Testing a Grammar Customization System with Sahaptin

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 254–262, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Testing a Grammar Customization System with Sahaptin Scott Drellishak University of Washington Seattle, WA, USA sfd@u.washington.edu Abstract I briefly describe a system for automatically creating an implemented grammar of a natu- ral language based on answers to a web-based questionnaire, then present a grammar of Sa- haptin, a language of the Pacific Northwest with complex argument-marking and agree- ment patterns, that was developed to test the system. The development of this grammar has proved useful in three ways: (1) verifying the correct functioning of the grammar cus- tomization system, (2) motivating the addition of a new pattern of agreement to the system, and (3) making detailed predictions that un- covered gaps in the linguistic descriptions of Sahaptin. 1 Introduction The LinGO Grammar Matrix (Bender et al., 2002) is a resource for building im- plemented precision HPSG (Pollard and Sag, 1994) grammars of natural languages. Grammars based on the Matrix are expressed in the Type Description Language (TDL) (Krieger and Schäfer, 1994), are interpretable by the Linguistic Knowledge Building system (LKB) (Copestake, 2002) (a software tool for developing constraint-based grammars), and have semantic representations that are compat- ible with Minimal Recursion Semantics (MRS) (Copestake et al., 2005). The Grammar Matrix project, in particular the customization system described below, has drawn on the linguistics and linguistic typology literature during its develop- ment; the system is now complex enough that it is capable making contributions back to linguistics. 1.1 Matrix Customization System In its earliest form, the Matrix provided a set of pre- defined types intended to give grammar engineers a head start, allowing them to avoid duplicating the effort required to develop analyses of linguistic structures thought to occur in all languages. How- ever, there exist many linguistic phenomena that are widespread, but not universal. If the Matrix were restricted to supporting only what is truly univer- sal, it would be a much less useful resource for grammar-writers working on languages containing such non-universal phenomena. Our solution has been to provide the Matrix customization system, which presents a linguist with a web-based typologi- cal questionnaire designed to elicit a description of a target language and, based on it, automatically pro- duce a grammar that parses and generates the target language.1 The grammars produced are not encum- bered by phenomena that do not occur in the target language; rather, they contain just enough complex- ity to model it as described. Although the grammars produced by the customization system are intended as a starting point for further grammar engineering, that starting point is now far enough along that even without enhancement the grammars can be used for interesting linguistic work. The customization system is conceived of as con- sisting of a set of libraries, each of which sup- ports a particular linguistic phenomenon, and in- cludes a section of the questionnaire and a syntac- tic analysis of …

Content truncated to 500 words