HPSG formalism in Tokyo University

Tsujii Laboratory in Tokyo University is working on the development of a framework and an environment based on Head-Driven Phrase Structure (HPSG) formalism, in order to enable applications of NLP like knowledge acquisition, machine translation or information extraction.

HPSG formalism is almost not used for practical applications because it is less efficient that other approaches, using for instance context-free grammars. However, it has several advantages like minimizing the size of the grammar, which only contains 6 rules for the Japanese grammar used in the laboratory of Professor Tsujii, and maximizing the role of lexical information.

The parser uses a two-step structural analysis:

• first analysis enumerates possible structures, using CFG rules compiled from HPSG lexical rules;
• then the parser solves the constraints that are not contained in the CFG grammar.

Several points were under development in 1998 in order to foster the use of HPSG formalism:

• creation of finite-state automata;
• creation of parallel algorithms to increase the speed;
• information extraction tools;
• automatic creation of a thesaurus.

The research on HPSG involves several collaborators, especially in Japan (e.g. EDR), in the USA (e.g. Stanford University) or in Europe (e.g. DFKE in Germany).