Grammar Modularity and its Impact on Grammar Documentation

Grammar Modularity and its Impact on Grammar Documentation Universit¨at PotsdamStefanie Dipper Humboldt-Universit¨at zu Berlin Institut f¨ur Linguistik Inst. f¨ur deutsche Sprache und Linguistik D-14415 Potsdam D-10099 Berlin Germany Germany dipper@ling.uni-potsdam.de stefanie.dipper@rz.hu-berlin.de Abstract This paper addresses the documentation of large-scale grammars.1 We argue that grammar implementation differs from ordinary software programs: the concept of modules, as known from software engineering, cannot be trans- ferred directly to grammar implementations, due to grammar-specific properties. These properties also put special constraints on the form of grammar documentation. To fulfill these constraints, we propose an XML-based, grammar-specific documentation technique. 1 Introduction Research in the fi eld of grammar development fo- cuses on grammar modularization, ambiguity man- agement, robustness, testing and evaluation, main- tainability and reusability. A point which has of- ten been neglected is the detailed documentation of large-scale grammars—despite the fact that thor- ough documentation of the grammar code is a pre- requisite for code maintainability and reusability. In this paper, we argue that documenting large- scale grammars is a complex task that requires spe- cial, grammar-specifi c documentation techniques. The line of reasoning goes as follows. We show that maintainability (and, hence, reusability) of a grammar depends to a large extent on the modular- ization of the grammar rules: a large-scale gram- mar remains maintainable only if linguistic gener- alizations are encoded explicitly, i.e., by modules (sec. 3.1). However, in contrast to modules in ordi- nary software programs, (certain) grammar modules cannot be black boxes (sec. 3.2). This property puts special constraints on the form of grammar docu- mentation (sec. 4). Finally, we present an XML- based documentation technique that allows us to ac- comodate these constraints (sec. 5). 1The paper is based on my doctoral dissertation (Dipper, 2003), which I wrote at the IMS Stuttgart. I am very grateful to Anette Frank for invaluable discussions of the dissertation. Many thanks go to Bryan Jurish and the anonymous reviewers for helpful comments on the paper. To illustrate the needs of documentation, we refer to a German LFG toy grammar (Lexical-Functional Grammar, cf. sec. 2). Our argumentation, however, applies not only to grammars in the LFG formalism but to any grammar that is modularized to a certain extent. 2 Lexical-Functional Grammar LFG is a constraint-based linguistic theory (Bresnan (2001), Dalrymple (2001)). It defi nes different lev- els of representation to encode syntactic, semantic and other information. The levels that are relevant here are constituent structure (c-structure) and functional structure (f- structure). The level of c-structure represents the constituents of a sentence and the order of the ter- minals. The level of f-structure encodes the func- tions of the constituents (e.g. subject, adjunct) and morpho-syntactic information, such as case, num- ber, and tense. The c-structure of a sentence is determined by a context-free phrase structure grammar and is repre- sented by a tree. In contrast, the f-structure is rep- resented by a matrix of attribute-value pairs. The structures are linked by a correspondence function (or mapping relation), called “φ-projection”. The (simplifi ed) analysis of the …

Content truncated to 500 words