ITRI Seminars - Autumn 2001


		ITRI seminars - Autumn 2001 ITRI seminars generally take place 12 noon on Thursdays in room W107 on the first floor of the Watts Building, University of Brighton (Moulsecoomb site). Occasional deviations from this pattern are indicated below. Information on how to find W107 is available on our contact page.
11 October abstract		Alexandre Direne Departamento de Informática, Universidade Federal do Paraná Authoring Intelligent Tutoring Systems for teaching visual concepts
18 October abstract		Fabrice Retkowsky Runtime Collective (Web Engineers), Brighton 18 months of Runtime: content management, new technologies, etc.
8 November abstract		Norman Creaney School of Information and Software Engineering, Faculty of Informatics, University of Ulster at Coleraine Generating Quantifiers
15 November abstract		Mark Levene School of Computer Science and Information Systems, Birkbeck College, University of London WWW Search and Navigation
29 November abstract		Amy Neale Information Technology Research Institute, University of Brighton Using a Systemic Functional Grammar to extend classifications for the most frequent verb senses in English
6 December abstract		Darren Pearce COGS, University of Sussex A Comparative Evaluation of Collocation Extraction Techniques

		Previous ITRI seminars Next Term ITRI seminars
		See also NLP seminars at COGS, University of Sussex

Abstracts

Alexandre Direne
Authoring Intelligent Tutoring Systems for teaching visual concepts

The talk addresses how high-level knowledge about visual images (e.g., X-rays and CT-scans) can be authored and further interpreted within the context of an Intelligent Tutoring System (ITS). Firstly, I shall explain some of the main issues related to learning and training of complex visual concepts, such as the ones of medical diagnostic imaging. I shall argue that, in highly visual domains, a learner's knowledge and expertise can be best developed through a computer tutor if he or she has access to the same surface knowledge representations created by experts, also using computer-based tools. Secondly, the main ideas of the RUI project will be presented. The approach adopted in the project consists of an expert authoring language for managing the complexity of ITS design representations and a domain-independent pedagogic model of dialogue interpretation for carrying out system-active interactions. The language and the model are both supported by implemented software tools. To evaluate the power of such tools, empirical observations have been conducted, which concentrate on the generality of the authoring language. I shall draw conclusions about the suitability of authoring tools in the context of ITS production, specifically aimed at visual diagnosis. Finally, future work in the RUI project will be outlined, focusing on interface as well as learner models.

Fabrice Retkowsky
18 months of Runtime: content management, new technologies, etc.

Runtime Collective, a Web services company, opened for business in April 2000. These 18 months have been pretty eventful, with quite many clients and contracts, and a dot.com bubble bursting. Noticeable amongst all was an evolution in Web development, away from custom and closed systems towards open source and free standards. To an extent this was the case in the area of content management systems, one of our centre of expertise (see www.infoconomy.com): off go the old-fashioned database-backed website written in a scripting language (with all its intricacies and surprises), and here comes the Java-supported standards that are XML, XSL et al., now ripe for wide-spread use. This talk will give an overview of content management and web publishing, including some technical aspects, and will also present the areas and research opportunities that a service company like Runtime can get involved in.

Norman Creaney
Generating Quantifiers

The talk describes a domain independent algorithm for generating descriptions involving quantification. Two significant, and related, problems are identified and given particular attention: choosing between the many alternative descriptions that are possible; and, controlling the search performed in the processing. An "aggregation tree" is used as the central data structure; and the "jump heuristic" is used to address the two problems mentioned above.

Mark Levene
WWW Search and Navigation

The navigation problem is a long standing problem in hypertext of users "getting lost in hyperspace". Although search engines are good at getting users to the home page of a Web site finding the information therein involve users "surfing" which is often frustrating and problematic. We will demonstrate a new tool called NavZone (www.navigationzone.com) which semi-automates user navigation by proposing to users a set of preferred trails that are relevant to their query. This is in contrast to a site search engine which merely outputs a list of pages which are relevant to the user query without addressing the problem of which trail the user should follow.

Amy Neale
Using a Systemic Functional Grammar to extend classifications for the most frequent verb senses in English

In this paper I will explore the possibilities of using a Systemic Functional Grammar (SFG) for producing a semantic classification of verb senses - or, in SFG terms, of 'experiential meaning' - which distinguishes them by very fine semantic differences, making the choices between senses as 'delicate' as possible (Halliday, 1961).

My paper will focus on three main areas:

Compiling a list of most frequently occurring verb senses in English.
The semantic classification of these senses for modeling in a 'system network'.
Obtaining 'probabilities' for the likelihood of occurrence of these senses.

The backdrop for my work is the Cardiff Grammar (Fawcett, forthcoming), a large computer-based grammar of English, which is used by GENESYS (the sentence generation component of the COMMUNAL Project).

Fawcett, R.P. (forthcoming) Functional Semantics Handbook: Analyzing English at the Level of Meaning. London: Continuum.

Halliday, M.A.K. (1961) 'Categories of the Theory of Grammar'. WORD 17.3, pp. 241- 292.

Darren Pearce
A Comparative Evaluation of Collocation Extraction Techniques

In this talk, I present a common framework for evaluation of collocation extraction techniques and discuss the results for a number of existing methods. I compare these to the performance of new techniques based on tests for lexical substitutability.

Over the last thirty years, there has been much discussion in the linguistics literature on the exact nature of collocation. Even now, there is no widely-accepted definition. One by-product of this lack of agreement is the lack of a consistent evaluation methodology. Some employ the skills of a professional lexicographer, others use native speaker judgements. Computational resources such as term banks and machine-readable dictionaries have also been used. This is the approach taken in this work.

I begin by briefly describing the various computationally-tractable definitions of collocation that have been investigated empirically as well as details of the new techniques I have developed. I then compare implementations of each of these approaches and investigate the interaction of precision and recall.

Finally, I discuss the relative merits of each of the existing techniques as well as compare the performance of these new approaches measured within this common framework. The discussion also places a particular emphasis on the efficacy of parse (or dependency) information in collocation extraction. In light of the experimental results, I discuss future directions for improving these new techniques.

Maintained by Paul Piwek (Paul.Piwek@itri.brighton.ac.uk ).
Last updated Tuesday September 18 2001

©Information Technology Research Inst itute