CGN to Grail: Extracting a Type-logical Lexicon From the CGN Annotation

The tag set for the CGN syntactic annotation is designed in such a way as to enable a transparent mapping to the derivational structures of current ‘lexicalized’ grammar formalisms. Through such translations, the CGN tree bank can be used to train and evaluate computational grammars within these frameworks. In this paper we will discuss some preliminary work on the mapping between the CGN annotation graphs and the proof net format of the Grail parser/theorem prover (Moot 2001, Moot 1999). Grail is a general grammar development environment for typelogical categorial grammars (TLG, (Moortgat 1997, Morrill 1994, Carpenter 1998)). To a large extent, there is a straightforward transfer between the type-logical format and the analyses provided by other lexicalized grammar formalisms such as LTAG (lexicalized Tree Adjoining Grammars, (Sarkar 2001)) and MG (computational versions of Minimalist Grammars, (Stabler 1997)). An attractive feature of TLG, which is not shared by these other frameworks, is its full support for hypothetical reasoning. In this paper, we exploit the hypothetical reasoning facilities to extract a type-logical grammar from the CGN annotation graphs. This task can be naturally divided in two subtasks. The first of these consists in solving type equations: in the TLG setting this means breaking up the CGN annotation graph into the subgraphs that correspond to lexical type assignments. In the presence of discontinuous dependencies, the lexical type assignments will not always be compatible with surface word order. The second subtask then consists in calibrating the lexicon in such a way that it has controlled access to the structural reasoning component of the grammar.