HERALD Hybrid Environment for Robust Analysis of Language Data

This project addresses the problem of performing structural and semantic analysis of data where the syntactic and semantic models of the domain are inadequate, and robust methods must be employed to perform a “best approximation” to a complete analysis. This problem is particularly pertinent in the domain of text analysis. The ability to deal with large amounts of possibly ill-formed or unforeseen text is one of the principal objectives of current research in Natural Language Processing by computer (NLP), an ability which is particularly necessary for advanced information extraction and retrieval from large textual corpora. The results of this work can, however, be applied in other domains where a mix of partial grammatical and semantic models exist, such as in image analysis. The project builds on previous FNSRS projects by the proposers. In particular it integrates discourse analysis methods is a direct continuation of FNSRS project ROTA which addressed the problems of developing robust grammatical analysis on noisy or partially described data. While the proposers have had much success in this latter project on the development of efficient robust techniques for grammar-based structural analysis of data, these techniques must be supplemented by semantic analysis, because many analysis problems cannot be resolved in any other way. This project proposes the investigation of such methods and their integration with structural analysis into a hybrid architecture.