Anette Frank, Tracy Holloway King, Jonas Kuhn, John Maxwell
Peter Sells (ed): Optimality Theoretic Syntax, CSLI Publications (to appear)
Broad-coverage computational (LFG) grammars are faced with problems of massive ambiguity and robustness
in the context of large-scale natural language processing. To confront these problems, we propose an
extension of the LFG projection architecture that incorporates ideas from the theoretical literature on Optimality
Theory (see Prince and Smolensky (1993), Bresnan (1996, 1998)). In this approach, a new projection,
o-structure, is defined on top of the classical constraint system of existing LFG grammars. For each reading
of a sentence, the o-structure will contain a record of the violable constraints applied to construct the analysis
for this reading. A dominance ranking of the violable constraints determines the winner among the competing
candidate analyses. The Optimality constraints are overlaid on the existing grammar and hence do not alter
the basic tenets of LFG theory. We discuss various usages of this Optimality Theory-style constraint ranking
mechanism in large-scale grammar processing and discuss how our approach relates to the framework of
Report number: