Professor Manabu Okumura
Precision and Intelligence Laboratory
Advanced Information Processing Division: Okumura Group
Tokyo Institute of Technology
NOTE: Japanese page is here.
Section: Natural Language Processing, Automated Text Summarization,
Computer Assisted Language Learning, Text Data Mining
Objective: Development of the technique of natural language processing
and its application systems
- Current Topics:
- Incremental Language Understanding Model(Robust Semantic and Discourse
- Automated Text Summarization
- Development of Communication Assistive Technology for People with
- Animation Control through Natural Language Understanding
Natural Langauge Processing, Natural Language Understanding,
Our research focuses on the development of a real-time and user-friendly speech
dialogue system. We are currently working on fundamental issues for it: incremental
interpretation, robustness improvement, and lexical acquisition.
- 1. Incremental interpretation model for Japanese
- In a real-time dialogue system, texts need to be incrementally analyzed
while they are inputed. We present such an interpretation model that uses
syntactic, semantic, contextual, and commonsense knowledge cooperatively.
In particular, those systems are under development which performs syntactic
and semantic analysis in integrated way using case frame information of verbs,
and which resolves word sense ambiguity and finds coherence between sentences
simultaneously in terms of associativity between words.
- 2. Ill-formed input analysis
- A user-friendly dialogue system should be robust and flexible in that it
can analyze any user inputs with various constructions and unlimited vocabulary.
Therefore, it needs to be able to cope with ill-formed sentences. In Japanese
ellipses are typical and so for a start we present a method to fill the gaps
for various types of ellipses.
- 3. Semi-automatic lexical acquisition
- Large scale dictionaries are indispensable for practical natural language
understanding systems. However, because it is considered to be difficult to
construct a large lexicon by hand, it is an important challenge to extract
lexical information semi-automatically from text data such as large corpora
and machine readable dictionaries. Now those projects are in progress where
information of associative relation between verbs and between nouns is extracted
from a machine readable dictionary of verbs and adjectives respectively. Semi-automatic
lexical acquisition from bilingual tagged corpora is also planned.
- ``Towards Incremental Disambiguation with a Generalized Discrimination
Network''; Proc. of the Eighth National Conference on Artificial Intelligence,
- ``Incremental Analysis of Japanese Dependency Relations with a Generalized
Discrimination Network''; Proc. of the second Pacific Rim International Conference
on Artificial Intelligence, pp.787-793(1992).
- ``Towards Japanese Ellipsis Resolution with a Generalized Discrimination
Network''; AI'92 Proc. of the 5th Australian Joint Conference on Artificial
Intelligence, A. Adams and L. Stirling eds.), World Scientific, pp.212-217,
- ``Word Sense Disambiguation and Text Segmentation Based on Lexical Cohesion'';
Proc. of the 15th International Conference on Computational Linguistics, pp.755-761,
1994. Postscript File (60 K)
- ``Word Sense Disambiguation by Marker Passing on Very Large Semantic Networks'';
Proc. of the Natural Language Processing Pacific Rim Symposium'95, pp.71-76,
1995. Postscript File (57 K)
- ``Zero Pronoun Resolution in Japanese Discourse Based on Centering Theory'';
Proc. of the 16th International Conference on Computational Linguistics, pp.871-876,
1996. Postscript File (50 K)
Okumura Lab. Homepage 1999-2002 Copyright Okumura Lab.
Please send any requests and comments to the following address: