Designing the Human Machine Interface in the ATIS Domain

Designing the Human Machine Interface in the ATIS Domain B. Bly P. J. Price S. Park S. Tepper SRI International 333 Ravenswood Ave. Menlo Park, CA 94025 E. Jackson V. Abrash Abstract Spoken language systems for the near future will not handle all of English, but, rather, will be limited to a domain-specific sub-language. Accurate modeling of the sub-language will depend on analysis of domain-specific data. Since no spoken language systems currently have a wide range of users, and since variability across users is expected to be large, we are simulating applications in which a large population of potential users can be sampled. The data resulting from the simulations can be used for system development and for system evaluation. The application discussed here is the air travel domain using the Official Airline Guide (OAG) reformatted in a relational structure. This study assesses the effects of changes in the simu- lations on the speech and language of the experimental subjects. These results are relevant to both the exper- imental conditions for data collection and the design of the human interface for spoken language systems. We re- port here on five experiments: (1) the effect of longer in- structions with examples vs. shorter instructions, using our earlier data collection system, (2) a baseline experi- ment using a functional equivalent of the data collection effort at Texas Instruments (TI), (3) the use of a more specific version of the scenario used in the baseline ex- periment, (4) the use of a short, simple familiarization scenario before the main scenario, and (5) in addition to the short familiarization scenario, the use of a finite vocabulary with rejection of sentences with extra-lexical items. Introduction The data reported here are part of an endeavor whose goal is to design an appropriate human-machine inter- face by examining various parameters in a simulated in- teraction involving air travel planning. The design of the system is such that either a spoken language system (SLS) or a simulation of one can be inserted between the user and the relational database version of the Offi- cial Airline Guide data for North American flights and fares. In this way we can gather data for development and evaluation of both the SLS and the user interface. Perhaps the greatest source of variability in the system is that across subjects. Individuals differ greatly in their language skills, in their problem solving skills, and in their attention spans. It is therefore important to sam- ple a variety of subjects from the relevant population. Individuals are also very adaptable. In many cases, it may be easier to rely on subject adaptability than to try to find technological solutions. However, the dimensions along which humans might adapt are largely unknown for spoken language interfaces. Thus, the simulations provide us with a mechanism to test experimentally var- ious interface strategies that may be appropriate for SLS technology as it develops. We describe here five experiments aimed at answering various questions about the interface. Our first exper- iment, the …

Content truncated to 500 words