Fig 3 - uploaded by Mathieu Lafourcade
Content may be subject to copyright.
(1) Linking of each meaning equivalent to its hyperonym. If an hyperonym is by itself ambiguous, its proper meaning is selected by minimizing the thematic distance between vectors. (2) Trimming of redundant meanings. When several meaning equivalents are competing, the one linked to the most specific hyperonym is selected. Other meanings are deleted as they related to upper items in the partial hierarchy.
Source publication
The NLP team of LIRMM currently works on lexical dis- ambiguation and thematic text analysis (Lafourcade, 2001). We built a system, with automated learning capabilities, based on conceptual vec- tors for meaning representation. Vectors are supposed to encode ideas associated to words or expressions. In the framework of knowledge and lexical meaning...
Context in source publication
Context 1
... selected hyperonym candidate is the term which vector is the closest (in thematic distance terms) to V (Hyper-cand/meaning i )). We are now able to create a node for each Hyper-cand/meaning i with a link to the corresponding disambiguated hyperonym candidate (cf Fig. 3 stage 1). If the term doesn't exist in the lexical database, it is added along its computed vector. (1) Linking of each meaning equivalent to its hyperonym. If an hyperonym is by itself ambiguous, its proper meaning is selected by minimizing the thematic dis- tance between vectors. (2) Trimming of redundant meanings. When several meaning ...
Citations
... OPALES provides several tools for indexing and retrieving data, including queries based on descriptors, on keywords, on text similarity, or conceptual graphs and so on. Let us suppose we also want to support the Conceptual Vectors [13] querying technique. The data type and its associated set of tools are first implemented in the small in Java, say, in an IUHM [21] compliant manner. ...
Traditional techniques for Programming in the Large, especially Object-Oriented approaches, have been used for a considerable time and with great success in the implementation of service-based information systems. However, the systems for which these techniques have been used are static, in that the user-services and the data available to users are fixed by the system, with a strict separation between system and user. Our interest lies in currently emerging dynamic systems, where both the data and the services available to users are freely extensible by the users and the strict distinction between system and user no longer exists. We describe why traditional object-oriented approaches are not suitable for modelling such dynamic systems. We discuss a new architectural model, the Information Unit Hypermedia Model, IUHM, which we have designed for modelling and implementing such dynamic systems. IUHM is based upon the application of structural computing to a hypermedia structure, which thereby operates as a service-based architecture. We discuss the details of this model, and illustrate its features by describing some aspects of a large-scale system which was built by using this architecture.
Hierarchy building from dictionaries and free texts is often viewed as an application of NLP for domain modeling. The reversal (i.e. building and using such hierarchy for Word Sense Disambiguation) is also definitively useful in NLP. Indeed, we do observe that, even in very specialized texts, polysemous terms as well as blurring linguistic phenom-ena like metonymy or metaphor are frequent. Conceptual vectors are part of a model for meaning representation applicable to lexical disambigua-tion [Lafourcade, 2001]. We devise some strategies combining vectors and relation templates to automatically construct lexical network able to dis-criminate between various is-a and part-of relations.
In the framework of the Papillon project, there are accep-tions that are not lexicalized in a given language. They correspond at best to some hypernyms. Moreover, in order to easily supervise a transla-tion process, we would like to be able to name meanings instead of refer-ring to definitions by numbers. Dictionaries define words using a "genus + differentia" approach and can be exploited for new compound extrac-tion. This approach is relates for various research efforts in sense naming. The conceptual vector model (CVM) aim to represent meanings in a non lexical way and vectors are calculated through the analyses of multiple dictionary resources. The following article describes how our work on sense naming can be easily applied to the Papillon project and could offer simultaneously a lexical augmentation approach, a disambiguation checking process and a new lexical resource. Using this information, an automatic process helps building a mixed lexical and acception network.
In the framework of the Papillon project, there are acceptions that are not lexicalized in a given language. They correspond at best to some hypernyms. Moreover, in order to easily supervise a translation process, we would like to be able to name meanings instead of referring to definitions by numbers. Dictionaries define words using a "genus differentia" approach and can be exploited for new compound extraction.
