Section: New Results

lexical knowledge acquisition and visualization

Participants : Éric Villemonte de La Clergerie, Mickael Morardo, Benoît Sagot.

In relation with our collaboration with Lingua & Machina (cf section  4.4 ), Mikael Morardo has enriched the interfaces of the WEB platform Libellex for the visualization and validation of more complex lexical resource. In particular, the focus has been on the development of a graph-based view with the javascript Library d3.js to represent large lexical networks. The current implementation is powerful enough to deal with large networks of several teens of thousands of connections, allowing the visualization of fragments of the network and an easy navigation. Because the graph-view proved to be both intuitive and efficient, the previous list-based view for terminology was partially re-implemented in the new graph-view. It was also extended for visualizing and validating more complex lexical networks, like the French Wordnet WOLF coupled with the original English WordNet (cf  5.9 ).

The graph-based view was used to explore several networks built using Harris' distributional hypothesis (through a clustering algorithm) on the output of FRMG for several corpora. Because terminology was now be visualized at the same time, the clustering algorithm was modified to be able to take into account a list of terms (also automatically extracted from the parsed corpora) .