Temporal Expression Identification Based on Semantic Roles

Following TimeML (TIMEX3) specifications, we present a study analyzing to what extent are semantic roles useful in temporal expression identification task, as well as, a list of the potential applications of this combination. For that purpose, two approaches of a temporal expression identification system based on semantic roles have been developed: (1) Baseline and (2) TIPSem-Full. The first approach is a direct conversion from a temporal semantic role to a temporal expression. The second one processes and converts all temporal roles into correct TIMEX3, using a set of transformation rules defined in this paper. These two approaches have been evaluated using Time Bank corpus. The evaluation results confirm that the application of semantic roles to temporal expression identification task can be valuable, obtaining, for TIPSem-Full, an 73.4% in F1 for relaxed span and a 65.9% in F1 for strict span.