Seminaire: Alpage, Joakim Nivre, Training Deterministic Parsers Using Non-Deterministic Oracles, 27 juin, 10h30

Thierry Hamon thierry.hamon at UNIV-PARIS13.FR
Sun Jun 23 10:16:15 UTC 2013


Date: Thu, 20 Jun 2013 10:04:52 +0200
From: Marie Candito <marie.candito at gmail.com>
Message-ID: <CAKCM-9FORzbxSO_a7JkqHrO5yqUp6OfrFB-o1kLwVQasVYxJFQ at mail.gmail.com>

************** Séminaire de l'équipe Alpage *********

Il s'agit du séminaire de recherche en linguistique informatique
organisé par l'équipe Alpage, équipe mixte INRIA - Paris Diderot,
spécialisée en analyse syntaxique automatique et en traitement du
discours.

*Le prochain séminaire se tiendra exceptionnellement :

JEUDI 27 juin

de 10h30 à 12h (une demi-heure plus tôt qu'habituellement),

en salle 265*

Bâtiment Olympe de Gouges
rue Albert Einstein
75013 Paris
(dernier bâtiment de la rue Albert Einstein, numéro non attribué...)

Toute personne intéressée est la bienvenue.

**********************************************************

*Joakim Nivre, Uppsala University*
Joint work with Yoav Goldberg, Bar-Ilan University.

*Title *: Training Deterministic Parsers Using Non-Deterministic Oracles

*Abstract* :
Greedy transition-based dependency parsers are among the fastest
available systems for natural language parsing but tend to suffer from
search errors and subsequent error propagation.  This problem is
aggravated by the fact that they are normally trained using oracles that
are deterministic and incomplete in the sense that they assume a unique
canonical derivation and are only valid as long as the parser does not
stray from this path. In this talk, I will introduce oracles that are
nondeterministic and complete in the sense that they allow different
derivations of the same parse tree and make optimal predictions for all
parser states, including states from which the gold parse is no longer
reachable.  Experimental evaluation on a wide range of data sets shows
that using these oracles to train greedy parsers gives substantial
improvements in accuracy because of reduced error propagation.
Moreover, this improvement comes at no cost in terms of efficiency,
unlike other techniques like beam search.

**********************************************************


-------------------------------------------------------------------------
Message diffuse par la liste Langage Naturel <LN at cines.fr>
Informations, abonnement : http://www.atala.org/article.php3?id_article=48
English version       : 
Archives                 : http://listserv.linguistlist.org/archives/ln.html
                                http://liste.cines.fr/info/ln

La liste LN est parrainee par l'ATALA (Association pour le Traitement
Automatique des Langues)
Information et adhesion  : http://www.atala.org/
-------------------------------------------------------------------------



More information about the Ln mailing list