Compositionality in a simple corpus

Manuel Vargas Guzmán (1) , Maria Boritchev (2) , Jakub Szymanik (3) , Maciej Malicki (2)

1 University of Warsaw, Krakowskie Przedmieście 26/28, 00-927 Warsaw, Poland
2 Mathematical Institute of the Polish Academy of Sciences
3 CIMeC and DISI - University of Trento Rovereto (TN) - Italy

We investigate the capacity of neural networks (NNs) to learn compositional structures by focusing on a well-defined simple logical corpus, and on proof-centered compositionality. We conduct our investigation in a minimal setting by creating a simple logical corpus, where all compositionality-related phenomena come from the structure of proofs as all the sentences of the corpus are propositional logic implications. By training NNs on this corpus we test different aspects of compositionality, through variations of proof lengths and permutations of the constants.

Mots clés: Compositionality; logic; reasoning; neural networks.

Domaines: Informatique et langage [cs.CL]

Manuel Vargas Guzmán, Maria Boritchev, Jakub Szymanik, Maciej Malicki. Compositionality in a simple corpus. Journées Jointes des Groupements de Recherche Linguistique Informatique, Formelle et de Terrain (LIFT) et Traitement Automatique des Langues (TAL), Nov 2022, Marseille, France. pp.55-63. ⟨hal-03846838⟩