A semantic approach to antecedent selection in verb phrase ellipsis

Consider this example of Verb Phrase Ellipsis (VPE): "The man [ANT1 stood up because the door bell [ANT2 rang]], but his son [VPE didn't]." Of the two possible antecedents, ant1 is the correct one. In earlier studies by Hardt (1997) and Nielsen (2005), syntactical features of candidate antecedents were used to determine the most plausible antecedent. With their methods, they reached accuracies of 84% and 79% respectively. Inspired by the ongoing theoretical debate on whether ellipsis is resolved syntactically or semantically, the research described in this thesis elaborates on these studies by adding a number of semantic features. To acquire semantic information from discourse, I use Boxer (Bos, 2005), a semantic parser that constructs Discourse Representation Structures (Kamp and Reyle, 1993) from syntactically parsed discourse. These semantic features are (1) semantic similarity of VPE and antecedent subjects, (2) parallelism of propositional phrases, (3) similarity in tense and (4) similarity in modality. To determine semantic similarity of nouns in (1) and (2), I useWordNets path distance measure. Like in Hardt (1997), a scoring mechanism is used to determine which of the possible antecedents is the most plausible one. Each feature that an antecedent may or may not have contributes to the score of an antecedent with a particular positive or negative value. These values are optimized using a Genetic Algorithm with close to 400 manually annotated examples of VPE from the Wall Street Journal part of the Penn Treebank. The added features have shown no improvement over baseline performance. A number of possible reasons for this low performance and suggestions for improvement are given in the discussion section.