Integrating Finite-state Technology with Deep LFG Grammars1

Researchers at PARC were pioneers in developing finite-state methods for applications in computational linguistics, and one of the original motivations was to provide a coherent architecture for the integration of lower-level lexical processing with higher-level syntactic analysis (Kaplan and Kay, 1981; Karttunen et al., 1992; Kaplan and Kay, 1994). Finite-state methods for tokenizing and morphological analysis were initially incorporated into the Grammar-writer’s Workbench for Lexical Functional Grammar (LFG) (Dalrymple, 2001; Kaplan and Bresnan, 1982), a relatively complete LFG parsing system but for relatively small-scale grammars (Kaplan and Maxwell, 1996). Finite-state transducers, also of a relatively small scale, were constructed automatically in this system from specifications provided by the LFG grammar writer, using the compilation techniques described by (Kaplan and Kay, 1994).