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

Download:

Current browse context:

cs.CL

Change to browse by:

cs

References & Citations

DBLP - CS Bibliography

Bookmark

(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: The Effectiveness of Masked Language Modeling and Adapters for Factual Knowledge Injection

Authors: Sondre Wold
Abstract: This paper studies the problem of injecting factual knowledge into large pre-trained language models. We train adapter modules on parts of the ConceptNet knowledge graph using the masked language modeling objective and evaluate the success of the method by a series of probing experiments on the LAMA probe. Mean P@K curves for different configurations indicate that the technique is effective, increasing the performance on subsets of the LAMA probe for large values of k by adding as little as 2.1% additional parameters to the original models.
Comments: Camera ready version for the 16th TextGraphs workshop, located at Coling 2022
Subjects: Computation and Language (cs.CL)
Cite as: arXiv:2210.00907 [cs.CL]
  (or arXiv:2210.00907v1 [cs.CL] for this version)

Submission history

From: Sondre Wold [view email]
[v1] Mon, 3 Oct 2022 13:08:09 GMT (6350kb,D)

Link back to: arXiv, form interface, contact.