Developing Tree-Adjoining Grammars with Lexical Descriptions

LTAG is an appealing formalism for representing various phenomena (especially syntactic phenomena) in natural languages because of its linguistic and computational properties such as the Extended Domain of Locality, stronger generative capacity and lexicalized elementary trees. Because templates (i.e., elementary trees with the lexical items removed) in an LTAG grammar often share some common structures, as the number of templates increases building and maintaining templates by hand presents two major problems. First, the reuse of tree structures in many templates creates redundancy. To make a single change in a grammar, all the related templates have to be manually checked. The process is inefficient and cannot guarantee consistency (Vijay-Shanker and Schabes, 1992). Second, the underlying linguistic information (e.g., the analysis of wh-movement) is not expressed explicitly. As a result, from the grammar itself (i.e., hundreds of templates plus the lexicon), it is hard to grasp the characteristics of a particular language, to compare languages, and to build a grammar for a new language given existing grammars for other languages. To address these problems, we designed a grammar development system named LexOrg, which automatically generate LTAG grammars from abstract specifications. The system is based on the ideas expressed in (Vijay-Shanker and Schabes, 1992), for using tree descriptions in specifying a grammar by separately defining pieces of tree structure that encode independent syntactic principles. Various individual specifications are then combined to form the elementary trees of the grammar. We have carefully designed our system to be as language-independent as possible and tested its performance by constructing both English and Chinese