"LazImpa": Lazy and Impatient neural agents learn to communicate efficiently - Inria - Institut national de recherche en sciences et technologies du numérique
Communication Dans Un Congrès Année : 2020

"LazImpa": Lazy and Impatient neural agents learn to communicate efficiently

Résumé

Previous work has shown that artificial neural agents naturally develop surprisingly non-efficient codes. This is illustrated by the fact that in a referential game involving a speaker and a listener neural networks optimizing accurate transmission over a discrete channel, the emergent messages fail to achieve an optimal length. Furthermore, frequent messages tend to be longer than infrequent ones, a pattern contrary to the Zipf Law of Abbreviation (ZLA) observed in all natural languages. Here, we show that near-optimal and ZLA-compatible messages can emerge, but only if both the speaker and the listener are modified. We hence introduce a new communication system, "LazImpa", where the speaker is made increasingly lazy, i.e. avoids long messages, and the listener impatient, i.e.,~seeks to guess the intended content as soon as possible.
Fichier principal
Vignette du fichier
2010.01878.pdf (2.31 Mo) Télécharger le fichier
Origine Fichiers produits par l'(les) auteur(s)

Dates et versions

hal-03070404 , version 1 (15-12-2020)

Identifiants

Citer

Mathieu Rita, Rahma Chaabouni, Emmanuel Dupoux. "LazImpa": Lazy and Impatient neural agents learn to communicate efficiently. CONLL 2020 - The SIGNLL Conference on Computational Natural Language Learning, Nov 2020, Virtual, France. ⟨hal-03070404⟩
45 Consultations
36 Téléchargements

Altmetric

Partager

More