In Support Of The Equal Rights Movement For Literal And Figurative Language -- A Parallel Search And Preferential Choice Model

is more stringent, however, as it takes other concepts apart from A.C into account. All candidates, whether literal or not, are processed in parallel. The interpretation, which fits best with the discourse, is selected by preferential choice (cf. Table 1), a criterion not spelled out by Recanati. The overwhelming majority of approaches to metonymy processing follow some variant of the In Fass' s and Norvig' s work figurative interpretations are only triggered when a sortal conflict is encountered, thereby explicitly subscribing to a two-stage model. In Hobbs et al.' s work the literal interpretation of a sentence is regarded as a special case of metonymic resolution , thus making a unified treatment of literal and figurative language possible, in principle. Nevertheless, the special case of the literal interpretation of a sentence is strictly preferred when no sortal conflicts are encountered, regardless of discourse constraints. Pustejovsky and Verspoor both concentrate on logical metonymy, which apart from semantic effects has also syntactic ones. Nevertheless, type coercion is regarded as a typically semantic operation applied after encountering sortal conflicts (see, e.g., Pustejovsky (1991), pp.425). In all these studies, the interaction of anaphora and metonymies is not an issue, though some of them (Norvig, 1989; Hobbs et al., 1993) consider nominal anaphora in a text understanding system framework like ours. Consistent with our goal to avoid overgeneration, lexical semanticists like Verspoor (1996) or Godard & Jayez (1993) have proposed additional, highly sophisticated constraints on coercion they assign to the lexicon. Verspoor argues that purely pragmatic constraints cannot be sufficiently strict or explain language-specific differences. This argument cannot be defeated in principle, but we have shown that there are at least a range of pragmatic regularities, which do not require additional modelling in the lexicon but mirror general constraints at the knowledge level and are worth being considered when constraining coercion. Supplementing them with lexical and language-specific constraints is a worthwhile direction for future work. Conclusion We have introduced a model of metonymy resolution that treats metonymies on a par with literal interpretations. Furthermore , the mutual influence metonymies and anaphora have upon each other in noun phrases is adequately accounted for. In particular, rejecting the LMF hypothesis in favor of a parallel approach leads us to a more effective procedure. Our empirical data have led us to several follow-up questions. In particular, the high percentage of non-anaphoric noun phrases not marked by a sortal …