On The Problem With Anaphora Resolution Mechanism In DRT And Its Solutions

This thesis addresses a particular instance of syntax-semantics interface problems,commonly known as anaphora resolution. Anaphora is a ubiquitous phenomenon innatural language. Simply stated, anaphora resolution, which most commonly appears aspronoun resolution, is the problem of resolving references to earlier or latter items in thediscourse. Anaphora is one of the most complex phenomena in verbal communication. Itnot only exists at the intra-sentential level, but is widespread at the inter-sentential levelas well. The fact that anaphora resolution calls for a synthetic consideration of factorssuch as government and binding, context of various types, and even those concerningpsychology and cognition makes it one of the most complicated issues in naturallanguage processing.Discourse Representation Theory (DRT for short) proposed by Hans Kamp in1981overcomes the fatal weakness encountered in the static semantics by introducingoriginally a discourse representation structure (DRS for short) as a new level ofpresentation along with a series of construction rules into the model-theoretic semantics.At this level, the conflict between the referential interpretation of noun phrases and theirquantificational interpretation is solved and thus the anaphora types such as referentialanaphora, bound-variables, and E-type anaphora are resolved effectively, whether theyare inter-sentential or intra-sentential, even if they span the entire discourse. What themost significant impact that DRT has made on linguistics is probably the innovative wayit brings with respect to formalization of context to usual model-theoretic account ofanaphora. It provides linguists with a way out of the current embarrassing situation inwhich natural language is used both as a tool and as an object in study and thus freesthem from the bondage of natural language.Nonetheless, further research indicates that although DRT has successfully resolvedthe referential anaphora and some sorts of quantificational anaphora including that ofdefinite noun phrases and indefinite noun phrases, it fails to give an effectiveinterpretation to those types of deep-anaphora which require considerations of pragmaticinferences. The results of this study show that the DRS ought to fulfill a dual function, asa mental model of the discourse on the one hand and as a mental model of the world on the other. Unfortunately, DRT has not yet given the two separate considerations and thusgreatly reduced its interpretation force with respect to deep anaphora resolution. Withoutgoing into formal details, however, it is not difficult to be convinced that the DRTapproach suffers from the lack of generality. Upon careful reflection of the problemsrevealed in DRT, I suggest that DRT model is obviously an oversimplication in boththeory and practice comparing with the complexity inherited from natural language andthus a major revision of DRT is called for. In view of the above, the present research setsabout extending the original DRT to a Two-Dimensional Discourse RepresentationStructure Model (2D-RS for short) to give the mental model of the world in DRT a duediligence.The theoretical innovation of this dissertation is that at baseline, the researchprovides a supplementary way to remedy the failures by separating the two differentroles played by the original DRS and to introduce yet another level of representation, alevel of knowledge-based representation structure (KRS for short) to accommodate themental model of the world and to use the original DRS only as a discourse model. KRSin2D-RS model is constructed on the basis of stereotypical relation. By employingabductive reasoning, the relation between an anaphor and its external anchor isestablished. By doing so,2D-RS has successfully depicted the way an addresseecompensates the implicated information conveyed in verbal communication throughpragmatic inferences.As for the empirical innovation of the dissertation, first of all, stereotypical relationand abductive reasoning are introduced into DRT as a natural deductive system in whichdiscourse anaphora is resolved through the identification of pragmatic inference. Itbroadens the original binding resolution in DRT by introducing a new concept ofaccommodation resolution. Secondly, semantic presupposition is treated as a necessarycondition in judging the truth value of a discourse, which perfects the original modeltheoretic semantic interpretation in DRT. Finally, the CR.PRO is amended by referring tothe Binding Principles in Chomsky’s Government and Binding Theory.In terms of methodology, our research adopts the analytical method of naturallanguage which is commonly used by analytic philosophy. To be more specific, takingChomsky’s “methodology of naturalism” as the model, we employ the method of hypothesis-testing based on formal mathematics formula and deductive reasoning inexploring the nature of anaphora resolution in our research.