Reasoning with Transformer-based Models: Deep Learning, but Shallow Reasoning

Chadi Helwe (1) , Chloé Clavel (2, 3, 4) , Fabian Suchanek (2, 5, 6)

1 Télécom Paris
2 IP Paris - Institut Polytechnique de Paris
3 IDS - Département Images, Données, Signal
4 S2A - Signal, Statistique et Apprentissage
5 INFRES - Département Informatique et Réseaux
6 DIG - Data, Intelligence and Graphs

Recent years have seen impressive performance of transformer-based models on different natural language processing tasks. However, it is not clear to what degree the transformers can reason on natural language. To shed light on this question, this survey paper discusses the performance of transformers on different reasoning tasks, including mathematical reasoning, commonsense reasoning, and logical reasoning. We point out successes and limitations, of both empirical and theoretical nature.

Domains: Databases [cs.DB]; Web

Chadi Helwe, Chloé Clavel, Fabian Suchanek. Reasoning with Transformer-based Models: Deep Learning, but Shallow Reasoning. International Conference on Automated Knowledge Base Construction (AKBC), 2021, online, United States. ⟨hal-03344668⟩