Sixth International Conference on Spoken Language Processing
(ICSLP 2000)

Beijing, China
October 16-20, 2000

Grammar Partitioning and Parser Composition for Natural Language Understanding

Po Chui Luk (1), Helen Meng (1), Filung Wang (2)

(1) Human-Computer Communication Laboratory, Chinese University of Hong Kong
(2) Intel China Research Center, China

This paper presents an approach for natural language understanding, which integrates multiple sub-grammars and sub-parsers; in contrast with the traditional single grammar and parser approach. The use of GLR(k) parsers for natural language understanding is hampered by the problem of exponential growth of the parsing table size as the size of grammar rules increases. Hence, we propose to partition a grammar into multiple sub-grammars. For each sub-grammar we generate its own parsing table together with its specialized GLR sub-parser. The total size of the sub-grammars parsing tables is much smaller than the size of the parsing table of the unpartitioned grammar. A parser composition algorithm then combines the sub-parsers' outputs to produce an overall parse that is identical to that produced by a single parser. Results based on natural language queries in the Air Travel Information System (ATIS) domain shows that this is a viable and efficient approach applicable to both English and Chinese.


Full Paper

Bibliographic reference.  Luk, Po Chui / Meng, Helen / Wang, Filung (2000): "Grammar partitioning and parser composition for natural language understanding", In ICSLP-2000, vol.3, 486-489.