Six Month Plan
Continued refinement of Mokusei components
- Improve translation of system responses, for more natural phrasing
- Expand language understanding coverage, and add robust parsing
- Develop initial speech recognition capability using available telephone-quality corpora and seeding from Jupiter models
- Perform read-speech data collection from native speakers in U.S. and Japan, using 600 sentence set to create initial templates
- Begin wizard-based data collection efforts to obtain spontaneous speech for improved language modeling and understanding