We gratefully acknowledge support from
the Simons Foundation and member institutions.
Full-text links:


Current browse context:


Change to browse by:

References & Citations

DBLP - CS Bibliography


(what is this?)
CiteULike logo BibSonomy logo Mendeley logo del.icio.us logo Digg logo Reddit logo ScienceWISE logo

Computer Science > Computation and Language

Title: Generated Knowledge Prompting for Commonsense Reasoning

Abstract: Despite their ability to capture large amount of knowledge during pretraining, large-scale language models often benefit from incorporating external knowledge bases, especially on commonsense reasoning tasks. This motivates us to explore how we can best leverage knowledge elicited from language models themselves. We propose generating knowledge statements directly from a language model with a generic prompt format, then selecting the knowledge which maximizes prediction probability. Despite its simplicity, this approach improves performance of both off-the-shelf and finetuned language models on four commonsense reasoning tasks, improving the state-of-the-art on numerical commonsense (NumerSense), general commonsense (CommonsenseQA 2.0), and scientific commonsense (QASC) benchmarks. Notably, we find that a model's predictions can improve when using its own generated knowledge, demonstrating the importance of symbolic knowledge representation in neural reasoning processes.
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2110.08387 [cs.CL]
  (or arXiv:2110.08387v1 [cs.CL] for this version)

Submission history

From: Jiacheng Liu [view email]
[v1] Fri, 15 Oct 2021 21:58:03 GMT (10608kb,D)

Link back to: arXiv, form interface, contact.