Lyne Da Sylva

Finding out: a system for providing rapid and reliable answers to questions in the construction sector

The construction sector is notorious for the dichotomy between its intensive use of information in its decision‐making processes and its limited access to, and insufficient use of, the pertinent information that is potentially available, e.g. on the internet. This paper seeks to examine this issue. To solve this problem (the ‘problem of information aboutinformation’), a multidisciplinary team developed an online question‐answering (Q.‐A.)system that uses natural language for the query and the reply. The system provides a direct answer to questions posed by building industry participants, instead of providing a list of references (as is the case with most online information retrieval systems), much as if onewere asking a question of, and receiving a response from, an expert.It has the capabilitiesto process questions in natural language, to find appropriate fragments of answers indifferent web sites and to condense them into a paragraph, also written in natural language. The main features of the system are that it uses domain‐specific knowledge (in the form ofa hierarchical specialized thesaurus complemented by terms of fieldwork parlance),semantic categorization, a database of filtered and indexed web sites, and an online interface that is adapted to different profiles of actors in the construction sector. The testing process shows that the system goes beyond the lists of references and links provided by traditional search engines on the web.The Q.‐A.system already gives 70% of satisfactory answers. The Q.‐A.system can be applied to other business domains apart from information retrieval and decision‐making in the building sector. It is also possible to apply it to the exploitation of in‐house knowledge management database.

Text compression by syntactic pruning

We present a method for text compression, which relies on pruning of a syntactic tree. The syntactic pruning applies to a complete analysis of sentences, performed by a French dependency grammar. Sub-trees in the syntactic analysis are pruned when they are labelled with targeted relations. Evaluation is performed on a corpus of sentences which have been manually compressed. The reduction ratio of extracted sentences averages around 70%, while retaining grammaticality or readability in a proportion of over 74%. Given these results on a limited set of syntactic relations, this shows promise for any application which requires compression of texts, including text summarization.

A document browsing tool: using lexical classes to convey information

This research project is a contribution to the global field of information discovery in digital documents We aim to provide the user with a tool for flexible access to the contents of digital documents: a text browsing facility inspired by traditional “back-of-the-book” style indexes It gives at a glance the main topics discussed in the document, and presents certain kinds of relationships between these topics These are captured automatically by exploiting certain lexical classes Previous research on this and similar topics is reviewed, followed by the main characteristics of a research prototype, which relies on modeling of professionally produced indexes Experimental results are presented, as well as remaining hurdles and potential applications.

Using Ancillary Text to Index Web-based Multimedia Objects

PeriCulture is the name of a research project at the Universite de Montreal which is part of a larger project based at the Universite de Sherbrooke. The parent project aimed to form a research network for managing Canadian digital cultural content. The general research objective of PeriCulture was to study indexing methods for web-based non-textual cultural content, specifically still images. The research results reported here build on work in image indexing and automatic (text) indexing by studying properties of text associated with images in a networked environment to try to gain some understanding of how the ancillary text associated with images on web pages can be exploited to index the corresponding images. We studied this question in the context of selected web sites, i.e. that contained multimedia objects, that had text associated with these objects (broader than file names and captions), that were bilingual (English and French), and that housed Canadian digital cultural content. We identified keywords that were useful in indexing and studied their proximity to the object described. Potential indexing terms were identified in various HTML tags and full text (each considered a different source of ancillary text). Our study found that a large number of useful indexing terms are available in the ancillary text of many web sites with cultural content, and that ancillary text of different sources have variable usefulness in retrieval. Our results suggest that these terms can be manipulated in a number of ways in automated retrieval systems to improve search results.

The implementation of a computational grammar of French using the grammar development environment

The design and implementation of a large-coverage computational grammar of French is described. This grammar is compared to a comprehensive computational grammar of English which was implemented using the same computer workbench. Although many similarities may be observed in the two grammars, there are important structural differences which can be traced back to features specific to the French language, notably agreement and eliticization.