A semantic approach to temporal information processing

Autor: Llorens Martínez, Héctor
Jazyk: Catalan<br />English
Rok vydání: 2011
Předmět:
Druh dokumentu: Doctoral Thesis
Popis: This thesis is focused on temporal information processing, which is a task framed in the field of natural language processing. The aim of this task is to obtain the temporal location and ordering of the events expressed in text or discourse, which requires the automatic interpretation of temporal expressions, events, and the temporal relations between them. The majority of current approaches are based on morphosyntactic knowledge. However, temporal entities are often ambiguous at that language analysis level. Our hypothesis is that the linguistic expression of time is a semantic phenomenon and therefore, to achieve a better extraction performance, temporal information must be processed using semantics. To prove this hypothesis, we present a semantic approach to temporal information processing: TIPSem. This is an automated system that includes features based on lexical semantics, semantic roles, and temporal semantics, in addition to morphosyntactic features. TIPSem has been empirically evaluated through the participation in the TempEval-2 international evaluation exercise and subsequent experiments based on this test. The results obtained firmly support the presented hypothesis and their analysis demonstrates that semantic features aid in handling morphosyntactic ambiguity and favour generalization capabilities. These conclusions have been reached for different languages (i.e., English, Spanish, Italian, and Chinese), which supports the defended hypothesis at a multilingual level. Finally, as compared with the state of the art, TIPSem obtains a very competitive performance level and introduces a remarkable improvement in event processing. TIPSem has been applied to the problem of graphical representation of temporal information. We have developed a dynamic interface which brings users time-based access to information: Time-Surfer. The results obtained through a user-oriented evaluation of this interface demonstrate that TIPSem’s performance is also satisfactory from an extrinsic standpoint.
This PhD thesis has been supported by the Spanish Government, in projects: TIN-2006-15265-C06-01, TIN-2009-13391-C04-01 and PROMETEO/2009/119, where Hector Llorens is funded under a FPI grant (BES-2007-16256).
Databáze: Networked Digital Library of Theses & Dissertations