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

Download:

Current browse context:

cs.CL

Change to browse by:

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: Measuring Fairness with Biased Rulers: A Survey on Quantifying Biases in Pretrained Language Models

Abstract: An increasing awareness of biased patterns in natural language processing resources, like BERT, has motivated many metrics to quantify `bias' and `fairness'. But comparing the results of different metrics and the works that evaluate with such metrics remains difficult, if not outright impossible. We survey the existing literature on fairness metrics for pretrained language models and experimentally evaluate compatibility, including both biases in language models as in their downstream tasks. We do this by a mixture of traditional literature survey and correlation analysis, as well as by running empirical evaluations. We find that many metrics are not compatible and highly depend on (i) templates, (ii) attribute and target seeds and (iii) the choice of embeddings. These results indicate that fairness or bias evaluation remains challenging for contextualized language models, if not at least highly subjective. To improve future comparisons and fairness evaluations, we recommend avoiding embedding-based metrics and focusing on fairness evaluations in downstream tasks.
Comments: 15 pages, 4 figures, 3 tables
Subjects: Computation and Language (cs.CL); Computers and Society (cs.CY); Machine Learning (cs.LG)
Cite as: arXiv:2112.07447 [cs.CL]
  (or arXiv:2112.07447v1 [cs.CL] for this version)

Submission history

From: Pieter Delobelle [view email]
[v1] Tue, 14 Dec 2021 15:04:56 GMT (524kb,D)

Link back to: arXiv, form interface, contact.