Conclusion

The results given above lead to the following conclusions.

• Sophisticated grammatical analysis is fast enough for practical spoken dialogue systems.
• Moreover, grammatical analysis is effective for the purposes of the present application. Almost all user utterances can be analysed correctly. This is somewhat surprising. Typically, linguistic ambiguities are a major obstacle for practical NLP systems. The current application is very simple in the sense that such linguistic ambiguities do not seem to play a significant role. The ambiguities introduced by the speech recognizer (as multiple paths in the word-graph) are a far more important problem.
• Grammatical analysis does not seem to help much to solve the problem of disambiguating the word-graph. The best method incorporating grammatical analysis performs about as well as a method which solely uses N-gram statistics and acoustic scores for disambiguation of the word-graph. However, in the latter case grammatical analysis of the type proposed here is useful in providing a robust analysis of the best path.

We have argued in this paper that sophisticated grammatical analysis in combination with a robust parser can be applied successfully as an ingredient of a spoken dialogue system. Grammatical analysis is thereby shown to be a viable alternative to techniques such as concept spotting. We showed that for a state-of-the-art application (public transport information system) grammatical analysis can be applied efficiently and effectively. It is expected that the use of sophisticated grammatical analysis will allow for easier construction of linguistically more complex spoken dialogue systems.