Skip to content

Using Linguistic Rule-based Methods and Autoencoder Neural Networks to generate Sarcastic one-liners.

License

Notifications You must be signed in to change notification settings

aravindvenu7/Sarcasm-Generation

Repository files navigation

Sarcasm-Generation

This repository contains code for a rule-based JAPE Generator for computational humour generation.For further info on the JAPE Generator,please refer to : The JAPE riddle generator: technical specification by Graeme Ritchie(2003)

Data

The dataset used for constucting one-liners consists of an alphabetically ordered text file containing homophones.The contents of the file have been scraped from the web using a simple Selenium Web Scraper Further,the data on the basis of which jokes have been generated consists of movie summaries of 50 movies from the Cornell Movie Summaries Corpus.

#Generator:

5 types of one-liners have been generated according to the following rules:

  1. NA Joke:

"What do you call a\an " + d1 " that is a\an d2 ? " "A\An" + w1 + w2 + "."

w1 - adjective w2 - noun

d1 - adjective definition d2 - noun definition

  1. N2A2 Joke:

"What do you call a\an" + d1 " that is a\an" d2 ?" "A\An" + w1 + w2 + "."

w1 - adjective w2 - noun

d1 - synonym of w2 d2 - synonym of w1

3.N2AN Joke:

"When is a\an " + d1 + " like a\an " + d2 + "? " "When it is a\an " w1 + w2 + "."

w1 - adjective w2 - noun

d1 and d2 are synonyms

4.N2V2 Joke:

"Why did someone " + d2 + "a\an " + d1 + "? " "So they could " + w1 + " a\an " + w2 + "."

w1 - verb w2 - noun

d1 - verb synonym d2 = noun synonym

5.N4 Joke:

"When is a/an" + d1 + " like a\an" + d2 + "? " "When it is a\an " + w2 + "."

d1 - noun d2 - noun w2 - Synonym of d1

Note here that d2 and w2 are homophones

Further Work and Reading:

As of now, a rule based joke generator has been created successfully.I propose two further steps

Step1: Build a generative model based on RNN architecture and train it on the dataset of all jokes have been generated by the rule-based generator. Step2: Build a seq2seq model with variational attention mechanism which is capable of generating jokes similar to the rule-based generator given a movie summary.Variational attention has been proposed here because it has been proved to generate a greater variety of questions.For further info,please refer to : Variational Attention for Sequence-to-Sequence Models - Hareesh Bahuleyan(2018).

About

Using Linguistic Rule-based Methods and Autoencoder Neural Networks to generate Sarcastic one-liners.

Resources

License

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published

Languages