Revised syntactic attributes for relative cause simplification and relative pronoun correction

Abstract

Using previous work for resolving relative clause attachment ambiguities as a starting point, we expanded an existing framework consisting mainly of semantic features to include new syntactic features. We used these new syntactic attributes on their own, as well as with the existing ones in two separate applications. First, we examined sentences containing restrictive and non-restrictive relative clauses and classified them according to whether the relative clauses could be removed to simplify the sentences. In the second application, we corrected the selection of relative pronouns in French to English machine translations. To test our revised syntactic attributes, we devised a two-stage system. In the first stage, we used binary classification to perform an initial screening of sentences, followed by a more detailed classification in the second stage. We then evaluated the performance of each stage as well as the system as a whole. In each application, we made specific adjustments and succeeded in demonstrating that the combination of syntactic and semantic attributes was more effective in these classification tasks than relying on one type of attribute exclusively.