Lightweight natural language processing

Device interfaces that can understand freeform input have long been the focus of both research and commercial effort. In recent years, many such interfaces have been produced with varying degrees of sophistication and capability. At the same time, devices have become ever smaller and relatively constrained in resources. A trend towards miniaturization and ubiquity has produced a class of devices which have far lesser processing power and memory than a conventional desktop machine; but are nonetheless expected to perform a natural language understanding task equivalent to a high performance computer. This has brought about the need for lightweight natural language processing optimized for resource efficiency, yet capable of providing acceptable results when used in ubiquitous devices. In this paper, we present our work on building a lightweight natural language processing engine for ubiquitous devices found around the household. Ubiquitous devices that are specifically targeted include televisions functioning as personal video recorders and refrigerators. Novel features of the natural language engine include an integrated scripting language for rule definition and the use of a semantic distance as a viable substitute for processor intensive natural language processing operations.