Since its inception in 1962, the Association for Computational Linguistics (ACL) has been “the premier international scientific and professional society for people working on computational problems involving human language, a field often referred to as either computational linguistics (CL) or natural language processing (NLP).”
ACL activities include the annual summer meeting and publication (via MIT Press) of the Computational Linguistics journal.The 60th Annual Meeting of the ACL took place from May 22–27, 2022 as a hybrid event, in Dublin and online.
These were the Top 10 themes in this year’s conference.
- Machine Learning for NLP
- Machine Translation and “Multilinguality” (Multilingualism)
- Interpretability and Analysis of Models for NLP
- Information Extraction
- NLP Applications
- Dialogue and Interactive Systems
- Resources and Evaluation
- Question Answering
Summarization, low-resource languages, speech technologies, multimodality, and ethics were also covered, albeit to a lesser extent.
And the Award for Best Paper Goes to…
More than 2,000 authors from all over the world contributed to the conference, either via long or short paper (604 long papers, 98 short papers).
Best Paper Award for 2022
Learned Incremental Representations for Parsing
The authors of this paper designed a representation for parsing that is “maximally speculation free.” As they noted, “human capabilities suggest that we should also be able to build accurate parsers that […] operate incrementally.”
Best Special Theme Paper Award
Requirements and Motivations of Low-Resource Speech Synthesis for Language Revitalization
Aidan Pine and Patrick William Littell, National Research Council Canada; Dan Wells, PhD student and Korin Richmond, Associate Professor, University of Edinburgh; Nathan Brinklow, Professor, Queen’s University
The authors aimed to revitalize three indigenous languages by developing speech synthesis systems for them, reevaluating the question of “how much data is required to build low-resource speech synthesis systems featuring state-of-the-art neural models.”
Best Resource Paper Award
DiBiMT: A Novel Benchmark for Measuring Word Sense Disambiguation Biases in Machine Translation
Niccolò Campolungo, Federico Martelli, and Roberto Navigli, Sapienza University of Rome; Francesco Saina, SSML Carlo Bo
This paper highlights the importance of bias in machine translation (MT). DiBiMT, is a “novel benchmark for measuring and understanding semantic biases in NMT, which goes beyond simple accuracy and provides novel metrics that summarize how biased NMT models are.”
Best Linguistic Insight Paper Award
This paper demonstrated the effectiveness of explicitly incorporating morphological information in language-model pretraining. The authors proposed a two-tier BERT architecture (first tier encodes morphological information; second one, sentence-level information) and evaluated it on the low-resource, morphologically rich Kinyarwanda language. The authors said this work “should motivate more research into morphology-aware language models.”
The richness and quality of the conference papers is reflected in the Outstanding Papers list as judged by the Best Paper Committee.
- Evaluating Factuality in Text Simplification by Ashwin Devaraj, William Berkeley Sheffield, and Junyi Jessy Li, University of Texas at Austin, and Byron C Wallace, Khoury College of Computer Sciences, Northeastern University
- Online Semantic Parsing for Latency Reduction in Task-Oriented Dialogue by Jiawei Zhou, Harvard University and Jason Eisner, Michael Newman, Emmanouil Antonios Platanios and Sam Thomson, Microsoft Semantic Machines
- Learning to Generalize to More: Continuous Semantic Augmentation for Neural Machine Translation by Xiangpeng Wei, Heng Yu, Rongxiang Weng, Weihua Luo and Rong Jin, Alibaba DAMO Academy, and Yue Hu, Chinese Academy of Sciences
- Fantastically Ordered Prompts and Where to Find Them: Overcoming Few-Shot Prompt Order Sensitivity by Yao Lu, Max Bartolo, Sebastian Riedel, and Pontus Stenetorp, UCL, and Alastair Moore, Mishcon de Reya LLP
- Inducing Positive Perspectives with Text Reframing by Caleb Ziems, Anthony Zhang, and Diyi Yang, Georgia Institute of Technology, and Minzhi Li from National University of Singapore
- Ditch the Gold Standard: Re-evaluating Conversational Question Answering by Huihan Li, Tianyu Gao, Manan Goenka, and Danqi Chen, Princeton University
- Active Evaluation: Efficient NLG Evaluation with Few Pairwise Comparisons by Akash Kumar Mohankumar, Microsoft; Mitesh M Khapra Indian Institute of Technology Madras
- Compression of Generative Pre-trained Language Models via Quantization by Chaofan Tao, Ping Luo, and Ngai Wong, University of Hong Kong, and Lu Hou, Wei Zhang, Lifeng Shang, Xin Jiang, and Qun Liu, Huawei Noah’s Ark Lab
The 60-60 Diversity and Inclusion Initiative
To celebrate its 60th year, ACL launched the 60-60 Initiative as an effort to “remove the ingrained linguistic bias in the scientific landscape in general and CL science in particular.”
The inauguration of this Diversity & Inclusion Special Initiative has already been embraced by a core group comprising academic teams from across the globe (National University of Singapore, Yale University, University of Illinois Urbana Champaign, NYUAD, and King Saud University), big tech (Baidu, Meta), medium tech (AppTek) companies, non-profit organizations (AI2), startups (aiXplain), as well as annotation companies (YaiGlobal).
The 60-60 Initiative features some strategic milestones to be reached by 2023, including
- the complete translation of the entire ACL Anthology into 60 languages;
- a comprehensive standardized scientific and CL terminology list in 60 languages;
- live cross-lingual closed captions and dubbing into 60 languages;
- a comprehensive repository for all the talks and videos from the CL community curated and translated into 60 languages.