The progress of NLP applications in this decade has been mainly accomplished by the rapid development of corpus-based and statistical techniques, while rather simple techniques have been used as far as the structural aspects of language are concerned.
In this paper, we will discuss how we can combine more sophisticated, linguistically elaborate techniques with the current statistical techniques and what kinds of improvement we can expect from such an integration of different knowledge types and methods.
This apparent progress in spoken language technology has been fuelled by a number of developments: the relentless increase in desktop computing power, the introduction of statistical modelling techniques, the availability of vast quantities of recorded speech material, and the institution of public system evaluations.
However, our understanding of the fundamental patterning in speech has progressed at a much slower pace, not least in the area of its high-level linguistic properties. Spoken language understanding continues to be an elusive goal, and the prosodic linkage between acoustic and linguistic patterning is still something of a mystery.
This talk will illuminate these issues, and will conclude with an analysis of the options for future spoken language R&D.