Séminaire de Recherche en Linguistique

This talks provides a look at the capabilities of deep linguistic
natural language processing in the context of ParGram, an
international effort at building grammars for languages as diverse as
English, French, Welsh, Wolof, Georgian, Urdu, Japanese, Indonesian,
Arabic and Murrinh-Patha.  The aim is to build these grammars using
not only the same theoretical assumptions (based on Lexical-Functional
Grammar) but also the same computational methods.  The underlying
development platform XLE/XFR (used by Powerset and in the ParGram
effort) was developed at PARC over a number of years.  The talk will
present the various powerful analysis possibilities this system allows
for, including a sophisticated treebanking infrastruture, a version of
Optimality Theory and an integration of statistical information to
constrain and inform both parsing and generation.  The talk will also
discuss the assumptions about the underlying grammar architecture that
are made, e.g., with respect to the morphology-syntax and the
syntax-semantics interface and the fact that the underlying grammar
architecture allows both parsing and generation.