In this homework, you will analyze some syntactic structures in the Twitter part-of-speech corpus. You will select sentences from the training file oct27.train. You can select any sentences you like, but you may find it easiest to work with relatively short sentences. You may consult the Stanford parser to get some sense of how an automatic parser might treat a sentence like yours, but keep in mind that like all automatic parsers, it makes mistakes.
Our canonical example “They eat fish with chopsticks” has prepositional phrase attachment ambiguity. Find two examples of prepositional phrase attachment ambiguity in the corpus, choose the attachment site that you feel is correct, and explain why. You can find prepositions by searching for the “P” tag. However, unlike the Penn Treebank, this corpus tags “to” as “P”, even when it is functioning in an infinitive verb phrase (e.g., “we are going to ace the midterm”). You are more likely to find prepositional phrases by searching for other prepositions, like “in”, “at”, “of”, etc.
Another common source of syntactic ambiguity is the scope of coordinators like “and” and “or”. For example, in the sentence “She likes English music and tacos”, there are two interpretations:
- She likes ((English music) and (tacos))
- She likes (English (music and tacos))
Find two examples of scope ambiguity in the Twitter dataset, showing the possible interpretations and indicating which one you think is correct.
You have now gathered four examples. Provide full syntactic analysis for two of them, using the grammar fragment described in the lecture slides. You may need to make up a few productions of your own to deal with Twitter-specific phenomena like hashtags.