Speeding Up Syntactic Learning Using Contextual Information

Leonor Becerra-Bonache (1) , Elisa Fromont (1) , Amaury Habrard (1) , Michael Perrot (1) , Marc Sebban (1)

1 LHC - Laboratoire Hubert Curien

It has been shown in (Angluin and Becerra-Bonache, 2010, 2011) that interactions between a learner and a teacher can help language learning. In this paper, we make use of additional contextual information in a pairwise-based generative approach aiming at learning (situation,sentence)-pair-hidden markov models. We show that this allows a significant speed-up of the convergence of the syntactic learning. We apply our model on a toy natural language task in Spanish dealing with geometric objects.

Domaines: Apprentissage [cs.LG]

Leonor Becerra-Bonache, Elisa Fromont, Amaury Habrard, Michael Perrot, Marc Sebban. Speeding Up Syntactic Learning Using Contextual Information. International Conference on Grammatical Inference, Sep 2012, United States. pp.49-53. ⟨hal-00717809⟩