Deep Syntax for French



Deep Syntax for French: from syntactic trees to semantic graphs using rewriting techniques

Corentin Ribeyre

22nd February 2016 | Room 432 | Battelle | 16h00

Syntactic parsing has been the focus of intensive research over  the last decades, leading to current state-of-the-art parsers that provide high performance. However, extracting more semantically-oriented high-quality information from syntactic parses is not straightforward, due to the numerous syntactic-semantic divergences. “Deep syntax” is generally intended as an intermediate representation between what is actually observable (surface syntax) and a semantic representation, which abstracts away from syntactic variation, such as diathesis alternations or non-canonical word order, and which can thus serve as an easier basis for semantic analysis.  In this talk, I will present the deep annotation scheme we designed for French and talk about the way two corpora (the French Treebank and the Sequoia Treebank) were annotated.  First, I focus on the annotated phenomena and the annotation methodology.  Then, I describe and evaluate the semi automatic pre-annotation of those corpora using a graph-rewriting system called OGRE.



16 févr. 2016