims to address these problems, but unfortunately, it is not yet widely used due to a lack of understanding about how it impacts performance, when it is most useful, and whether it is stable. Recently, Negrinho et al. (2018) proposed a meta-algorithm that captures beam-aware training algorithms and suggests new ones, but unfortunately did not provide empirical results. In this paper, we begin an empirical investigation: we train the supertagging model of Vaswani et al. (2018) and a simpler model with instantiations of the meta-algorithm. We explore the influence of various design choices and make recommendations for choosing them. We observe that beam-aware training improves performance for both models, with large improvements for the simpler model which must effectively manage uncertainty during decoding. Our results suggest that a model must be learned with search to maximize its effectiveness.</abstract>
      <url hash="69c33efb">2020.findings-emnlp.406</url>
      <attachment type="OptionalSupplementaryMaterial" hash="c6237e2e">2020.findings-emnlp.406.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.406</doi>
    </paper>
    <paper id="407">
      <title>Improving Aspect-based Sentiment Analysis with Gated Graph Convolutional Networks and Syntax-based Regulation</title>
      <author><first>Amir</first><last>Pouran Ben Veyseh</last></author>
      <author><first>Nasim</first><last>Nouri</last></author>
      <author><first>Franck</first><last>Dernoncourt</last></author>
      <author><first>Quan Hung</first><last>Tran</last></author>
      <author><first>Dejing</first><last>Dou</last></author>
      <author><first>Thien Huu</first><last>Nguyen</last></author>
      <pages>4543–4548</pages>
      <abstract>Aspect-based Sentiment Analysis (ABSA) seeks to predict the sentiment polarity of a sentence toward a specific aspect. Recently, it has been shown that dependency trees can be integrated into deep learning models to produce the state-of-the-art performance for ABSA. However, these models tend to compute the hidden/representation vectors without considering the aspect terms and fail to benefit from the overall contextual importance scores of the words that can be obtained from the dependency tree for ABSA. In this work, we propose a novel graph-based deep learning model to overcome these two issues of the prior work on ABSA. In our model, gate vectors are generated from the representation vectors of the aspect terms to customize the hidden vectors of the graph-based models toward the aspect terms. In addition, we propose a mechanism to obtain the importance scores for each word in the sentences based on the dependency trees that are then injected into the model to improve the representation vectors for ABSA. The proposed model achieves the state-of-the-art performance on three benchmark datasets.</abstract>
      <url hash="e2396da9">2020.findings-emnlp.407</url>
      <doi>10.18653/v1/2020.findings-emnlp.407</doi>
    </paper>
    <paper id="408">
      <title><fixed-case>D</fixed-case>ecoding Language Spatial Relations to 2<fixed-case>D</fixed-case> Spatial Arrangements</title>
      <author><first>Gorjan</first><last>Radevski</last></author>
      <author><first>Guillem</first><last>Collell</last></author>
      <author><first>Marie-Francine</first><last>Moens</last></author>
      <author><first>Tinne</first><last>Tuytelaars</last></author>
      <pages>4549–4560</pages>
      <abstract>We address the problem of multimodal spatial understanding by decoding a set of language-expressed spatial relations to a set of 2D spatial arrangements in a multi-object and multi-relationship setting. We frame the task as arranging a scene of clip-arts given a textual description. We propose a simple and effective model architecture Spatial-Reasoning Bert (SR-Bert), trained to decode text to 2D spatial arrangements in a non-autoregressive manner. SR-Bert can decode both explicit and implicit language to 2D spatial arrangements, generalizes to out-of-sample data to a reasonable extent and can generate complete abstract scenes if paired with a clip-arts predictor. Finally, we qualitatively evaluate our method with a user study, validating that our generated spatial arrangements align with human expectation.</abstract>
      <url hash="7d2223ae">2020.findings-emnlp.408</url>
      <attachment type="OptionalSupplementaryMaterial" hash="26c48b07">2020.findings-emnlp.408.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.408</doi>
    </paper>
    <paper id="409">
      <title>The Dots Have Their Values: Exploiting the Node-Edge Connections in Graph-based Neural Models for Document-level Relation Extraction</title>
      <author><first>Hieu</first><last>Minh Tran</last></author>
      <author><first>Minh Trung</first><last>Nguyen</last></author>
      <author><first>Thien Huu</first><last>Nguyen</last></author>
      <pages>4561–4567</pages>
      <abstract>The goal of Document-level Relation Extraction (DRE) is to recognize the relations between entity mentions that can span beyond sentence boundary. The current state-of-the-art method for this problem has involved the graph-based edge-oriented model where the entity mentions, entities, and sentences in the documents are used as the nodes of the document graphs for representation learning. However, this model does not capture the representations for the nodes in the graphs, thus preventing it from effectively encoding the specific and relevant information of the nodes for DRE. To address this issue, we propose to explicitly compute the representations for the nodes in the graph-based edge-oriented model for DRE. These node representations allow us to introduce two novel representation regularization mechanisms to improve the representation vectors for DRE. The experiments show that our model achieves state-of-the-art performance on two benchmark datasets.</abstract>
      <url hash="4ac0ac3d">2020.findings-emnlp.409</url>
      <doi>10.18653/v1/2020.findings-emnlp.409</doi>
    </paper>
    <paper id="410">
      <title>Why and when should you pool? Analyzing Pooling in Recurrent Architectures</title>
      <author><first>Pratyush</first><last>Maini</last></author>
      <author><first>Keshav</first><last>Kolluru</last></author>
      <author><first>Danish</first><last>Pruthi</last></author>
      <author><first/><last>Mausam</last></author>
      <pages>4568–4586</pages>
      <abstract>Pooling-based recurrent neural architectures consistently outperform their counterparts without pooling on sequence classification tasks. However, the reasons for their enhanced performance are largely unexamined. In this work, we examine three commonly used pooling techniques (mean-pooling, max-pooling, and attention, and propose *max-attention*, a novel variant that captures interactions among predictive tokens in a sentence. Using novel experiments, we demonstrate that pooling architectures substantially differ from their non-pooling equivalents in their learning ability and positional biases: (i) pooling facilitates better gradient flow than BiLSTMs in initial training epochs, and (ii) BiLSTMs are biased towards tokens at the beginning and end of the input, whereas pooling alleviates this bias. Consequently, we find that pooling yields large gains in low resource scenarios, and instances when salient words lie towards the middle of the input. Across several text classification tasks, we find max-attention to frequently outperform other pooling techniques.</abstract>
      <url hash="a35d5cd2">2020.findings-emnlp.410</url>
      <doi>10.18653/v1/2020.findings-emnlp.410</doi>
    </paper>
    <paper id="411">
      <title>Structural and Functional Decomposition for Personality Image Captioning in a Communication Game</title>
      <author><first>Minh Thu</first><last>Nguyen</last></author>
      <author><first>Duy</first><last>Phung</last></author>
      <author><first>Minh</first><last>Hoai</last></author>
      <author><first>Thien Huu</first><last>Nguyen</last></author>
      <pages>4587–4593</pages>
      <abstract>Personality image captioning (PIC) aims to describe an image with a natural language caption given a personality trait. In this work, we introduce a novel formulation for PIC based on a communication game between a speaker and a listener. The speaker attempts to generate natural language captions while the listener encourages the generated captions to contain discriminative information about the input images and personality traits. In this way, we expect that the generated captions can be improved to naturally represent the images and express the traits. In addition, we propose to adapt the language model GPT2 to perform caption generation for PIC. This enables the speaker and listener to benefit from the language encoding capacity of GPT2. Our experiments show that the proposed model achieves the state-of-the-art performance for PIC.</abstract>
      <url hash="f784a546">2020.findings-emnlp.411</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f7d27a9e">2020.findings-emnlp.411.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.411</doi>
    </paper>
    <paper id="412">
      <title>Long Document Ranking with Query-Directed Sparse Transformer</title>
      <author><first>Jyun-Yu</first><last>Jiang</last></author>
      <author><first>Chenyan</first><last>Xiong</last></author>
      <author><first>Chia-Jung</first><last>Lee</last></author>
      <author><first>Wei</first><last>Wang</last></author>
      <pages>4594–4605</pages>
      <abstract>The computing cost of transformer self-attention often necessitates breaking long documents to fit in pretrained models in document ranking tasks. In this paper, we design Query-Directed Sparse attention that induces IR-axiomatic structures in transformer self-attention. Our model, QDS-Transformer, enforces the principle properties desired in ranking: local contextualization, hierarchical representation, and query-oriented proximity matching, while it also enjoys efficiency from sparsity. Experiments on four fully supervised and few-shot TREC document ranking benchmarks demonstrate the consistent and robust advantage of QDS-Transformer over previous approaches, as they either retrofit long documents into BERT or use sparse attention without emphasizing IR principles. We further quantify the computing complexity and demonstrates that our sparse attention with TVM implementation is twice more efficient that the fully-connected self-attention. All source codes, trained model, and predictions of this work are available at https://github.com/hallogameboy/QDS-Transformer.</abstract>
      <url hash="4ef8d0bd">2020.findings-emnlp.412</url>
      <doi>10.18653/v1/2020.findings-emnlp.412</doi>
    </paper>
    <paper id="413">
      <title>Visuo-Linguistic Question Answering (<fixed-case>VLQA</fixed-case>) Challenge</title>
      <author><first>Shailaja Keyur</first><last>Sampat</last></author>
      <author><first>Yezhou</first><last>Yang</last></author>
      <author><first>Chitta</first><last>Baral</last></author>
      <pages>4606–4616</pages>
      <abstract>Understanding images and text together is an important aspect of cognition and building advanced Artificial Intelligence (AI) systems. As a community, we have achieved good benchmarks over language and vision domains separately, however joint reasoning is still a challenge for state-of-the-art computer vision and natural language processing (NLP) systems. We propose a novel task to derive joint inference about a given image-text modality and compile the Visuo-Linguistic Question Answering (VLQA) challenge corpus in a question answering setting. Each dataset item consists of an image and a reading passage, where questions are designed to combine both visual and textual information i.e., ignoring either modality would make the question unanswerable. We first explore the best existing vision-language architectures to solve VLQA subsets and show that they are unable to reason well. We then develop a modular method with slightly better baseline performance, but it is still far behind human performance. We believe that VLQA will be a good benchmark for reasoning over a visuo-linguistic context. The dataset, code and leaderboard is available at https://shailaja183.github.io/vlqa/.</abstract>
      <url hash="9cefea9c">2020.findings-emnlp.413</url>
      <attachment type="OptionalSupplementaryMaterial" hash="4b0644d5">2020.findings-emnlp.413.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.413</doi>
    </paper>
    <paper id="414">
      <title>Byte Pair Encoding is Suboptimal for Language Model Pretraining</title>
      <author><first>Kaj</first><last>Bostrom</last></author>
      <author><first>Greg</first><last>Durrett</last></author>
      <pages>4617–4624</pages>
      <abstract>The success of pretrained transformer language models (LMs) in natural language processing has led to a wide range of pretraining setups. In particular, these models employ a variety of subword tokenization methods, most notably byte-pair encoding (BPE) (Sennrich et al., 2016; Gage, 1994), the WordPiece method (Schuster and Nakajima, 2012), and unigram language modeling (Kudo, 2018), to segment text. However, to the best of our knowledge, the literature does not contain a direct evaluation of the impact of tokenization on language model pretraining. We analyze differences between BPE and unigram LM tokenization, finding that the latter method recovers subword units that align more closely with morphology and avoids problems stemming from BPE’s greedy construction procedure. We then compare the fine-tuned task performance of identical transformer masked language models pretrained with these tokenizations. Across downstream tasks and two languages (English and Japanese), we find that the unigram LM tokenization method matches or outperforms BPE. We hope that developers of future pretrained LMs will consider adopting the unigram LM method over the more prevalent BPE.</abstract>
      <url hash="5d0e9799">2020.findings-emnlp.414</url>
      <doi>10.18653/v1/2020.findings-emnlp.414</doi>
    </paper>
    <paper id="415">
      <title>Exploring <fixed-case>BERT</fixed-case>’s Sensitivity to Lexical Cues using Tests from Semantic Priming</title>
      <author><first>Kanishka</first><last>Misra</last></author>
      <author><first>Allyson</first><last>Ettinger</last></author>
      <author><first>Julia</first><last>Rayz</last></author>
      <pages>4625–4635</pages>
      <abstract>Models trained to estimate word probabilities in context have become ubiquitous in natural language processing. How do these models use lexical cues in context to inform their word probabilities? To answer this question, we present a case study analyzing the pre-trained BERT model with tests informed by semantic priming. Using English lexical stimuli that show priming in humans, we find that BERT too shows “priming”, predicting a word with greater probability when the context includes a related word versus an unrelated one. This effect decreases as the amount of information provided by the context increases. Follow-up analysis shows BERT to be increasingly distracted by related prime words as context becomes more informative, assigning lower probabilities to related words. Our findings highlight the importance of considering contextual constraint effects when studying word prediction in these models, and highlight possible parallels with human processing.</abstract>
      <url hash="7c27b733">2020.findings-emnlp.415</url>
      <doi>10.18653/v1/2020.findings-emnlp.415</doi>
    </paper>
    <paper id="416">
      <title>Multi-hop Question Generation with Graph Convolutional Network</title>
      <author><first>Dan</first><last>Su</last></author>
      <author><first>Yan</first><last>Xu</last></author>
      <author><first>Wenliang</first><last>Dai</last></author>
      <author><first>Ziwei</first><last>Ji</last></author>
      <author><first>Tiezheng</first><last>Yu</last></author>
      <author><first>Pascale</first><last>Fung</last></author>
      <pages>4636–4647</pages>
      <abstract>Multi-hop Question Generation (QG) aims to generate answer-related questions by aggregating and reasoning over multiple scattered evidence from different paragraphs. It is a more challenging yet under-explored task compared to conventional single-hop QG, where the questions are generated from the sentence containing the answer or nearby sentences in the same paragraph without complex reasoning. To address the additional challenges in multi-hop QG, we propose Multi-Hop Encoding Fusion Network for Question Generation (MulQG), which does context encoding in multiple hops with Graph Convolutional Network and encoding fusion via an Encoder Reasoning Gate. To the best of our knowledge, we are the first to tackle the challenge of multi-hop reasoning over paragraphs without any sentence-level information. Empirical results on HotpotQA dataset demonstrate the effectiveness of our method, in comparison with baselines on automatic evaluation metrics. Moreover, from the human evaluation, our proposed model is able to generate fluent questions with high completeness and outperforms the strongest baseline by 20.8% in the multi-hop evaluation. on. The code is publicly availableat https://github.com/HLTCHKU</abstract>
      <url hash="d134f0f8">2020.findings-emnlp.416</url>
      <doi>10.18653/v1/2020.findings-emnlp.416</doi>
    </paper>
    <paper id="417">
      <title><fixed-case>MMFT-BERT</fixed-case>: <fixed-case>M</fixed-case>ultimodal <fixed-case>F</fixed-case>usion <fixed-case>T</fixed-case>ransformer with <fixed-case>B</fixed-case><fixed-case>ERT</fixed-case> <fixed-case>E</fixed-case>ncodings for <fixed-case>V</fixed-case>isual <fixed-case>Q</fixed-case>uestion <fixed-case>A</fixed-case>nswering</title>
      <author><first>Aisha</first><last>Urooj</last></author>
      <author><first>Amir</first><last>Mazaheri</last></author>
      <author><first>Niels</first><last>Da vitoria lobo</last></author>
      <author><first>Mubarak</first><last>Shah</last></author>
      <pages>4648–4660</pages>
      <abstract>We present MMFT-BERT(MultiModal FusionTransformer with BERT encodings), to solve Visual Question Answering (VQA) ensuring individual and combined processing of multiple input modalities. Our approach benefits from processing multimodal data (video and text) adopting the BERT encodings individually and using a novel transformer-based fusion method to fuse them together. Our method decomposes the different sources of modalities, into different BERT instances with similar architectures, but variable weights. This achieves SOTA results on the TVQA dataset. Additionally, we provide TVQA-Visual, an isolated diagnostic subset of TVQA, which strictly requires the knowledge of visual (V) modality based on a human annotator’s judgment. This set of questions helps us to study the model’s behavior and the challenges TVQA poses to prevent the achievement of super human performance. Extensive experiments show the effectiveness and superiority of our method.</abstract>
      <url hash="89d8711d">2020.findings-emnlp.417</url>
      <doi>10.18653/v1/2020.findings-emnlp.417</doi>
    </paper>
    <paper id="418">
      <title>Thinking Like a Skeptic: Defeasible Inference in Natural Language</title>
      <author><first>Rachel</first><last>Rudinger</last></author>
      <author><first>Vered</first><last>Shwartz</last></author>
      <author><first>Jena D.</first><last>Hwang</last></author>
      <author><first>Chandra</first><last>Bhagavatula</last></author>
      <author><first>Maxwell</first><last>Forbes</last></author>
      <author><first>Ronan</first><last>Le Bras</last></author>
      <author><first>Noah A.</first><last>Smith</last></author>
      <author><first>Yejin</first><last>Choi</last></author>
      <pages>4661–4675</pages>
      <abstract>Defeasible inference is a mode of reasoning in which an inference (X is a bird, therefore X flies) may be weakened or overturned in light of new evidence (X is a penguin). Though long recognized in classical AI and philosophy, defeasible inference has not been extensively studied in the context of contemporary data-driven research on natural language inference and commonsense reasoning. We introduce Defeasible NLI (abbreviated <tex-math>\delta</tex-math>-NLI), a dataset for defeasible inference in natural language. Defeasible NLI contains extensions to three existing inference datasets covering diverse modes of reasoning: common sense, natural language inference, and social norms. From Defeasible NLI, we develop both a classification and generation task for defeasible inference, and demonstrate that the generation task is much more challenging. Despite lagging human performance, however, generative models trained on this data are capable of writing sentences that weaken or strengthen a specified inference up to 68% of the time.</abstract>
      <url hash="c7dccb48">2020.findings-emnlp.418</url>
      <doi>10.18653/v1/2020.findings-emnlp.418</doi>
    </paper>
    <paper id="419">
      <title>Guiding Attention for Self-Supervised Learning with Transformers</title>
      <author><first>Ameet</first><last>Deshpande</last></author>
      <author><first>Karthik</first><last>Narasimhan</last></author>
      <pages>4676–4686</pages>
      <abstract>In this paper, we propose a simple and effective technique to allow for efficient self-supervised learning with bi-directional Transformers. Our approach is motivated by recent studies demonstrating that self-attention patterns in trained models contain a majority of non-linguistic regularities. We propose a computationally efficient auxiliary loss function to guide attention heads to conform to such patterns. Our method is agnostic to the actual pre-training objective and results in faster convergence of models as well as better performance on downstream tasks compared to the baselines, achieving state of the art results in low-resource settings. Surprisingly, we also find that linguistic properties of attention heads are not necessarily correlated with language modeling performance.</abstract>
      <url hash="6870d6ac">2020.findings-emnlp.419</url>
      <attachment type="OptionalSupplementaryMaterial" hash="2dd05b7c">2020.findings-emnlp.419.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.419</doi>
    </paper>
    <paper id="420">
      <title>Language-<fixed-case>C</fixed-case>onditioned <fixed-case>F</fixed-case>eature <fixed-case>P</fixed-case>yramids for <fixed-case>V</fixed-case>isual <fixed-case>S</fixed-case>election <fixed-case>T</fixed-case>asks</title>
      <author><first>Taichi</first><last>Iki</last></author>
      <author><first>Akiko</first><last>Aizawa</last></author>
      <pages>4687–4697</pages>
      <abstract>Referring expression comprehension, which is the ability to locate language to an object in an image, plays an important role in creating common ground. Many models that fuse visual and linguistic features have been proposed. However, few models consider the fusion of linguistic features with multiple visual features with different sizes of receptive fields, though the proper size of the receptive field of visual features intuitively varies depending on expressions. In this paper, we introduce a neural network architecture that modulates visual features with varying sizes of receptive field by linguistic features. We evaluate our architecture on tasks related to referring expression comprehension in two visual dialogue games. The results show the advantages and broad applicability of our architecture. Source code is available at https://github.com/Alab-NII/lcfp .</abstract>
      <url hash="ef3eca83">2020.findings-emnlp.420</url>
      <doi>10.18653/v1/2020.findings-emnlp.420</doi>
    </paper>
    <paper id="421">
      <title>Learning to Classify Events from Human Needs Category Descriptions</title>
      <author><first>Haibo</first><last>Ding</last></author>
      <author><first>Zhe</first><last>Feng</last></author>
      <pages>4698–4704</pages>
      <abstract>We study the problem of learning an event classifier from human needs category descriptions, which is challenging due to: (1) the use of highly abstract concepts in natural language descriptions, (2) the difficulty of choosing key concepts. To tackle these two challenges, we propose LeaPI, a zero-shot learning method that first automatically generate weak labels by instantiating high-level concepts with prototypical instances and then trains a human needs classifier with the weakly labeled data. To filter noisy concepts, we design a reinforced selection algorithm to choose high-quality concepts for instantiation. Experimental results on the human needs categorization task show that our method outperforms baseline methods, producing substantially better precision.</abstract>
      <url hash="68601998">2020.findings-emnlp.421</url>
      <attachment type="OptionalSupplementaryMaterial" hash="6edbd0b0">2020.findings-emnlp.421.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.421</doi>
    </paper>
    <paper id="422">
      <title>Automatic Term Name Generation for Gene Ontology: Task and Dataset</title>
      <author><first>Yanjian</first><last>Zhang</last></author>
      <author><first>Qin</first><last>Chen</last></author>
      <author><first>Yiteng</first><last>Zhang</last></author>
      <author><first>Zhongyu</first><last>Wei</last></author>
      <author><first>Yixu</first><last>Gao</last></author>
      <author><first>Jiajie</first><last>Peng</last></author>
      <author><first>Zengfeng</first><last>Huang</last></author>
      <author><first>Weijian</first><last>Sun</last></author>
      <author><first>Xuanjing</first><last>Huang</last></author>
      <pages>4705–4710</pages>
      <abstract>Terms contained in Gene Ontology (GO) have been widely used in biology and bio-medicine. Most previous research focuses on inferring new GO terms, while the term names that reflect the gene function are still named by the experts. To fill this gap, we propose a novel task, namely term name generation for GO, and build a large-scale benchmark dataset. Furthermore, we present a graph-based generative model that incorporates the relations between genes, words and terms for term name generation, which exhibits great advantages over the strong baselines.</abstract>
      <url hash="98a57758">2020.findings-emnlp.422</url>
      <doi>10.18653/v1/2020.findings-emnlp.422</doi>
    </paper>
    <paper id="423">
      <title>Compressing Transformer-Based Semantic Parsing Models using Compositional Code Embeddings</title>
      <author><first>Prafull</first><last>Prakash</last></author>
      <author><first>Saurabh Kumar</first><last>Shashidhar</last></author>
      <author><first>Wenlong</first><last>Zhao</last></author>
      <author><first>Subendhu</first><last>Rongali</last></author>
      <author><first>Haidar</first><last>Khan</last></author>
      <author><first>Michael</first><last>Kayser</last></author>
      <pages>4711–4717</pages>
      <abstract>The current state-of-the-art task-oriented semantic parsing models use BERT or RoBERTa as pretrained encoders; these models have huge memory footprints. This poses a challenge to their deployment for voice assistants such as Amazon Alexa and Google Assistant on edge devices with limited memory budgets. We propose to learn compositional code embeddings to greatly reduce the sizes of BERT-base and RoBERTa-base. We also apply the technique to DistilBERT, ALBERT-base, and ALBERT-large, three already compressed BERT variants which attain similar state-of-the-art performances on semantic parsing with much smaller model sizes. We observe 95.15% 98.46% embedding compression rates and 20.47% 34.22% encoder compression rates, while preserving &gt;97.5% semantic parsing performances. We provide the recipe for training and analyze the trade-off between code embedding sizes and downstream performances.</abstract>
      <url hash="5941a8a9">2020.findings-emnlp.423</url>
      <attachment type="OptionalSupplementaryMaterial" hash="c1e6fbf0">2020.findings-emnlp.423.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.423</doi>
    </paper>
    <paper id="424">
      <title><fixed-case>BERT-QE</fixed-case>: <fixed-case>C</fixed-case>ontextualized <fixed-case>Q</fixed-case>uery <fixed-case>E</fixed-case>xpansion for <fixed-case>D</fixed-case>ocument <fixed-case>R</fixed-case>e-ranking</title>
      <author><first>Zhi</first><last>Zheng</last></author>
      <author><first>Kai</first><last>Hui</last></author>
      <author><first>Ben</first><last>He</last></author>
      <author><first>Xianpei</first><last>Han</last></author>
      <author><first>Le</first><last>Sun</last></author>
      <author><first>Andrew</first><last>Yates</last></author>
      <pages>4718–4728</pages>
      <abstract>Query expansion aims to mitigate the mismatch between the language used in a query and in a document. However, query expansion methods can suffer from introducing non-relevant information when expanding the query. To bridge this gap, inspired by recent advances in applying contextualized models like BERT to the document retrieval task, this paper proposes a novel query expansion model that leverages the strength of the BERT model to select relevant document chunks for expansion. In evaluation on the standard TREC Robust04 and GOV2 test collections, the proposed BERT-QE model significantly outperforms BERT-Large models.</abstract>
      <url hash="1fff2b0d">2020.findings-emnlp.424</url>
      <doi>10.18653/v1/2020.findings-emnlp.424</doi>
    </paper>
    <paper id="425">
      <title><fixed-case>ZEN</fixed-case>: Pre-training <fixed-case>C</fixed-case>hinese Text Encoder Enhanced by N-gram Representations</title>
      <author><first>Shizhe</first><last>Diao</last></author>
      <author><first>Jiaxin</first><last>Bai</last></author>
      <author><first>Yan</first><last>Song</last></author>
      <author><first>Tong</first><last>Zhang</last></author>
      <author><first>Yonggang</first><last>Wang</last></author>
      <pages>4729–4740</pages>
      <abstract>The pre-training of text encoders normally processes text as a sequence of tokens corresponding to small text units, such as word pieces in English and characters in Chinese. It omits information carried by larger text granularity, and thus the encoders cannot easily adapt to certain combinations of characters. This leads to a loss of important semantic information, which is especially problematic for Chinese because the language does not have explicit word boundaries. In this paper, we propose ZEN, a BERT-based Chinese text encoder enhanced by n-gram representations, where different combinations of characters are considered during training, thus potential word or phrase boundaries are explicitly pre-trained and fine-tuned with the character encoder (BERT). Therefore ZEN incorporates the comprehensive information of both the character sequence and words or phrases it contains. Experimental results illustrated the effectiveness of ZEN on a series of Chinese NLP tasks, where state-of-the-art results is achieved on most tasks with requiring less resource than other published encoders. It is also shown that reasonable performance is obtained when ZEN is trained on a small corpus, which is important for applying pre-training techniques to scenarios with limited data. The code and pre-trained models of ZEN are available at https://github.com/sinovation/ZEN.</abstract>
      <url hash="f4b41562">2020.findings-emnlp.425</url>
      <doi>10.18653/v1/2020.findings-emnlp.425</doi>
    </paper>
    <paper id="426">
      <title>Finding <fixed-case>F</fixed-case>riends and Flipping Frenemies: Automatic Paraphrase Dataset Augmentation Using Graph Theory</title>
      <author><first>Hannah</first><last>Chen</last></author>
      <author><first>Yangfeng</first><last>Ji</last></author>
      <author><first>David</first><last>Evans</last></author>
      <pages>4741–4751</pages>
      <abstract>Most NLP datasets are manually labeled, so suffer from inconsistent labeling or limited size. We propose methods for automatically improving datasets by viewing them as graphs with expected semantic properties. We construct a paraphrase graph from the provided sentence pair labels, and create an augmented dataset by directly inferring labels from the original sentence pairs using a transitivity property. We use structural balance theory to identify likely mislabelings in the graph, and flip their labels. We evaluate our methods on paraphrase models trained using these datasets starting from a pretrained BERT model, and find that the automatically-enhanced training sets result in more accurate models.</abstract>
      <url hash="30da7833">2020.findings-emnlp.426</url>
      <doi>10.18653/v1/2020.findings-emnlp.426</doi>
    </paper>
    <paper id="427">
      <title>Probabilistic Case-based Reasoning for Open-World Knowledge Graph Completion</title>
      <author><first>Rajarshi</first><last>Das</last></author>
      <author><first>Ameya</first><last>Godbole</last></author>
      <author><first>Nicholas</first><last>Monath</last></author>
      <author><first>Manzil</first><last>Zaheer</last></author>
      <author><first>Andrew</first><last>McCallum</last></author>
      <pages>4752–4765</pages>
      <abstract>A case-based reasoning (CBR) system solves a new problem by retrieving ‘cases’ that are similar to the given problem. If such a system can achieve high accuracy, it is appealing owing to its simplicity, interpretability, and scalability. In this paper, we demonstrate that such a system is achievable for reasoning in knowledge-bases (KBs). Our approach predicts attributes for an entity by gathering reasoning paths from similar entities in the KB. Our probabilistic model estimates the likelihood that a path is effective at answering a query about the given entity. The parameters of our model can be efficiently computed using simple path statistics and require no iterative optimization. Our model is non-parametric, growing dynamically as new entities and relations are added to the KB. On several benchmark datasets our approach significantly outperforms other rule learning approaches and performs comparably to state-of-the-art embedding-based approaches. Furthermore, we demonstrate the effectiveness of our model in an “open-world” setting where new entities arrive in an online fashion, significantly outperforming state-of-the-art approaches and nearly matching the best offline method.</abstract>
      <url hash="7c9c3991">2020.findings-emnlp.427</url>
      <doi>10.18653/v1/2020.findings-emnlp.427</doi>
    </paper>
    <paper id="428">
      <title><fixed-case>TLDR</fixed-case>: Extreme Summarization of Scientific Documents</title>
      <author><first>Isabel</first><last>Cachola</last></author>
      <author><first>Kyle</first><last>Lo</last></author>
      <author><first>Arman</first><last>Cohan</last></author>
      <author><first>Daniel</first><last>Weld</last></author>
      <pages>4766–4777</pages>
      <abstract>We introduce TLDR generation, a new form of extreme summarization, for scientific papers. TLDR generation involves high source compression and requires expert background knowledge and understanding of complex domain-specific language. To facilitate study on this task, we introduce SCITLDR, a new multi-target dataset of 5.4K TLDRs over 3.2K papers. SCITLDR contains both author-written and expert-derived TLDRs, where the latter are collected using a novel annotation protocol that produces high-quality summaries while minimizing annotation burden. We propose CATTS, a simple yet effective learning strategy for generating TLDRs that exploits titles as an auxiliary training signal. CATTS improves upon strong baselines under both automated metrics and human evaluations. Data and code are publicly available at https://github.com/allenai/scitldr.</abstract>
      <url hash="9071ddb1">2020.findings-emnlp.428</url>
      <attachment type="OptionalSupplementaryMaterial" hash="f8d44aee">2020.findings-emnlp.428.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.428</doi>
    </paper>
    <paper id="429">
      <title>Tri-Train: Automatic Pre-Fine Tuning between Pre-Training and Fine-Tuning for <fixed-case>S</fixed-case>ci<fixed-case>NER</fixed-case></title>
      <author><first>Qingkai</first><last>Zeng</last></author>
      <author><first>Wenhao</first><last>Yu</last></author>
      <author><first>Mengxia</first><last>Yu</last></author>
      <author><first>Tianwen</first><last>Jiang</last></author>
      <author><first>Tim</first><last>Weninger</last></author>
      <author><first>Meng</first><last>Jiang</last></author>
      <pages>4778–4787</pages>
      <abstract>The training process of scientific NER models is commonly performed in two steps: i) Pre-training a language model by self-supervised tasks on huge data and ii) fine-tune training with small labelled data. The success of the strategy depends on the relevance between the data domains and between the tasks. However, gaps are found in practice when the target domains are specific and small. We propose a novel framework to introduce a “pre-fine tuning” step between pre-training and fine-tuning. It constructs a corpus by selecting sentences from unlabeled documents that are the most relevant with the labelled training data. Instead of predicting tokens in random spans, the pre-fine tuning task is to predict tokens in entity candidates identified by text mining methods. Pre-fine tuning is automatic and light-weight because the corpus size can be much smaller than pre-training data to achieve a better performance. Experiments on seven benchmarks demonstrate the effectiveness.</abstract>
      <url hash="ea5f5ad6">2020.findings-emnlp.429</url>
      <attachment type="OptionalSupplementaryMaterial" hash="072907e3">2020.findings-emnlp.429.OptionalSupplementaryMaterial.zip</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.429</doi>
    </paper>
    <paper id="430">
      <title>Hierarchical Region Learning for Nested Named Entity Recognition</title>
      <author><first>Xinwei</first><last>Long</last></author>
      <author><first>Shuzi</first><last>Niu</last></author>
      <author><first>Yucheng</first><last>Li</last></author>
      <pages>4788–4793</pages>
      <abstract>Named Entity Recognition (NER) is deeply explored and widely used in various tasks. Usually, some entity mentions are nested in other entities, which leads to the nested NER problem. Leading region based models face both the efficiency and effectiveness challenge due to the high subsequence enumeration complexity. To tackle these challenges, we propose a hierarchical region learning framework to automatically generate a tree hierarchy of candidate regions with nearly linear complexity and incorporate structure information into the region representation for better classification. Experiments on benchmark datasets ACE-2005, GENIA and JNLPBA demonstrate competitive or better results than state-of-the-art baselines.</abstract>
      <url hash="0328026c">2020.findings-emnlp.430</url>
      <doi>10.18653/v1/2020.findings-emnlp.430</doi>
    </paper>
    <paper id="431">
      <title>Understanding User Resistance Strategies in Persuasive Conversations</title>
      <author><first>Youzhi</first><last>Tian</last></author>
      <author><first>Weiyan</first><last>Shi</last></author>
      <author><first>Chen</first><last>Li</last></author>
      <author><first>Zhou</first><last>Yu</last></author>
      <pages>4794–4798</pages>
      <abstract>Persuasive dialog systems have various usages, such as donation persuasion and physical exercise persuasion. Previous persuasive dialog systems research mostly focused on analyzing the persuader’s strategies and paid little attention to the persuadee (user). However, understanding and addressing users’ resistance strategies is an essential job of a persuasive dialog system. So, we adopt a preliminary framework on persuasion resistance in psychology and design a fine-grained resistance strategy annotation scheme. We annotate the PersuasionForGood dataset with the scheme. With the enriched annotations, we build a classifier to predict the resistance strategies. Furthermore, we analyze the relationships between persuasion strategies and persuasion resistance strategies. Our work lays the ground for developing a persuasive dialogue system that can understand and address user resistance strategy appropriately. The code and data will be released.</abstract>
      <url hash="47212a35">2020.findings-emnlp.431</url>
      <doi>10.18653/v1/2020.findings-emnlp.431</doi>
    </paper>
    <paper id="432">
      <title>On the Sub-layer Functionalities of Transformer Decoder</title>
      <author><first>Yilin</first><last>Yang</last></author>
      <author><first>Longyue</first><last>Wang</last></author>
      <author><first>Shuming</first><last>Shi</last></author>
      <author><first>Prasad</first><last>Tadepalli</last></author>
      <author><first>Stefan</first><last>Lee</last></author>
      <author><first>Zhaopeng</first><last>Tu</last></author>
      <pages>4799–4811</pages>
      <abstract>There have been significant efforts to interpret the encoder of Transformer-based encoder-decoder architectures for neural machine translation (NMT); meanwhile, the decoder remains largely unexamined despite its critical role. During translation, the decoder must predict output tokens by considering both the source-language text from the encoder and the target-language prefix produced in previous steps. In this work, we study how Transformer-based decoders leverage information from the source and target languages – developing a universal probe task to assess how information is propagated through each module of each decoder layer. We perform extensive experiments on three major translation datasets (WMT En-De, En-Fr, and En-Zh). Our analysis provides insight on when and where decoders leverage different sources. Based on these insights, we demonstrate that the residual feed-forward module in each Transformer decoder layer can be dropped with minimal loss of performance – a significant reduction in computation and number of parameters, and consequently a significant boost to both training and inference speed.</abstract>
      <url hash="70919c1b">2020.findings-emnlp.432</url>
      <doi>10.18653/v1/2020.findings-emnlp.432</doi>
    </paper>
    <paper id="433">
      <title>Extremely Low Bit Transformer Quantization for On-Device Neural Machine Translation</title>
      <author><first>Insoo</first><last>Chung</last></author>
      <author><first>Byeongwook</first><last>Kim</last></author>
      <author><first>Yoonjung</first><last>Choi</last></author>
      <author><first>Se Jung</first><last>Kwon</last></author>
      <author><first>Yongkweon</first><last>Jeon</last></author>
      <author><first>Baeseong</first><last>Park</last></author>
      <author><first>Sangha</first><last>Kim</last></author>
      <author><first>Dongsoo</first><last>Lee</last></author>
      <pages>4812–4826</pages>
      <abstract>The deployment of widely used Transformer architecture is challenging because of heavy computation load and memory overhead during inference, especially when the target device is limited in computational resources such as mobile or edge devices. Quantization is an effective technique to address such challenges. Our analysis shows that for a given number of quantization bits, each block of Transformer contributes to translation quality and inference computations in different manners. Moreover, even inside an embedding block, each word presents vastly different contributions. Correspondingly, we propose a mixed precision quantization strategy to represent Transformer weights by an extremely low number of bits (e.g., under 3 bits). For example, for each word in an embedding block, we assign different quantization bits based on statistical property. Our quantized Transformer model achieves 11.8× smaller model size than the baseline model, with less than -0.5 BLEU. We achieve 8.3× reduction in run-time memory footprints and 3.5× speed up (Galaxy N10+) such that our proposed compression strategy enables efficient implementation for on-device NMT.</abstract>
      <url hash="594aba36">2020.findings-emnlp.433</url>
      <doi>10.18653/v1/2020.findings-emnlp.433</doi>
    </paper>
    <paper id="434">
      <title>Robust <fixed-case>B</fixed-case>acked-off <fixed-case>E</fixed-case>stimation of <fixed-case>O</fixed-case>ut-of-<fixed-case>V</fixed-case>ocabulary <fixed-case>E</fixed-case>mbeddings</title>
      <author><first>Nobukazu</first><last>Fukuda</last></author>
      <author><first>Naoki</first><last>Yoshinaga</last></author>
      <author><first>Masaru</first><last>Kitsuregawa</last></author>
      <pages>4827–4838</pages>
      <abstract>Out-of-vocabulary (oov) words cause serious troubles in solving natural language tasks with a neural network. Existing approaches to this problem resort to using subwords, which are shorter and more ambiguous units than words, in order to represent oov words with a bag of subwords. In this study, inspired by the processes for creating words from known words, we propose a robust method of estimating oov word embeddings by referring to pre-trained word embeddings for known words with similar surfaces to target oov words. We collect known words by segmenting oov words and by approximate string matching, and we then aggregate their pre-trained embeddings. Experimental results show that the obtained oov word embeddings improve not only word similarity tasks but also downstream tasks in Twitter and biomedical domains where oov words often appear, even when the computed oov embeddings are integrated into a bert-based strong baseline.</abstract>
      <url hash="2627fb23">2020.findings-emnlp.434</url>
      <doi>10.18653/v1/2020.findings-emnlp.434</doi>
    </paper>
    <paper id="435">
      <title><fixed-case>E</fixed-case>xploiting <fixed-case>U</fixed-case>nsupervised <fixed-case>D</fixed-case>ata for <fixed-case>E</fixed-case>motion <fixed-case>R</fixed-case>ecognition in <fixed-case>C</fixed-case>onversations</title>
      <author><first>Wenxiang</first><last>Jiao</last></author>
      <author><first>Michael</first><last>Lyu</last></author>
      <author><first>Irwin</first><last>King</last></author>
      <pages>4839–4846</pages>
      <abstract>Emotion Recognition in Conversations (ERC) aims to predict the emotional state of speakers in conversations, which is essentially a text classification task. Unlike the sentence-level text classification problem, the available supervised data for the ERC task is limited, which potentially prevents the models from playing their maximum effect. In this paper, we propose a novel approach to leverage unsupervised conversation data, which is more accessible. Specifically, we propose the Conversation Completion (ConvCom) task, which attempts to select the correct answer from candidate answers to fill a masked utterance in a conversation. Then, we Pre-train a basic COntext-Dependent Encoder (Pre-CODE) on the ConvCom task. Finally, we fine-tune the Pre-CODE on the datasets of ERC. Experimental results demonstrate that pre-training on unsupervised data achieves significant improvement of performance on the ERC datasets, particularly on the minority emotion classes.</abstract>
      <url hash="3a828a97">2020.findings-emnlp.435</url>
      <doi>10.18653/v1/2020.findings-emnlp.435</doi>
    </paper>
    <paper id="436">
      <title>Tensorized Embedding Layers</title>
      <author><first>Oleksii</first><last>Hrinchuk</last></author>
      <author><first>Valentin</first><last>Khrulkov</last></author>
      <author><first>Leyla</first><last>Mirvakhabova</last></author>
      <author><first>Elena</first><last>Orlova</last></author>
      <author><first>Ivan</first><last>Oseledets</last></author>
      <pages>4847–4860</pages>
      <abstract>The embedding layers transforming input words into real vectors are the key components of deep neural networks used in natural language processing. However, when the vocabulary is large, the corresponding weight matrices can be enormous, which precludes their deployment in a limited resource setting. We introduce a novel way of parameterizing embedding layers based on the Tensor Train decomposition, which allows compressing the model significantly at the cost of a negligible drop or even a slight gain in performance. We evaluate our method on a wide range of benchmarks in natural language processing and analyze the trade-off between performance and compression ratios for a wide range of architectures, from MLPs to LSTMs and Transformers.</abstract>
      <url hash="c050b12c">2020.findings-emnlp.436</url>
      <doi>10.18653/v1/2020.findings-emnlp.436</doi>
    </paper>
    <paper id="437">
      <title>Speaker or Listener? The Role of a Dialog Agent</title>
      <author><first>Yafei</first><last>Liu</last></author>
      <author><first>Hongjin</first><last>Qian</last></author>
      <author><first>Hengpeng</first><last>Xu</last></author>
      <author><first>Jinmao</first><last>Wei</last></author>
      <pages>4861–4869</pages>
      <abstract>For decades, chitchat bots are designed as a listener to passively answer what people ask. This passive and relatively simple dialogue mechanism gains less attention from humans and consumes the interests of human beings rapidly. Therefore some recent researches attempt to endow the bots with proactivity through external knowledge to transform the role from a listener to a speaker with a hypothesis that the speaker expresses more just like a knowledge disseminator. However, along with the proactive manner introduced into a dialogue agent, an issue arises that, with too many knowledge facts to express, the agent starts to talks endlessly, and even completely ignores what the other expresses in dialogue sometimes, which greatly harms the interest of the other chatter to continue the conversation. To the end, we propose a novel model named Initiative-Imitate to interact with adaptive initiative throughout a dialogue. It forces the agent to express in parallel with the appropriate role during the whole conversation. The corresponding experiments show the proposed Initiative-Imitate obtains competitive results both on the automatic and manual metrics. And the fluency and engagement of the chatbot have also been improved significantly. Besides, the case study indicates the Initiative-Imitate can constantly transfer to appropriate role timely and response more properly during the whole continuous conversation.</abstract>
      <url hash="73ca2930">2020.findings-emnlp.437</url>
      <doi>10.18653/v1/2020.findings-emnlp.437</doi>
    </paper>
    <paper id="438">
      <title>Bridging Textual and Tabular Data for Cross-Domain Text-to-<fixed-case>SQL</fixed-case> Semantic Parsing</title>
      <author><first>Xi Victoria</first><last>Lin</last></author>
      <author><first>Richard</first><last>Socher</last></author>
      <author><first>Caiming</first><last>Xiong</last></author>
      <pages>4870–4888</pages>
      <abstract>We present BRIDGE, a powerful sequential architecture for modeling dependencies between natural language questions and relational databases in cross-DB semantic parsing. BRIDGE represents the question and DB schema in a tagged sequence where a subset of the fields are augmented with cell values mentioned in the question. The hybrid sequence is encoded by BERT with minimal subsequent layers and the text-DB contextualization is realized via the fine-tuned deep attention in BERT. Combined with a pointer-generator decoder with schema-consistency driven search space pruning, BRIDGE attained state-of-the-art performance on the well-studied Spider benchmark (65.5% dev, 59.2% test), despite being much simpler than most recently proposed models for this task. Our analysis shows that BRIDGE effectively captures the desired cross-modal dependencies and has the potential to generalize to more text-DB related tasks. Our model implementation is available at https://github.com/ salesforce/TabularSemanticParsing.</abstract>
      <url hash="785d2b49">2020.findings-emnlp.438</url>
      <doi>10.18653/v1/2020.findings-emnlp.438</doi>
    </paper>
    <paper id="439">
      <title>Do Language Embeddings capture Scales?</title>
      <author><first>Xikun</first><last>Zhang</last></author>
      <author><first>Deepak</first><last>Ramachandran</last></author>
      <author><first>Ian</first><last>Tenney</last></author>
      <author><first>Yanai</first><last>Elazar</last></author>
      <author><first>Dan</first><last>Roth</last></author>
      <pages>4889–4896</pages>
      <abstract>Pretrained Language Models (LMs) have been shown to possess significant linguistic, common sense and factual knowledge. One form of knowledge that has not been studied yet in this context is information about the scalar magnitudes of objects. We show that pretrained language models capture a significant amount of this information but are short of the capability required for general common-sense reasoning. We identify contextual information in pre-training and numeracy as two key factors affecting their performance, and show that a simple method of canonicalizing numbers can have a significant effect on the results.</abstract>
      <url hash="6d66605d">2020.findings-emnlp.439</url>
      <doi>10.18653/v1/2020.findings-emnlp.439</doi>
    </paper>
    <paper id="440">
      <title>Paraphrasing vs Coreferring: Two Sides of the Same Coin</title>
      <author><first>Yehudit</first><last>Meged</last></author>
      <author><first>Avi</first><last>Caciularu</last></author>
      <author><first>Vered</first><last>Shwartz</last></author>
      <author><first>Ido</first><last>Dagan</last></author>
      <pages>4897–4907</pages>
      <abstract>We study the potential synergy between two different NLP tasks, both confronting predicate lexical variability: identifying predicate paraphrases, and event coreference resolution. First, we used annotations from an event coreference dataset as distant supervision to re-score heuristically-extracted predicate paraphrases. The new scoring gained more than 18 points in average precision upon their ranking by the original scoring method. Then, we used the same re-ranking features as additional inputs to a state-of-the-art event coreference resolution model, which yielded modest but consistent improvements to the model’s performance. The results suggest a promising direction to leverage data and models for each of the tasks to the benefit of the other.</abstract>
      <url hash="ff847aca">2020.findings-emnlp.440</url>
      <doi>10.18653/v1/2020.findings-emnlp.440</doi>
    </paper>
    <paper id="441">
      <title>Active Sentence Learning by Adversarial Uncertainty Sampling in Discrete Space</title>
      <author><first>Dongyu</first><last>Ru</last></author>
      <author><first>Jiangtao</first><last>Feng</last></author>
      <author><first>Lin</first><last>Qiu</last></author>
      <author><first>Hao</first><last>Zhou</last></author>
      <author><first>Mingxuan</first><last>Wang</last></author>
      <author><first>Weinan</first><last>Zhang</last></author>
      <author><first>Yong</first><last>Yu</last></author>
      <author><first>Lei</first><last>Li</last></author>
      <pages>4908–4917</pages>
      <abstract>Active learning for sentence understanding aims at discovering informative unlabeled data for annotation and therefore reducing the demand for labeled data. We argue that the typical uncertainty sampling method for active learning is time-consuming and can hardly work in real-time, which may lead to ineffective sample selection. We propose adversarial uncertainty sampling in discrete space (AUSDS) to retrieve informative unlabeled samples more efficiently. AUSDS maps sentences into latent space generated by the popular pre-trained language models, and discover informative unlabeled text samples for annotation via adversarial attack. The proposed approach is extremely efficient compared with traditional uncertainty sampling with more than 10x speedup. Experimental results on five datasets show that AUSDS outperforms strong baselines on effectiveness.</abstract>
      <url hash="a49de01f">2020.findings-emnlp.441</url>
      <doi>10.18653/v1/2020.findings-emnlp.441</doi>
    </paper>
    <paper id="442">
      <title><fixed-case>C</fixed-case>oming to <fixed-case>T</fixed-case>erms: <fixed-case>A</fixed-case>utomatic <fixed-case>F</fixed-case>ormation of <fixed-case>N</fixed-case>eologisms in <fixed-case>H</fixed-case>ebrew</title>
      <author><first>Moran</first><last>Mizrahi</last></author>
      <author><first>Stav</first><last>Yardeni Seelig</last></author>
      <author><first>Dafna</first><last>Shahaf</last></author>
      <pages>4918–4929</pages>
      <abstract>Spoken languages are ever-changing, with new words entering them all the time. However, coming up with new words (neologisms) today relies exclusively on human creativity. In this paper we propose a system to automatically suggest neologisms. We focus on the Hebrew language as a test case due to the unusual regularity of its noun formation. User studies comparing our algorithm to experts and non-experts demonstrate that our algorithm is capable of generating high-quality outputs, as well as enhance human creativity. More broadly, we seek to inspire more computational work around the topic of linguistic creativity, which we believe offers numerous unexplored opportunities.</abstract>
      <url hash="46be12dc">2020.findings-emnlp.442</url>
      <doi>10.18653/v1/2020.findings-emnlp.442</doi>
    </paper>
    <paper id="443">
      <title>Dual Inference for Improving Language Understanding and Generation</title>
      <author><first>Shang-Yu</first><last>Su</last></author>
      <author><first>Yung-Sung</first><last>Chuang</last></author>
      <author><first>Yun-Nung</first><last>Chen</last></author>
      <pages>4930–4936</pages>
      <abstract>Natural language understanding (NLU) and Natural language generation (NLG) tasks hold a strong dual relationship, where NLU aims at predicting semantic labels based on natural language utterances and NLG does the opposite. The prior work mainly focused on exploiting the duality in model training in order to obtain the models with better performance. However, regarding the fast-growing scale of models in the current NLP area, sometimes we may have difficulty retraining whole NLU and NLG models. To better address the issue, this paper proposes to leverage the duality in the inference stage without the need of retraining. The experiments on three benchmark datasets demonstrate the effectiveness of the proposed method in both NLU and NLG, providing the great potential of practical usage.</abstract>
      <url hash="fd15d888">2020.findings-emnlp.443</url>
      <doi>10.18653/v1/2020.findings-emnlp.443</doi>
    </paper>
    <paper id="444">
      <title>Joint Intent Detection and Entity Linking on Spatial Domain Queries</title>
      <author><first>Lei</first><last>Zhang</last></author>
      <author><first>Runze</first><last>Wang</last></author>
      <author><first>Jingbo</first><last>Zhou</last></author>
      <author><first>Jingsong</first><last>Yu</last></author>
      <author><first>Zhenhua</first><last>Ling</last></author>
      <author><first>Hui</first><last>Xiong</last></author>
      <pages>4937–4947</pages>
      <abstract>Continuous efforts have been devoted to language understanding (LU) for conversational queries with the fast and wide-spread popularity of voice assistants. In this paper, we first study the LU problem in the spatial domain, which is a critical problem for providing location-based services by voice assistants but is without in-depth investigation in existing studies. Spatial domain queries have several unique properties making them be more challenging for language understanding than common conversational queries, including lexical-similar but diverse intents and highly ambiguous words. Thus, a special tailored LU framework for spatial domain queries is necessary. To the end, a dataset was extracted and annotated based on the real-life queries from a voice assistant service. We then proposed a new multi-task framework that jointly learns the intent detection and entity linking tasks on the with invented hierarchical intent detection method and triple-scoring mechanism for entity linking. A specially designed spatial GCN is also utilized to model spatial context information among entities. We have conducted extensive experimental evaluations with state-of-the-art entity linking and intent detection methods, which demonstrated that can outperform all baselines with a significant margin.</abstract>
      <url hash="e434b8db">2020.findings-emnlp.444</url>
      <attachment type="OptionalSupplementaryMaterial" hash="e9c162d3">2020.findings-emnlp.444.OptionalSupplementaryMaterial.pdf</attachment>
      <doi>10.18653/v1/2020.findings-emnlp.444</doi>
    </paper>
    <paper id="445">
      <title><fixed-case>I</fixed-case>ndic<fixed-case>NLPS</fixed-case>uite: Monolingual Corpora, Evaluation Benchmarks and Pre-trained Multilingual Language Models for <fixed-case>I</fixed-case>ndian Languages</title>
      <author><first>Divyanshu</first><last>Kakwani</last></author>
      <author><first>Anoop</first><last>Kunchukuttan</last></author>
      <author><first>Satish</first><last>Golla</last></author>
      <author><first>Gokul</first><last>N.C.</last></author>
      <author><first>Avik</first><last>Bhattacharyya</last></author>
      <author><first>Mitesh M.</first><last>Khapra</last></author>
      <author><first>Pratyush</first><last>Kumar</last></author>
      <pages>4948–4961</pages>
      <abstract>In this paper, we introduce NLP resources for 11 major Indian languages from two major language families. These resources include: (a) large-scale sentence-level monolingual corpora, (b) pre-trained word embeddings, (c) pre-trained language models, and (d) multiple NLU evaluation datasets (<i>IndicGLUE</i> benchmark). The monolingual corpora contains a total of 8.8 billion tokens across all 11 languages and Indian English, primarily sourced from news crawls. The word embeddings are based on <i>FastText</i>, hence suitable for handling morphological complexity of Indian languages. The pre-trained language models are based on the compact ALBERT model. Lastly, we compile the (<i>IndicGLUE</i> benchmark for Indian language NLU. To this end, we create datasets for the following tasks: Article Genre Classification, Headline Prediction, Wikipedia Section-Title Prediction, Cloze-style Multiple choice QA, Winograd NLI and COPA. We also include publicly available datasets for some Indic languages for tasks like Named Entity Recognition, Cross-lingual Sentence Retrieval, Paraphrase detection, <i>etc.</i> Our embeddings are competitive or better than existing pre-trained embeddings on multiple tasks. We hope that the availability of the dataset will accelerate Indic NLP research which has the potential to impact more than a billion people. It can also help the community in evaluating advances in NLP over a more diverse pool of languages. The data and models are available at <url>https://indicnlp.ai4bharat.org</url>.</abstract>
      <url hash="2a22f7f6">2020.findings-emnlp.445</url>
      <doi>10.18653/v1/2020.findings-emnlp.445</doi>
    </paper>
    <paper id="446">
      <title>Weakly-Supervised Modeling of Contextualized Event Embedding for Discourse Relations</title>
      <author><first>I-Ta</first><last>Lee</last></author>
      <author><first>Maria Leonor</first><last>Pacheco</last></author>
      <author><first>Dan</first><last>Goldwasser</last></author>
      <pages>4962–4972</pages>
      <abstract>Representing, and reasoning over, long narratives requires models that can deal with complex event structures connected through multiple relationship types. This paper suggests to represent this type of information as a narrative graph and learn contextualized event representations over it using a relational graph neural network model. We train our model to capture event relations, derived from the Penn Discourse Tree Bank, on a huge corpus, and show that our multi-relational contextualized event representation can improve performance when learning script knowledge without direct supervision and provide a better representation for the implicit discourse sense classification task.</abstract>
      <url hash="33aeb376">2020.findings-emnlp.446</url>
      <doi>10.18653/v1/2020.findings-emnlp.446</doi>
    </paper>
    <paper id="447">
      <title>Enhancing Generalization in Natural Language Inference by Syntax</title>
      <author><first>Qi</first><last>He</last></author>
      <author><first>Han</first><last>Wang</last></author>
      <author><first>Yue</first><last>Zhang</last></author>
      <pages>4973–4978</pages>
      <abstract>Pre-trained language models such as BERT have achieved the state-of-the-art performance on natural language inference (NLI). However, it has been shown that such models can be tricked by variations of surface patterns such as syntax. We investigate the use of dependency trees to enhance the generalization of BERT in the NLI task, leveraging on a graph convolutional network to represent a syntax-based matching graph with heterogeneous matching patterns. Experimental results show that, our syntax-based method largely enhance generalization of BERT on a test set where the sentence pair has high lexical overlap but diverse syntactic structures, and do not degrade performance on the standard test set. In other words, the proposed method makes BERT more robust on syntactic changes.</abstract>
      <url hash="556acea6">2020.findings-emnlp.447</url>
      <doi>10.18653/v1/2020.findings-emnlp.447</doi>
    </paper>
  </volume>
</collection>