DEFI : The English-to-French direction


 

The English-to-French direction of the defi  project is described in six collections of documents.

 

The first provides a general introduction to defi  and a discussion of the workings of the matcher :

 

A. Michiels - The DEFI project: start here

A. Michiels - The DEFI matcher: an introduction (Euralex’98 paper)

A. Michiels - Target Selection and Word Sense Discrimination in DEFI  (useful     background reading but describes an older version of DEFI)

A. Michiels - The treatment of the INDIC field

 

The second deals with the preparation of the lexical resources used in the project :

 

A. Michiels and N. Dufour - From SGML tape to Dic Clauses

N. Dufour - From the Oxford Hachette SGML tape to DEFI dictionaries

N. Dufour - Merging two Defi dictionaries

N. Dufour - LKP: a user's guide

N. Dufour - DEFIDIC, a lexical database for computerized translation selection

Extracts from DEFIDIC under LKP

 

The third is concerned with the input file format and the preprocessing operations :

 

                       

An example of input text file

An example of enhanced engcg parse

 

 

The fourth illustrates the lexical resources in defi format :

 

Extract from the single word lexeme DEFI dictionary (SDIC)

Extract from the MWU DEFI dictionary (DIC)

Extract from the tree of keys accessing the MWU dictionary

Extract from the monolingual DEFI dictionary

 

The fifth shows the results obtained by running the defi matchers on test files :

 

The first defi test suite

The COBUILD-derived test bed

The COBUILD-derived test bed in zip format

Explanation of the debugging information provided by the matcher

 

 

The sixth contains only one document, giving the Prolog code of the defi matcher :

           

Prolog Code of the DEFI matcher

Prolog Code of the DEFI matcher in zip format

 

The following two documents are in French :

 

DEFI, un outil d'aide à la compréhension

Le traitement de la phraséologie dans DEFI

 

Back to the English Department's Home Page