Seminar: Sentence rewriting for text processing

20160622_113349Richard Evans, University of Wolverhampton

Date and time: Wednesday, June 22nd, 11:30
Room: MC224
Evidence from clinical linguistics and NLP shows that sentences which are propositionally dense and syntactically complex are relatively difficult to process by humans and machines. Sentences containing compound clauses and nominally bound relative clauses are likely to be propositionally dense and syntactically complex. In this project, I propose an NLP pipeline to automatically rewrite these types of sentences as sequences of simpler sentences that preserve as much of the meaning of the originals as possible.

I present an automatic iterative sentence rewriting method which exploits handcrafted patterns and information about explicit signs of syntactic complexity such as conjunctions, complementisers, and punctuation marks. Evaluation of the method by comparison to sentences rewritten by humans is described. The difficulty of interpreting the obtained results motivates subsequent research in which the comparative evaluation method is reinforced using extrinsic methods exploiting independent NLP applications and human-centred reading tasks.