CLIN 2005 Abstracts
  • Some notes on referential modes and reference tracking criteria in automatic (pro)nominal anaphora
    Noureddine Elouazizi (Leiden University Centre for Linguistics (LUCL))
    Despite the existence of a great deal of important works on (pronominal) anaphora resolution and a plethora of models which attempt to formalize coreferential relations (Cf. Ariel (1990); Grosz Et al. (1995); Mitkov (1998, 2002); Byron (2002); Strube and Muller (2003), among many others), yet many areas of referential resolution in natural language remain poorly understood and accordingly poorly formalized for the sake of engineering and implementation within different NLP systems (e.g. dialogue systems, Questions/Answering systems etc.). For example, the engineering of the ([pro]nomainal) coreferential relations in NLP systems leaves much to desire as their underlying formal models do not always mirror the conceptual structure presupposed by the [pro]nominal system of natural languages. This paper is a theoretical and empirical contribution towards narrowing the gulf which separates the conceptual structure of the [pro]nominal system in natural language from its engineering in NLP systems with respect to reference tracking in [pro]nominal anaphora. I propose that the design of referential algorithms to resolve [pro]nominal coreference and automatically determine its antecedent needs to: (i) make a clear distinction between referential [pro]nominal types and modes, (ii) step on a taxonomy for antecedent indicators. The feature structure of this taxonomy need not be prescribed but rather it is made to follow naturally from formalizing the semantico-pragmatic relations that the [pro]nominal elements encode and instantiate.