13 May: fyi -- phd funding, France

Index of May 2008 | Index of year: 2008 | Full index


PhD Thesis Opportunity (Autumn 2008)

Contribution of Syntactic Analysis to Machine Translation

Machine Translation has tremendously progressed in tasks such as
lexical transfer (word translation), idiomatic expressions, with the
help of statistical lexical techniques, advanced multilingual
resources and the availability of aligned corpora. Nowadays, the new
challenge is to extend these skills to correct sentences building in
the target language, in order to produce grammatically acceptable
texts. Several methods are competing in this field: Surface structure
generators from a pivot representation as well as numerous grammatical
structures and sentence style learning techniques, using corpora.

The subject of this PhD thesis consists in studying, theoretically as
well as experimentally, the source language syntactic analysis
(through parsing) contribution to the generation of grammatically
correct sentences in the target language. Some of our team
contributions (namely Chauché and Prince 2006, Bonnin and Prince 2007,
available at the LIRMM the on-line documentation HAL ), show that the
syntactic generation of the target language, if the latter is not very
far from the source (pairs such as English-German, French-English,
French-Spanish...), might be obtained through the source language
units syntactic structure transformation.

The goals of this work include but are not restricted to:

Extending, enhancing and refurbishing the existing transformation
rules and models suggested in the cited words, possibly through
learning these transformations from aligned data.

Building a transformation grammar, from the existing French to English
translation protoype SYGFtoE. This grammar must result from a
theoretical and algorithmical reworking of the structures learnt in
step 1.

Evaluating this work with corpora.

Application: Candidates must apply at Pr. Dony's following URL :
http://www.lirmm.fr/~iss/specdoctinfo/InfoCandidatureAlloc.html before
May, 30th.

A good level in NLP and fluency in French are highly recommended.

Contact : prince@lirmm.fr

Index of May 2008 | Index of year: 2008 | Full index