Speech and corpora

The analysis of spontaneous speech is of paramount importance. It allows focusing on pragmatic, syntactic, prosodic and information structures and leads to a better understanding as how they interact in actual speech. Indeed, many applications and industrial development in speech synthesis and recognition badly need coherent models to be integrated in their software, whereas today even successful systems rely mainly on word spotting if recognized speech does not simply results from oral written text reading. Other applications in oral language learning are also important by departing from traditions linguistic approaches based on written text. New tools are now becoming available to execute the main tasks involved in spontaneous speech studies: data speech recording, transcription, alignment and annotation. None of these tasks are trivial and require a sound expertise, still they are essential for the future of linguistic studies. Current research in the domain of syntax-intonation interaction already revealed unexpected results for supposedly well-known prosodic items, such that sentence modality, congruence with syntax, stress clash, left and right dislocation, parenthesis, etc. These results could not have been discovered without careful analysis of actual speech data, as the traditional available linguistic models, particularly in syntax, were, and still are, highly conditioned by the analysis of written text, a very specialized and limited mode of linguistic communication indeed.