Resolving Demonstrative Anaphora in the TRAINS93 Corpus
Donna K. Byron, James F. Allen · 1998
This paper reports the results of a first-pass annotation exercise of third-person pronouns in the TRAINS spoken dialog corpus. Our research goal is to disambiguate demonstrative pronouns that/this, and the annotation project is undertaken with the intent to find surface features that can be used to strongly prefer/disprefer certain potential referents. Pragmatic differences between demonstrative pronouns and definite pronouns are also discussed. An analysis of the annotation data uncovered many statistically significant patterns that can be used to determine when a pronoun might be referring to an entity in the global focus, such as the plan or a discourse segment, versus an item in the task domain, such as an engine. These patterns can be used to constrain the search space in a reference resolution algorithm that resolves reference to both task and global-focus entities using only easy-to-compute syntactic features.