Subsections

• Introduction
• Description
• LE Uses

   
Resources from MemoData

  
Introduction

The Memodata resources can be described as a package containing definition dictionary for French, Multilingual dictionaries for French, Italian, Spanish, English and German and an ontology for French. The French side has been more developed but there is on on-going work for resources on other languages as well. All the existing resources can be linked together. For example one can start with the French definition dictionary, link it to the French ontology, next, access the different translations corresponding to the different meanings. As for applications, the dictionary is meant for semantic disambiguation and to be used as a thesaurus, where you start from the idea and get to the words.

  
Description

Dictionnaire intégral

is the core dictionary. It is meant to be used both for dictionary editing purpose and natural language processing. It exists in five languages: French (20.000 classes (links or sort of synonyms)) that give an ontological organization of around 180.000 words-meaning (130.000 words with grammatical categories), English, Italian, Spanish and German (around 35.000 words for each languages roughly corresponding to 45.000 word meanings). There are 350 labels (level, origin, domain, ...) and more than 40.000 concepts that organize the meanings.

Dicologique

It is the editorial version of the dictionary. It refers to the interface as well as to the content.

Lexidiom

It is a tool meant for lexicographers to edit/ change/enrich the dictionary.

Semiographe

It is the compiled version of the dictionary that can be used in applications. For instance a recent version of SEMIOGRAPH is used for disambiguation, genericity, indexing the French yellow pages. SEMIOGRAPH contains information and features that are not in DICOLOGIQUE.

How to integrate the different components

. The multiligual dictionaries are part of LEXIDIOM, attached to the same concepts as the French equivalents and linked via a relation of the type ``to be translated by" in that particular sense. There are two types of information in this dictionary.

• the conceptual network which associates words or expressions with sets of referents.
• the linguistic level (for choosing the translation for instance)

   
LE Uses

Semantic disambiguation. The Semiographe uses the context of words to choose between the different meanings of a word. For instance it can choose between the different meanings of the word ``renard" :
 

A- Voil`! Jean a (obturi) le renard (du barrage)
B- Regarde Justine qui (porte) (a revjtu) son magnifique renard
C- le (corbeau) et le renard
D- la France fait la (chasse) au renard pour lutter contre le diveloppement de la (rage).
 

Where parentheses indicate :

• words that are used to perform the disambiguation task
• for each sentence containing two parentheses that one word is optional

The corresponding meanings are given as follows:

• A: fente accidentelle par oy s'ichappe l'eau d'une installation (barrage, coque de navire, icluse....) (This meaning is not the default one)
• B: pour fourrure de renard (not default, it is an ellispis)
• The C example has just one word in between parenthesis. This word is optional as well, because, at the end, ``renard" has the default meaning.
• In the example D it is the animal again (fox) but with the feature ``gibier".

One can also uses the dictionary the other way around, starting from an hyponym and find its specific value. For instance one can ask couper un arbre (to cut a tree) and get back the answer abattre, or one can ask personne qui vend des fleurs (person who sells flowers) and get back fleuriste (florist). Another way of querying the system is to ask questions such as je veux louer un vihicule volant (I want to rent a flying machine), and get location d'avion (plane renting) as an answer.