Proceedings of the 2019 Workshop on Widening NLP

Amittai Axelrod, Diyi Yang, Rossana Cunha, Samira Shaikh, Zeerak WaseemACL 2019

TL;DR

This document is the program for the 2019 Widening NLP Workshop, held at ACL 2019 in Florence, Italy. It showcases research aimed at increasing diversity and inclusion within the NLP community, highlighting papers that address underrepresentation and promote equitable participation.

What Problem Does This Paper Address?

The workshop program directly addresses the persistent underrepresentation of women and other minorities in the Natural Language Processing (NLP) community. This issue was identified as a critical concern, leading to the establishment of the Widening NLP Workshop starting in 2017. The workshop aims to provide a platform for overlooked voices and work, fostering a more inclusive and diverse research landscape. This year's workshop specifically sought to highlight diversity not only in terms of gender and ethnicity but also in scientific background, discipline, training, obtained degrees, and seniority.

Methods

This document is a program, not a research paper, so it does not detail specific research methods. However, the listed papers cover a wide range of NLP topics, including:

  • Sentiment Lexicons: Developing resources for under-resourced languages.
  • Grammar Development: Creating grammatical resources for low-resource languages.
  • Speech Recognition: Utilizing deep neural networks for languages with limited data.
  • Word Sense Disambiguation: Employing distributional semantics for enhanced accuracy.
  • Aspect-based Sentiment Analysis: Developing models for nuanced opinion mining.
  • Sequence Prediction: Applying various models for language modeling tasks.
  • Relation Extraction: Building frameworks for cross-dataset and cross-domain extraction.
  • Bias Detection and Mitigation: Identifying and addressing sexism and gender bias in NLP models and data.
  • Machine Translation: Developing and evaluating translation systems, particularly for low-resource language pairs.
  • Chatbots and Dialogue Systems: Creating more inclusive and ethical conversational agents.
  • NLP for Social Good: Applying NLP techniques to understand and address societal issues.

The workshop also introduced a "workshopping feature" where authors could opt-in to provide feedback to peers before formal review, promoting a collaborative and supportive environment.

Datasets

The program lists numerous papers that likely utilize various datasets. Specific details on datasets used for training and evaluation are not provided in this program document. However, the titles suggest the use of:

  • General NLP Benchmarks: Standard datasets for tasks like sentiment analysis, machine translation, and named entity recognition.
  • Specialized Datasets: Datasets tailored for specific languages (e.g., Igbo, Runyankore, Rukiga, Tigrinya, Amharic, Urdu, Mixtec, Kurdish, China English) or specific phenomena (e.g., sexism, offensive language, child-directed speech, discourse relations, radiological reports).
  • Newly Introduced Resources: Some papers likely introduce new corpora or lexicons, such as a general-purpose sentiment lexicon for Igbo, a resource grammar for Runyankore and Rukiga, and a word analogy task dataset for Indonesian (KaWAT).

Key Insights and Ideas

  • The Importance of Diverse Data: Many papers focus on developing resources and models for under-resourced languages, highlighting the critical need to move beyond English-centric NLP.
  • Addressing Bias is Crucial: A significant portion of the research is dedicated to identifying and mitigating various forms of bias (gender, racial, etc.) in NLP models and datasets, underscoring the ethical imperative in AI development.
  • Inclusion Benefits Everyone: The workshop's existence and the breadth of topics demonstrate a growing recognition that a more diverse NLP community leads to more robust, equitable, and impactful research.
  • Beyond Core NLP Tasks: The program showcases NLP applications in diverse areas, from understanding children's language to analyzing political discourse and even detecting offensive language directed at specific groups.

Key Results

As this is a workshop program, specific quantitative results are not presented here. However, the sheer volume and diversity of accepted papers (110 submissions received, with encouraging statistics on the diversity of accepted papers) indicate a strong and growing interest in the workshop's themes. The program features invited talks from prominent researchers in the field, including Drs. Emily M. Bender, Mona Diab, and Alvin Grissom II, as well as a panel discussion, suggesting high-quality contributions and engagement.

Limitations and Future Work

The program itself does not explicitly list limitations or future work for the workshop as a whole. However, the themes of the presented papers implicitly point to ongoing challenges and future directions:

  • Data Scarcity for Low-Resource Languages: A recurring theme is the need for more data and linguistic resources for languages other than English. Future work will likely focus on developing more efficient methods for data collection and augmentation.
  • Subtle Forms of Bias: Identifying and mitigating bias is an ongoing challenge. Future research will likely explore more sophisticated methods for detecting and correcting subtle biases that may not be immediately apparent.
  • Ethical Considerations in NLP: The workshop highlights the increasing importance of ethical considerations in NLP research and development. Future work will continue to explore frameworks for evaluating and ensuring the ethical deployment of NLP technologies.
  • Broader Community Engagement: The workshop's success suggests a desire for continued efforts to broaden participation and engagement within the NLP community.

Summary updated

BibTeX
@inproceedings{axelrod2019proceedings,
  author = {Amittai Axelrod and Diyi Yang and Rossana Cunha and Samira Shaikh and Zeerak Waseem},
  title = {{Proceedings of the 2019 Workshop on Widening NLP}},
  booktitle = {Annual Meeting of the Association for Computational Linguistics},
  year = {2019},
}
Figure