Objective:

Discussions with NASA astronauts, trainers, management, and research groups has revealed a huge potential demand for spoken language interfaces. Complex spoken dialogue systems are becoming available, but construction of a language model for acceptable utterances still requires highly specialized expertise. The de facto standard method is to write a context free grammar (CFG) with semantic annotations that associate a meaning representation with each in-coverage utterance. Commercial platforms such as the Nuance Toolkit can then compile CFGs into efficient recognizers. Example-based methods can simplify the language modeling, as well as maintenance and porting to new domains. REGULUS, an Open Source project spearheaded by the RIALIST group, is developing an advanced toolkit to support this approach. Initial results have been extremely encouraging, and REGULUS has already been employed in one major NASA application. Investigators will now develop technology -- integrated with REGULUS -- to make language modeling accessible to domain experts who are not familiar with computational linguistics. This will include documentation, tutorial materials, and implemented examples. Work will be carried out in collaboration with other NASA groups who wish to develop spoken language interfaces.

Applications:

Hands-free information retrieval and commanding; automated support for checklist procedure execution.

NASA Benefit:

The Robonaut and Space Operations Computing (SpOC) groups at JSC have declared interest in working with this research. Many other groups will want spoken dialogue interfaces once the technology is sufficiently easy to implement. It has revolutionary potential benefits for safe and productive use of automation and for comprehensive knowledge creation, access, and sharing. Spoken interfaces to automated agents will be especially useful to busy or suited astronauts.

Keywords:

spoken dialogue understanding, speech interface, grammar construction, language models

Prior Technology:

Expensive hand construction of large grammars; difficulty with maintenance and with porting to new applications.

Contacts:

Beth A. Hockey (PI), UC Santa Clara at ARC.