Automating Reuse of Broad-Coverage Typed Feature Structure Grammars

2. Background Since the beginning of the Chomskyan tradition, grammar writers and implementors have taken up the challenge to produce as accurate and covering monolingual grammars as possible. Often, this has taken place within certain theoretical and practical frameworks, more or less relating to existing linguistic theories, sometimes even contributing to language theory. For computational applications, the obvious and natural strife to achieve broad coverage and use representations which facilitate efficient implementations has to some extent overshadowed other important aspects of monolingual grammars. The question of how much of a computational grammar is parochial, and how much belongs to a universal “core” is often overlooked. Although our scientific communities may agree on a certain collection of formal elements which constitute typed feature structure grammars (TFSG) (subsumption relations of type hierarchies, appropriateness conditions), and this collection of formal elements would prove adequate to describe relevant properties of every single language in the world, it is not given that TFSG constitutes an adequate form of representation in order to distinguish a common core from the rest of the formal elements. On the one hand, TFSG may not be fine-grained enough. On the other hand, the core grammar might be expressed as the combination of a huge type hierarchy allowing every possible kind of multiple inheritance, and a large disjunction of language-specific constraints, leaving it as a mathematical task to discover cross-linguistic generalizations by factoring out repeating terms.