Research objectives and content
The identification of grammatical and semantic dependencies between phrasal segment is indispensable for NLP components that have to handle spoken data or to structure partial analises produced by NLP tools. In this research project NLP components will be designed and implemented for the resolution of dependencies and will be applied to solve the ambiguities left unsolved by speech tools at the phonological level. The system will base on analogy-based models of acquisition of linguistic skills by extending its application to syntactic and semantic knowledge. Training content (objective, benefit and expected impact)
ENST is one of the leading european research centres for spoken NLP. The proposer expects to substantially benefit from being exposed to self-learning techniques currently developed at ENST. Self-learning techniques for language modelling are the focus of increasing investigation in speech technologies.
Links with industry / industrial relevance (22)
ENST has links with France Telecom. Applications that allows automatic handling of spoken data are of great interest for the communication industry. The development of components able to exploit the linguistic content of spoken data can be used to compress spoken data and to realize robust speech recognition systems. Self-learning systems for fast and accurate acquisition of syntactic and semantic information can be also fruitfully used for information retrieval/access, automatic summarisation, computer-aided translation.