Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
102 tokens/sec
GPT-4o
59 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
50 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Social Biases in NLP Models as Barriers for Persons with Disabilities (2005.00813v1)

Published 2 May 2020 in cs.CL, cs.AI, and cs.LG

Abstract: Building equitable and inclusive NLP technologies demands consideration of whether and how social attitudes are represented in ML models. In particular, representations encoded in models often inadvertently perpetuate undesirable social biases from the data on which they are trained. In this paper, we present evidence of such undesirable biases towards mentions of disability in two different English LLMs: toxicity prediction and sentiment analysis. Next, we demonstrate that the neural embeddings that are the critical first step in most NLP pipelines similarly contain undesirable biases towards mentions of disability. We end by highlighting topical biases in the discourse about disability which may contribute to the observed model biases; for instance, gun violence, homelessness, and drug addiction are over-represented in texts discussing mental illness.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (6)
  1. Ben Hutchinson (25 papers)
  2. Vinodkumar Prabhakaran (48 papers)
  3. Emily Denton (18 papers)
  4. Kellie Webster (14 papers)
  5. Yu Zhong (27 papers)
  6. Stephen Denuyl (1 paper)
Citations (275)

Summary

Overview of the ACL 2020 Proceedings Instructions

The document in review provides comprehensive guidelines for authors preparing manuscripts for the ACL 2020 proceedings. It serves as both an instructional manual and an exemplar, ensuring adherence to established formatting standards required for submissions. The structured nature of the document underscores several critical elements for authors submitting their work to the conference.

Key Sections and Instructions

  1. Document Formatting and Structure: The document outlines the necessary formatting prerequisites, including layout specifics such as the two-column format, margin measurements, and font requirements. Authors are reminded of the importance of consistent styling across submissions to maintain uniformity within the published proceedings.
  2. Submission Length and Anonymity: Strict page limits are dictated for both long and short papers. Crucially, the document emphasizes the anonymization of submissions to facilitate double-blind reviewing. Authors must omit any identifiable information and ensure citations of their previous work do not compromise their anonymity.
  3. Resources and Templates: Authors are encouraged to utilize the provided \LaTeX templates available on the ACL 2020 website. These templates are tailored to help conform submissions to the specific requirements of the conference, ensuring the inclusion of standard elements such as bibliographies and DOIs.
  4. Supplementary Materials and Multiple Submissions Policy: Submissions may include supplementary materials that aid in the understanding or replication of research, though these materials should not be integral to the paper. The guidance clarifies the policy regarding simultaneous submissions to multiple venues and emphasizes transparency concerning overlapping content between submissions.
  5. References and Citations: Documentation on the proper format for references and citations is provided, aligning with the goal of maintaining consistency and aiding in the readability and credibility of research work. Authors are guided on how to effectively cite prior work, ensuring the inclusion of DOIs when available.
  6. Accessibility Considerations: There is a directed effort towards ensuring that all figures and tables are distinguishable by all readers, including those who are colorblind, thereby enhancing the overall accessibility of the presentation materials.

Implications and Future Considerations

In the context of academic dissemination, such meticulously detailed guidelines greatly contribute to the standardization of research presentation, thereby aiding in the seamless exchange and review of ideas within the computational linguistics community. This procedural rigor not only facilitates efficient peer review but also elevates the professional standard to which submissions are held, thereby enhancing the overall quality of the conference outputs.

Looking forward, the continued evolution of submission guidelines could further integrate advancements in digital publishing and data accessibility, perhaps expanding the capacity for interactive content or integrating more robust data-sharing practices. As conferences like ACL continue to embrace technological advancements, future proceedings may explore innovative formats that further enhance collaborative opportunities among researchers.

Overall, this document serves a pivotal role in maintaining the high academic and professional standards synonymous with the ACL proceedings, exemplifying a commitment to academic excellence and innovation.