Recurrent Neural Network Sentence Parser for Multiple Languages with Flexible Meaning Representations for Home Scenarios

Xavier Hinaut (1) , Johannes Twiefel (2)

1 Mnemosyne - Mnemonic Synergy
2 KT - Knowledge Technology group [Hamburg]

We present a Recurrent Neural Network (RNN), namely an Echo State Network (ESN), that performs sentence comprehension and can be used for Human-Robot Interaction (HRI). The RNN is trained to map sentence structures to meanings (i.e. predicates). We have previously shown that this ESN is able to generalize to unknown sentence structures in English and French. The flexibility of the predicates it can learn to produce enables one to use the model to explore language acquisition in a developmental approach. This RNN has been encapsulated in a ROS module which enables one to use it in a cognitive robotic architecture. Here, for the first time, we show that it can be trained to learn to parse sentences related to home scenarios with higly flexible predicate representations and variable sentence structures. Moreover we apply it to various languages, including some languages that were never tried with the architecture before, namely German and Spanish. We conclude that the representations are not limited to predicates, other type of representations can be used.

Mots clés: recurrent neural network; echo state networks; human-robot interaction; language acquisition; home robotics; reservoir computing; thematic role assignment; knowledge representation

Domaines: Réseau de neurones [cs.NE]; Apprentissage [cs.LG]; Linguistique; Robotique [cs.RO]; Sciences cognitives

Xavier Hinaut, Johannes Twiefel. Recurrent Neural Network Sentence Parser for Multiple Languages with Flexible Meaning Representations for Home Scenarios. IROS Workshop on Bio-inspired Social Robot Learning in Home Scenarios, Oct 2016, Daejon, South Korea. ⟨hal-01417667⟩