A paraphrase of a sentence is a rewording – possibly using different words – which roughly means the same thing; a paraphrase of a sentence does not have to be longer or shorter than the original sentence. Paraphrase generation can be considered as a machine translation task – translating from one English sentence to another that means the same thing. This project investigates the effects of leveraging syntactic information in the paraphrase generation process, using a statistical phrase-based machine translation approach (Quirk, Brockett, & Dolan, 2004) as a baseline.

A paraphrase generation system could improve search engines for web browsing, create interactive computer systems, or even be used as a sentence-level thesaurus in word processing programs – instead of looking up a word to substitute, imagine looking up a possible sentence substitute.

BYU NLP Lab Tech Report #2: “Generating Paraphrases with Greater Syntactic Variation using Syntactic Phrases”

Please contact Rebecca Madsen at rmadsen(at)byu(dot)net or visit the Natural Language Processing research lab in room 3346 TMCB.

Volunteers