Section: Software

JSafran

Contact : Christophe Cerisara (Christophe.Cerisara@loria.fr).

J-Safran is the “Java Syntaxico-semantic French Analyser”. Its development has started in june 2009 from the collaboration between Parole and Talaris in the context of the RAPSODIS project. It is an open-source dependency parsing platform that is dedicated to oral speech. Its main interesting features, as compared to other similar software, are:

• It is designed for both manual and semi-automatic edition of dependency graphs, as well as for fully automatic parsing. To this end, it integrates two of the best state-of-the-art automatic parsers of the litterature, the Malt Parser and the MATE parser, as well as a third experimental Maximum Entropy Markov Model-based parser developed from November 2011 in the team. It further integrates three automatic Part-of-speech taggers: the TreeTagger, the OpenNLP and MATE taggers.

• It is smoothly interfaced with the JTrans platform, thus enabling the user to directly listen to the aligned speech segments when annotating, which is an important added value to help disambiguation. The interface between both software goes well beyond simple method calls, as they both share for instance parts of the tokenization process and access a common immutable text source from the disk or on the Web.

• It supports multi-layer annotations, such as dependency relations, semantic role labeling, named entities and coreference links for instance, as well as inter-layer projection facilites.

• It offers a powerful rule-based search and tree manipulation language to transform for instance the annotation schema of a large corpus with a few commands only.

• As it is written in pure Java, it can run on any modern computer, either as a standalone application or embedded in a web page.

A description of JSafran is published in [16] . JSafran is distributed under the Cecill-C licence, and can be downloaded at http://synalp.loria.fr/?n=Research.Software