Progress to June 1999
Language generation:
- Created initial translation capability for weather domain
- Estimate 70% correct translation of weather reports, with remainder being mostly comprehensible, but slightly ill-formed
- Required minor alterations to original meaning representation structure (e.g., preposition ‘in’ in locative vs temporal reference)
Data collection:
- Created initial 600 sentence set in Mokusei domain by translating English queries, and soliciting examples from NTT colleagues
Language understanding:
- Developed initial context-free grammar to parse 600 sentence set
- Estimate 50% of sentences are parsable, with remainder consisting of both out-of-domain queries and more complex queries which we expect to parse in the future