ABSTRACT
This paper1 surveys ParseTalk, a grammar model for natural language analysis that combines lexical organization of grammatical knowledge with lexicalized control of the corresponding parser in an object-oriented specification framework. Our research takes into account recent developments in the field of linguistic grammar theory that have yielded a radical lexical decomposition of linguistic knowledge. We consider this rigid form of lexical modularization just as a starting point for the incorporation of lexicalized control at the grammar level. Such a local, distributed perspective on
the behaviour of the grammar system contrasts with previous designs of lexicalized grammars (e.g. HPSG (Proudian & Pollard 1985), Lexicalized TAG (Schabes et al. 1988), Categorial Grammar (Hepple 1992)). In these approaches lexical items are considered as passive data containers whose contents are uniformly interpreted by globally defined operations (unification, tree adjunction, functional composition, etc.). Trying to exploit fully the potential of the object-oriented computation paradigm and thus pushing lexicalism to its limit, we assign complete procedural autonomy to lexical units. They are treated as active lexical processes, locally communicating with each other by message passing and dynamically establishing heterogeneous communication lines in order to determine each lexical item's functional role in the discourse. While the issue of lexicalized control has long been investigated in the paradigm of conceptual parsing (Riesbeck & Schank 1978), and in particular word expert parsing (Small & Rieger 1982), these approaches are limited in several ways. Systematic grammar development is hard as there is no common grammar theory available, generalizations over sets of lexical items cannot be expressed because of the lack of inheritance mechanisms, and lexical communication is based on entirely informal protocols that have no grounding in principles of distributed computing.
