A Hybrid Approach to Inferring a Consistent Temporal Relation Set in Natural Language Text
Lee, Chong Min
Katz, Graham E
This dissertation investigates the temporal relation identificationtask. The goal is to construct consistent temporal relations betweentemporal entities (e.g., events and time expressions) in anarrative. Constructing consistent temporal relations is challengingdue to the exponential increase in the number of candidates fortemporal relations proportional to the number of pairs of temporalentities. When we use transitive constraints to construct consistenttemporal relations, performance improvement can be expected becausethe application of transitive constraints reduces the number ofpossible relation candidates in a narrative.The primary objective of this study is to develop a temporal relationidentification (TRI) system that is composed of three modules: 1) amodule that classifies the temporal relation of a pair of temporalentities, 2) a module that extracts conflicting classified relationsusing transitive constraints, and 3) a module that restores consistenttemporal relations from the conflicting relations using transitiveconstraints. In developing a TRI system, this dissertation examineswhether the application of transitive constraints to such a system canlead to performance improvement.The first step in developing the system was to implement a rudimentarytemporal relation classification module. The module labels a pair oftemporal entities with a temporal relation among eleven possibletemporal relations. Next, a method for extracting conflictingrelations among classified relations is proposed. The extractionmethod is based on heuristics because of NP-hard complexity inextracting all conflicting relations. Finally, two heuristic methodsare proposed that restore consistent temporal structure fromconflicting relations using transitive constraints. The performance ofthe developed system is tested using TimeBank and AQUAINT temporalcorpora.The results of this work indicate that a performance improvementthrough the application of transitive constraints to TRI task is notguaranteed. Furthermore, this study empirically shows the limitationson performance improvement through the application of transitiveconstraints to the TRI task and identifies the bottlenecks in the TRItask.
MetadataShow full item record
Showing items related by title, author, creator and subject.
Inferred Propositions and the Expression of the Evidence Relation in Natural Language: Evidentiality in Central Alaskan Yup'ik Eskimo and English Krawczyk, Elizabeth Allyn (Georgetown University, 2012)<italic>Evidentiality</italic> has usually been defined as the grammaticalized expression