A Natural Language Interface for Mobile Devices

Humans express their needs almost effortlessly in natural language, and for this reason, constructing machines that can reliably respond to natural language requests has been a longstanding and significant goal in the design of intelligent systems. Early successes—for example, the LUNAR system of Woods, Kaplan, and Nash-Webber (1972); the LADDER system of Hendrix, Sacerdoti, Sagalowicz, and Slocum (1978); and the PLANES system of Waltz (1978)—relied on the use of application-specific grammars to encode various constraints of particular domains or datasets. While this approach enabled these systems to respond to a range of requests in their targeted domains, the systems could not easily be ported to work in related or expanded domains, and they had limited coverage of vocabulary and syntactic constructions used in expressing requests. More recently developed systems—such as those described in Harabagiu, Maiorano, and Pasca (2003); Katz (1997); Nyberg, Burger, Mardis, and Ferrucci (2004); and Weischedel, Xu, and Licuanan (2004)—address multiple domains and employ general-purpose grammars or statistical language interpretation to handle the variety of requests and phrasings of requests that might be submitted by users. With more and more information sources available in networked contexts, plus mobile devices acting as sources and targets of requests, recent systems have adopted a more distributed architecture. In this chapter, we first discuss some of the primary issues related to the design and construction of natural language interfaces, and in particular, interfaces to mobile devices. Then, we describe two systems in this space: the START information access system and the StartMobile natural language interface to mobile devices. Finally, we discuss recently deployed commercial systems and future directions.