Mini-Seminar: Being Pragmatic about Natural Language Processing =============================================================== 13:30 Parsing the `Map Task Corpus' Dialogues Afzal Ballim EPFL The Map Task Corpus is a collection of task-driven dialogues between pairs of people. In each dialogue, one person (the giver) has a map with a route marked on it, and the other person (the follower) has a similar map but without the route. Without seeing each other's maps, the interlocutors must attempt to recreate the route on the follower's map. The task is complicated by various differences between the maps. This talk is concerned with the practical problems involved in parsing dialogues such those found in the Map Task Corpus. It focuses on the parsing system LHIP, which we developed for this task, and which uses an extended Definite Clause Grammar (DCG) notation, making it possible to reuse DCGs that have been developed for more strict texts. ============================================================ 14:15 Analysing coherence of intention in natural language dialogue Paul Mc Kevitt Department of Computer Science University of Sheffield Much of the work on the computer processing of natural languages, or (ITALICS: natural language processing), has concentrated on studying the structure, meaning, and usage of individual utterances. One of the problems in natural language processing is to build theories, models and implementations of how individual utterances cling together into a coherent discourse. We have developed a theory of intention analysis for solving, in part, the problem of natural language dialogue processing. A central principle of the theory is that coherence of natural language dialogue can be modelled by analysing sequences of peoples intention. The theory is incorporated within Operating System CONsultant (OSCON) We demonstrate that the theory and computational model can be applied to the problem of modelling the level of expertise of a user. The results have implications for both the theory and engineering of natural language interfaces. ============================================================ 15:00-15:30 break ============================================================ 15:30 Belief-Based Views of Ill-Formed Input and Non-Literal Language Dan Fass Simon-Fraser University Ill-formed input is natural language that people really use, containing spelling errors, omissions, agreement failure, ungrammaticality, and so on. Three types of non-literal language -- metonymy, metaphor and anomaly -- are also types of ill-formed input. A belief-based view of language is developed that attempts to show how ill-formed input is related to non-literal and literal language. The view bridges the literal/non-literal language divide in the sense that it will show how non-literal language generated by one machine or person could be interpreted by another machine or person as literal, and vice versa. The belief-based view also throws light on the nature of ill-formed input and problems with treating it. ============================================================ 16:15 Integration of natural language and vision processing Paul Mc Kevitt Department of Computer Science University of Sheffield ENGLAND, EU Although there has been much progress in developing theories, models and systems in the areas of Natural Language Processing (NLP) and Vision Processing (VP) there has been little progress on integrating these two subareas of Artificial Intelligence (AI). It is not clear why there has not already been much activity in integrating NLP and VP. Is it because of the long-time reductionist trend in science up until the recent emphasis on chaos theory, non-linear systems, and emergent behaviour? Or, is it because the people who have tended to work on NLP tend to be in other Departments, or of a different ilk, to those who have worked on VP? NLP and VP can be integrated into a single model and computer system architecture. An application domain is that of reconstructing complex blood vessel structures called vasculatures. The proposed architecture incorporates a single knowledge representation for representing knowledge about blood vessel structures. The input to the system will be in the form of X-ray angiograms and clinical reports associated with those angiograms. We have developed a theory of intention analysis for solving, in part, the problem of natural language dialogue processing. A central principle of the theory is that coherence of natural language dialogue can be modelled by analysing sequences of peoples intention. The theory is incorporated within Operating System CONsultant (OSCON). We believe intention analysis to be a fundamental part of language and vision integration. The integration of vision and language processing has implications for artificial intelligence, psychology, philosophy, and computer science. ====================================================================== ====================================================================== Paul Mc Kevitt is 31 and from Dun Na nG\'{a}ll (Donegal), Ireland on the Northwest of the EU. He is a British EPSRC (Engineering and Physical Sciences Research Council) Advanced Fellow in the Department of Computer Science at the University of Sheffield in Sheffield, England. The Fellowship, commenced in 1994, and releases him from his tenured Lecturership (Associate Professorship) for 5 years to conduct full-time research on the integration of natural language, speech and vision processing. He is currently pursuing a Master's Degree in Education at the University of Sheffield. He completed his Ph.D. in Computer Science at the University of Exeter, England in 1991. His Master's Degree in Computer Science was obtained from New Mexico State University, New Mexico, USA in 1988 and his Bachelor's Degree in Computer Science from University College Dublin, Dublin, Ireland, EU in 1985. His primary research interests are in Natural Language Processing including the processing of pragmatics, beliefs and intentions in dialogue. He is also interested in Philosophy, Multimedia and the general area of Artificial Intelligence. ------------------------------------------------------------ Dan Fass has been a post doctoral fellow and then research scientist at the Natural Language Laboratory, Simon Fraser University, Canada since 1988; research specialist at the Computing Research Laboratory, New Mexico State University, USA, 1985-1988; and research officer at the Department of Computer Science, University of Essex, Great Britain, in 1984. He obtained a PhD in Computer Science from the University of Essex in 1988. Fass's research interests include the interpretation of metaphor and metonymy, treatment of ill-formed input, lexical ambiguity resolution, constituent attachment, machine-readable dictionaries, natural language interfaces to databases, machine translation, user modelling, and "knowledge-based" or "conceptual" semantics, i.e., knowledge representations based on semantic networks or frames, sometimes containing an explicit set of semantic primitives for actions, nominals, cases, etc. A book of his, "Processing Metonymy and Metaphor", is in press. ------------------------------------------------------------------------- Afzal Ballim recently joined LITH at the EPFL, having previously worked as a researcher for ISSCO (University of Geneva) from 1988 until 1995, and before that at the Computing Research Lab. of New Mexico State University from 1986 until 1988. He obtained his doctorate in computer science from the University of Geneva in 1992. His research interests: natural language processing including robust parsing, discourse and discourse processing, machine translation and computer assisted translation, user modelling, computer assisted collaborative work, nonmonotonic reasoning, and knowledge representation. He has written a book, with Yorick Wilks, entitled "Artificial Believers", on the subject of user modelling in dialogue. A new book of his, "Discourse through the ViewFinder" will be published in 1996 by Ablex. ---------------------------------------------------------------------------