The next frontier for computers is the automated personal assistant that can interact with people in intuitive and natural terms. One of the key capabilities to enable this is robust spoken language communication. Current techniques for automated natural language understanding, however, all depend on the notion of a complete sentence as the core unit on communication. This assumption makes sense for text understanding applications, but is very unnatural in spoken language applications. Human-human spoken communication occurs incrementally, using sentence fragments, with corrections, elaborations, and confirmations to enable efficient coordinated action. This Small Grant for Exploratory Research (SGER) explores the possibility for new models of incremental processing that redefine the basic communicative unit of communication and develop new understanding techniques that use these units.
One of the key ideas underlying this work is that communicative units should be defined in terms of identifying the speaker intentions (as opposed to the traditional view of analyzing language in terms of its structural properties). This project is exploring the feasibility of developing new computational models that identifying intentions incrementally. Exploratory testing is accomplished using the TRIPS natural language understanding system and comparing the performance of incremental vs non-incremental processing models.
This SGER will lay the foundations for significantly more natural human-computer interfaces in a new generation of automated personal assistants that will revolutionized how computers are used, allowing computers to adapt to people rather than forcing humans to adapt to computers.
