Progress Through December 1999
- Initial Mokusei prototype completed:
- Speech recognition (ASR) (2200 word vocabulary)
- Japanese phonological rules created
- Interpolated English and Japanese acoustic models
- Conventional word class and NL-derived n-grams explored
- Natural language (NL) understanding
- Improved coverage on training data (95% of 2400 utts)
- “Trace” mechanism used for faster parsing
- Additional 50 Japanese cities added to database
- 1000 spontaneous utterances collected from native speakers interacting with system