The MAYZ Toolkit provides tools to support the corpus-oriented development of lexicalized grammars. Corpus-oriented development is a new methodology of developing wide-coverage lexicalized grammars. In conventional grammar development, grammar writers build a large lexicon manually, while in our method we first make a derivation bank (derivbank). A derivbank is a treebank of a target grammar theory, e.g., HPSG treebank. It includes real sentences annotated with its syntactic structure. Annotations must conform to grammar rules (principles and schemas) assessed by the grammar theory. When derivation structures are properly assigned to real sentences, a lexicon is extract from them.
The MAYZ Toolkit helps the development of a derivbank, and extracts a lexicon from the derivbank. Users can make a large lexicalized grammar only by writing grammar rules (schemas) and heuristic rules for annotation. In addition, it provides tools to make disambiguation models based on maximum entropy models. You can make probabilistic disambiguation models only by writing pattern rules of extracting features of maximum entropy models.
The package also includes a general-purpose parser (Unification Parser, UP). This parser supports beam thresholding using probabilistic models. Users can readily apply grammars developed by MAYZ to real tasks.