Transparent combination of rule-based and data-driven approaches in speech understanding

Transparent combination of rule-based and data-driven approaches in a speech understanding architecture Manny Rayner and Beth Ann Hockey RIACS, Mail Stop T27A-2 NASA Ames Research Center Moffett Field, CA 94035-1000, USA fmrayner, bahockeyl@riacs.edu Abstract We describe a domain-independent se- mantic interpretation architecture suit- able for spoken dialogue systems, which uses a decision-list method to effect a transparent combination of rule-based and data-driven approaches. The ar- chitecture has been implemented and evaluated in the context of a medium- vocabulary command and control task. 1 Introduction As the field of spoken language understanding be- comes more mature, a clearer picture begins to emerge of the tradeoffs between rule-based and data-driven methods. Other things being equal, there are many reasons to prefer data-driven ap- proaches. They are more robust, and reduce the heavy authoring costs associated with rule-based systems; methods are moreover starting to emerge which enable data-driven approaches to be used in areas which previously were thought to require rules, such as dialogue management. A good overview of current work in this area is provided in (Young, 2002). Excellent as data-driven systems are, they have one obvious drawback: they require corpus data, usually in fairly substantial amounts. Academics basically interested in pure research are free to work within a domain for which the data has al- ready been collected, and for example can decide to use the Penn Treebank (Penn Treebank, 2002) or the ATIS corpus (Dahl et al., March 1994). If, on the other hand, the goal is to create a use- ful application for a specified new domain, there will in general be little or no available data at the start of the project. It is possible to create the data by using Wizard of Oz methods, or sim- ilar. Wizard of Oz data collection is unattrac- tive for many reasons: it is expensive and time- consuming, and once the data has been collected it is not easy to change the coverage of the system. For these reasons, commercial speech recognition platform vendors like Nuance and SpeechWorks have focussed on rule-based approaches, which al- low rapid prototyping of systems from only very modest quantities of corpus data. Although most commercial rule-based spoken language dialogue systems use directed dialogue strategies and mod- erately simple recognition grammars, the literature now contains descriptions of several research sys- tems built using rule-based methods, which suc- cessfully use mixed-initiative strategies and com- plex grammars (Stent et al., 1999; Rayner et al., 2000; Lemon et al., 2001; Rayner et al., 2001b). If a project of this kind is developed over a substantial period of time, corpus material accu- mulates automatically as input to the system is logged. The more corpus material there is, the stronger the reasons for moving towards data- driven processing; this will however only be easy if the architecture is originally set up to use statis- tics as well as rules. Summarising the argument so far, we would like an architecture which com- bines rule-based and data-driven methods as trans- parently as possible. …

Content truncated to 500 words