The elicitation of the ontology -- i.e. the objects of a domain -- is a key issue of conceptual modelling and therefore of knowledge acquisition. The Conceptual Graph Theory provides a knowledge representation formalism to be used in knowledgebased systems with an explicit "type lattice" to account for the ontology. Since knowledge is in most AI applications non formal, it has to be normalized to ensure that the formal exploitation of its representation conforms to its meaning in the domain. Noting the intensional nature of types, which reflect the essences of the objects they denote, this normalization relies on a commitment on type definitions by necessary and sufficient conditions at the knowledge level. Our claim is that the taxonomic structure that accounts for the intensional nature of the ontology can be nothing but a tree, precluding tangled taxonomies. From this starting point, we derive methodological principles to constrain the acquisition of the "type tree", thus...

INTRODUCTION Medical vocabularies are a fundamental resource for medical information processing. 1 They are faced with a difficult problem of coverage, 2 both in width, with different disciplines and new terms, and in depth, to produce more precise descriptions with modifiers and context. A promising way to extend vocabulary coverage is to examine medical corpora, such as patient narratives, with the help of robust natural language processing tools. 3 This can help propose new terms or modifiers for inclusion in existing vocabularies. An issue then arises of categorizing these new items. We aim to assess the relevance of advanced corpus linguistic tools to identify and structure semantic categories. zellig 4 is such a tool. It has been designed to automate the discovery of semantic classes in the spirit of Harris' work. 5 Harris claims it is possible, with a distributional analys

this paper, we examine some methodological and theoretical principles to enforce this conformity. These principles result from our experience in Menelas, a medical language understanding project.

ion The above experiment has shown the different kinds of knowledge that a concept model like GALEN can provide for NLP needs. However, the main challenge is to stress the distinction between information as it is formulated in medical texts and as it is expressed in concept models. This entails mediation between the large expressiveness, permissibility, and impliciteness of natural language on the one hand, and the generality, granularity, and conciseness of the concept model on the other hand. Such a gap between the "language of the texts" and "the language of concepts" can be filled in by considering what linguistic information must be attached to the conceptual level in order to manage the analysis of medical texts. Such syntactic attachments have been defined at different strategic points in the RECIT system. First, it is important to translate the model typology in the context of the analyzed texts. This is performed through the typology annotation which allows concepts to be an...

Medical language processing depends on large-coverage, fine-grained specialized lexicons. The vast majority of existing electronic lexicons concern the English language, and for other languages such as French, resources are scarce. In contrast, large medical thesauri exist in numerouslanguages, including French. Our goal was to study what kind of linguistic information could be extracted from thesauri into a lexicon, in which places human intervention is necessary, and what kind of issues arise in this process. We designed in this purpose a method to build a semantic lexicon from a subset of the Snomed axes in their French translation. Keywords: Natural language processing, SNOMED, semantic lexicon, nomenclature, French. Introduction Medical information processing crucially depends on the availability of large-coverage, fine-grained controlled vocabularies, nomenclatures or classifications. On the computer side, these terminologies specify the elements that constitute the basic sym...

This paper reports on this evaluation method and its application to the French MENELAS prototype (an evaluation of the Dutch MENELAS prototype is described by Spyns;

words, our hypothesis is that given a (supposedly) unknown word, its semantic category can be determined as the most salient among that of its neighbors. RESULTS We attempted to quantify the extent to which this process succeeds in proposing a correct category for a given word of the corpus while we vary several parameters of the method: the similarity measure, thresholds used to prune the graph, and the vote aggregation methods for ranking the categories of the immediate neighbors. With the currently examined parameters, the percentage of correctly categorized words (precision) ranges between 50 and 75%, while the best percentage of categorized words (recall) is 37% for the whole categorization process. More information on the precise experiments and their results is provided in Habert et al. 4 . CONCLUSIONS Whereas weighting was useful, using a threshold was not really desirable, and the different similarity measures tested did no

Although medical language processing (MLP) has achieved some success, the actual use and dissemination of data extracted from free text by MLP systems is still very limited. We claim that the adoption of an "enricheddocument " paradigm (or "document-centered" view) can help to address this issue. We present this paradigm and explain how it can be implemented, then discuss its expected benefits both for end-users and MLP researchers. 1 Introduction Some medical language processing (MLP) systems have achieved a reasonable level of performance and success from a technical point a view, the most well-known one being unarguably the Linguistic String Project Medical Language Processor (LSP/MLP, [1]). However, the actual use and dissemination of data extracted from free text by MLP systems is still very limited. In our opinion, the most crucial factor which impedes a wider use of these results by health care professionals involves the expected quality of the resulting data: one obviously can...

Medical language processing has focused until recently on a few types of textual documents. However, a much larger variety of document types are used in different settings. It has been showed that Natural Language Processing (NLP) tools can exhibit very different behavior on different types of texts. Without better informed knowledge about the differential performance of NLP tools on a variety of medical text types, it will be difficult to control the extension of their application to different medical documents. We endeavored to provide a basis for such informed assessment: the construction of a large corpus of medical text samples. We propose a framework for designing such a corpus: a set of descriptive dimensions and a standardized encoding of both meta-information (implementing these dimensions) and content. We present a proof of concept demonstration by encoding an initial corpus of text samples according to these principles.