Old site of the INRIA ARC RAPSODIS:
The ARC Rapsodis started in 2008 and lasted for two years. Its main objective has been the study of approaches to make use of syntactic and semantic information within speech recognition systems. Two main tracks were investigated: first, lexical semantics based on the distributional hypothesis have shown to provide relevant information for high-level semantic information; second, deep parsing has proven to be an important building block of many subsequent NLP processes, but it also happened to be very dependent on the domain. In particular, the lack of speech treebanks for French is a serious limitation for its effective use in this context. We thus continued this work from 2009 by addressing this specific task of French speech parsing.
Post-RAPSODIS (>2009) work
Our research objectives started in RAPSODIS continue in other projects. One of our result is the JSafran software for semi-automatic parsing in French, that can be freely downloaded below.
If you are interested in JSafran, or more generally in syntactic parsing, please let us know using the contact link; we'll be happy to hear from you !
JSafran software (free + open-source)
McNemar significance test to compare two taggers or classifiers
Live demo (limited functionalities - applet - requires java 1.6)