TrustNLP: Third Workshop on Trustworthy Natural Language Processing

Schedule

Invited Talk

Hal Daumé III: Trust in the Absence of Verifiability

When NLP systems, in particular large language models, generate claims that are easily verified by people to be untrue, trust is irrelevant. What matters is when they generate claims that are not (easily) verified to be true. Which raises two questions: (1) How much complementarity is there between what LLMs "know" and what people do, and (2) Can LLMs themselves provide missing complementary information. I'll discuss some good news and bad news that provide partial answers to these questions.

Hal Daumé III is a Volpi-Cupal Endowed Professor in Computer Science and Language Science at the University of Maryland, College Park; he has a joint appointment as a Senior Principal Researcher at Microsoft Research, New York City. His primary research interest is in developing new learning algorithms for prototypical problems that arise in the context of natural language processing and artificial intelligence, with a focus on interactive learning and understanding and minimizing social harms that can be caused or exacerbated by computational systems. He has received several awards, including best paper at AACL 2022, WMT 2020, ACL 2018, NAACL 2016, CEAS 2011 and ECML 2009, as well as best demo at NeurIPS 2015. He has been program chair for ICML 2020 (together with Aarti Singh) and for NAACL 2013 (together with Katrin Kirchhoff), and he was an inaugural diversity and inclusion co-chair at NeurIPS 2018 (with Katherine Heller).

Ramprasaath R. Selvaraju: Empowering Human-like Decision Making in AI Models through Explanations

Deep neural networks have achieved remarkable results in several Computer Vision tasks, but their increasing complexity poses challenges in terms of interpretability. In this talk, I will present my research on explainability in deep learning models, ranging from convolutional neural networks (CNNs) to multi-modal transformers, for tasks ranging from static image analysis to active perception, and demonstrate how it can enhance their human-likeness. I will focus on how interpretability can establish user trust, identify failure modes, provide targeted human feedback, debias models, ground representations, and facilitate compositional reasoning. Lastly, I will discuss future research directions. Ultimately, my talk aims to underscore the importance of interpretability in AI and its potential to advance the development of trustworthy, robust, and human-like machine learning models.

Ramprasaath is a Scientist at Apple TDG (Technology Development Group). Prior to this, he was a Sr. Research Scientist at Salesforce. He holds a PhD in Computer Science from the Georgia Institute of Technology, where he was advised by Devi Parikh and Dhruv Batra. His research lies at the intersection of computer vision, explainable AI and multi-modal pretraining. Specifically, his research focuses on building algorithms that provide explanations for decisions emanating from deep networks in order to build user trust, incorporate domain knowledge into AI, and correct for unwanted biases learned by deep AI models. Previously, he has held visiting positions at Brown, Oxford, Virginia Tech, Facebook, Samsung, Tesla and Microsoft. He obtained his Bachelor's degree in Electrical and Electronics Engineering and his Master's degree in Physics from Birla Institute of Technology and Science, Pilani.

Rachel Rudinger: Trustworthy Commonsense Inference for NLP

We use language to communicate about real-world situations, so it follows that trustworthy NLP systems ought to have basic knowledge of and ability to reason about real-world situations. Imbuing NLP systems with “common sense” requires models to learn useful generalizations about the world while avoiding the pitfalls of overgeneralization. In this talk, I will discuss a number of cases where models overgeneralize in ways that harm robustness and fairness. I will discuss work on uncovering these failure modes, as well as identifying their root causes.

Rachel Rudinger is an Assistant Professor in the Department of Computer Science at the University of Maryland, College Park. She holds joint appointments in the Department of Linguistics and the Institute for Advanced Computer Studies (UMIACS). In 2019, Rachel completed her Ph.D. in Computer Science at Johns Hopkins University in the Center for Language and Speech Processing. From 2019-2020, she was a Young Investigator at the Allen Institute for AI in Seattle, and a visiting researcher at the University of Washington. Her research interests include computational semantics, common-sense reasoning, and issues of social bias and fairness in NLP.

Sunipa Dev: Contextualizing Evaluations with Cross Cultural Perspectives

Evaluation of language models rely heavily on semi-structured data annotated by humans. Both the data and human perspectives involved in the process, thus play a key role in what is taken as ground truth by models. Historically, this perspective has been Western-oriented which leads to a lack of representation of global contexts and identities in models as well as evaluation strategies, and the risk of disregarding marginalized groups that are most significantly affected by implicit harms. Accounting for cross-cultural differences in interacting with technology is an important step for building and evaluating AI holistically. We will talk through different strategies including community engaged approaches, along with survey experiments to capture a more diverse set of perspectives in data curation and benchmarking efforts. We will also zoom in on how socio-culturally aware AI research can fill in the gaps in fairness evaluations.

Sunipa Dev is a Research Scientist at Google at their Responsible AI and Human Centered Technologies organization. She works at the intersection of language and society, trying to push the frontiers of inclusivity, equity, and representation in language technologies, and making them more socio-culturally aware. Some of her current work investigates what representation means across languages and cultures, and how to make emerging technologies more useful across the globe. Her work has received awards such as the NSF Computing Innovation Fellowship, and she was named as one of '100 Brilliant Women in AI Ethics' in 2022. She is also passionate about inclusivity in opportunities in NLP and is a core organizer of Widening NLP.

Accepted Papers

Towards Faithful Explanations for Text Classification with Robustness Improvement and Explanation Guided Training, Dongfang Li, Baotian Hu, Qingcai Chen and Shan He

Driving Context into Text-to-Text Privatization, Stefan Arnold, Dilara Yesilbas and Sven Weinzierl

Automated Ableism: An Exploration of Explicit Disability Biases in Sentiment and Toxicity Analysis Models, Pranav Narayanan Venkit, Mukund Srinath and Shomir Wilson

Pay Attention to the Robustness of Chinese Minority Language Models! Syllable-level Textual Adversarial Attack on Tibetan Script, Xi Cao, Dolma Dawa, Nuo Qun and Trashi Nyima

Can we trust the evaluation on ChatGPT?, Rachith Aiyappa, Jisun An, Haewoon Kwak and Yong-Yeol Ahn

Improving Factuality of Abstractive Summarization via Contrastive Reward Learning, I-Chun Chern, Zhiruo Wang, Sanjan Das, Bhavuk Sharma, Pengfei Liu and Graham Neubig

Examining the Causal Impact of First Names on Language Models: The Case of Social Commonsense Reasoning, Sullam Jeoung, Jana Diesner and Halil Kilicoglu

Reliability Check: An Analysis of GPT-3's Response to Sensitive Topics and Prompt Wording, Aisha Khatun and Daniel G. Brown

On the Privacy Risk of In-context Learning, Haonan Duan, Adam Dziedzic, Mohammad Yaghini, Nicolas Papernot and Franziska Boenisch

Sample Attackability in Natural Language Adversarial Attacks, Vyas Raina and Mark Gales

A Keyword Based Approach to Understanding the Overpenalization of Marginalized Groups by English Marginal Abuse Models on Twitter, Kyra Yee, Alice Schoenauer Sebag, Olivia Redfield, Matthias Eck, Emily Sheng and Luca Belli

An Empirical Study of Metrics to Measure Representational Harms in Pre-Trained Language Models, Saghar Hosseini, Hamid Palangi and Ahmed Hassan Awadallah

Linguistic Properties of Truthful Response, Bruce W. Lee, Benedict Florance Arockiaraj and Helen Jin

Debunking Biases in Attention, Shijing Chen, Usman Naseem and Imran Razzak

Guiding Text-to-Text Privatization by Syntax, Stefan Arnold and Dilara Yesilbas

Differentially Private In-Context learning, Ashwinee Panda, Tong Wu, Jiachen Wang and Prateek Mittal

Are fairness metric scores enough to assess discrimination biases in machine learning?, Fanny Jourdan, Laurent Risser, Jean-Michel Loubes and Nicholas Asher

DEPTH+: An Enhanced Depth Metric for Wikipedia Corpora Quality, Saied Alshahrani, Norah Alshahrani and Jeanna Matthews

Distinguishing Fact from Fiction: A Benchmark Dataset for Identifying Machine-Generated Scientific Papers in the LLM Era., Edoardo Mosca, Mohamed Hesham Ibrahim Abdalla, Paolo Basso, Margherita Musumeci and Georg Groh

Detecting Personal Information in Training Corpora: an Analysis, Nishant Subramani, Sasha Luccioni, Jesse Dodge and Margaret Mitchell

Enhancing textual counterfactual explanation intelligibility through Counterfactual Feature Importance, Milan Bhan, Jean-Noel Vittaut, Nicolas Chesneau and Marie-Jeanne Lesot

Privacy- And Utility-Preserving NLP with Anonymized data: A case study of Pseudonymization, Oleksandr Yermilov, Vipul Raheja and Artem Chernodub

Can NLP Models Identify, Distinguish, and Justify Questions that Don't have a Definitive Answer?, Ayushi Agarwal, Nisarg Patel, Neeraj Varshney, Mihir Parmar, Pavan Mallina, Aryan Shah, Srihari Raju Sangaraju, Tirth Patel, Nihar Thakkar and Chitta Baral

SMoA: Sparse Mixture of Adapters to Mitigate Multiple Dataset Biases, Yanchen Liu, Jing Yan, Yan Chen, Jing Liu and Hua Wu

GPTs Don't Keep Secrets: Searching for Backdoor Watermark Triggers in Autoregressive Language Models, Evan Lucas and Timothy Havens

Make Text Unlearnable: Exploiting Effective Patterns to Protect Personal Data, Xinzhe Li and Ming Liu

Bias Beyond English: Counterfactual Tests for Bias in Sentiment Analysis in Four Languages, Seraphina Goldfarb-Tarrant, Adam Lopez, Roi Blanco and Diego Marcheggiani

Training Data Extraction From Pre-trained Language Models: A Survey, Shotaro Ishihara

Expanding Scope: Adapting English Adversarial Attacks to Chinese, Hanyu Liu, Chengyuan Cai and Yanjun Qi

IMBERT: Making BERT Immune to Insertion-based Backdoor Attacks, Xuanli He, Jun Wang, Benjamin Rubinstein and Trevor Cohn

Large Language Models with Controllable Working Memory, Daliang Li, Ankit Singh Rawat, Manzil Zaheer, Xin Wang, Michal Lukasik, Andreas Veit, Felix Yu and Sanjiv Kumar

On The Real-world Performance of Machine Translation: Exploring Social Media Post-authors' Perspectives, Ananya Gupta, Jae D. Takeuchi and Bart Knijnenburg

ActiveAED: A Human in the Loop Improves Annotation Error Detection, Leon Weber and Barbara Plank

Shielded Representations: Protecting Sensitive Attributes Through Iterative Gradient-Based Projection, Shadi Iskander, Kira Radinsky and Yonatan Belinkov

Keeping Up with the Language Models: Robustness-Bias Interplay in NLI Data and Models, Ioana Baldini, Chhavi Yadav, Payel Das and Kush Varshney

Enabling Classifiers to Make Judgements Explicitly Aligned with Human Values, Yejin Bang, Tiezheng Yu, Andrea Madotto, Zhaojiang Lin, Mona Diab and Pascale Fung

This Prompt is Measuring : Evaluating Bias Evaluation in Language Models, Seraphina Goldfarb-Tarrant, Eddie Ungless, Esma Balkir and Su Lin Blodgett

COCKATIEL: COntinuous Concept ranKed ATtribution with Interpretable ELements for explaining neural net classifiers on NLP tasks, Fanny Jourdan, Agustin Picard, Laurent Risser, Jean-Michel Loubes and Nicholas Asher

Strength in Numbers: Estimating Confidence of Large Language Models by Prompt Agreement, Gwenyth Portillo Wightman, Alexandra DeLucia and Mark Dredze

Adversarial Named-Entity Recognition with Word Attributions and Disentanglement, Xiaomeng Jin, Bhanukiran Vinzamuri, Sriram Venkatapathy, Heng Ji and Pradeep Natarajan

Characterizing Political Bias in Automatic Summaries: A Case Study of Trump and Biden, Karen Zhou and Chenhao Tan

Model-tuning Via Prompts Makes NLP Models Adversarially Robust, Mrigank Raman, Pratyush Maini, Zico Kolter, Zachary C. Lipton and Danish Pruthi

Mitigating Bias for Question Answering Models by Tracking Bias Influence, Mingyu Derek Ma, Jiun-Yu Kao, Arpit Gupta, Yu-Hsiang Lin, Wenbo Zhao, Tagyoung Chung, Kai-Wei Chang and Nanyun Peng

PromptAttack: Probing Dialogue State Trackers with Adversarial Prompts, Xiangjue Dong, Yun He, Ziwei Zhu and James Caverlee

Defending against Insertion-based Textual Backdoor Attacks via Attribution, Jiazhao Li, Zhuofeng Wu, Wei Ping, Chaowei Xiao, V.G. Vinod Vydiswaran

FORK: A Bite-Sized Test Set for Probing Culinary Cultural Biases in Commonsense Reasoning Models, Shramay Palta and Rachel Rudinger

Call for papers

Overview

We invite papers which focus on developing models that are “explainable, fair, privacy-preserving, causal, and robust” (Trustworthy ML Initiative). Topics of interest include (but are not limited to):

• Differential Privacy
• Fairness and Bias: Evaluation and Treatments
• Model Explainability and Interpretability
• Accountability, Safety, and Robustness
• Ethics
• Industry applications of Trustworthy NLP
• Causal Inference and Fair ML
• Secure, Faithful, Trustworthy Data/Language Generation
• Toxic Language Detection and Mitigation
We also welcome contributions which draw upon interdisciplinary knowledge to advance Trustworthy NLP. This may include working with, synthesizing, or incorporating knowledge across expertise, sociopolitical systems, cultures, or norms.

Important Dates

• Apr 24, 2023: Workshop Paper Due Date (Direct Submission)

• May 10, 2023: Workshop Paper Due Date (Fast-Track)

• May 22, 2023: Notification of Acceptance

• June 2, 2023: Camera-ready papers due

• July 14, 2023: Workshop

Submission Policy

All submissions will be double-blind peer-reviewed (with author names and affiliations removed) by the program committee and judged by their relevance to the workshop themes.
Accepted and under-review papers are allowed to submit to the workshop but will not be included in the proceeding. Submitted manuscripts must be 8 pages long for full papers, and 4 pages long for short papers. Please follow ACL submission policies. Both full and short papers can have unlimited pages for references and appendices. Please note that at least one of the authors of each accepted paper must register for the workshop and present the paper. Template files can be found here: https://aclrollingreview.org/cfp#long-papers.

We also ask authors to include a limitation section and broader impact statement, following guidelines from the main conference.

Please submit to Softconf via this link

Fast-Track Submission

If your paper has been reviewed by ACL, EMNLP, EACL, or ARR and the average rating is higher than 2.5 (either avg soundness or excitement score), the paper is qualified to be submitted to the fast-track. In the appendix, please include the reviews and a short statement discussing what parts of the paper have been revised.

Non-Archival option

ACL workshops are traditionally archival. To allow dual submission of work, we are also including a non-archival track. If accepted, these submissions will still participate and present their work in the workshop. A reference to the paper will be hosted on the workshop website (if desired), but will not be included in the official proceedings. Please submit through softconf but indicate that this is a cross submission at the bottom of the submission form. You can also skip this step and inform us of your non-archival preference after the reviews.

Anonymity Period

We will follow NAACL’s anonymity policy, and require full anonymity until time of acceptance.

Speakers

Diyi Yang, Assistant Professor, School of Interactive Computing, Georgia Tech

Diyi Yang is an assistant professor in the School of Interactive Computing at Georgia Tech. Diyi has broad interests in NLP and Computational Social Science, including dialogue summarization, limited data learning, hate speech and bias, as well as responsible NLP for mental health. Her work has received multiple awards (or nominations) at EMNLP, ICWSM, SIGCHI, and CSCW. She is a Microsoft Research Faculty Fellow, a Forbes 30 under 30 in Science, an IEEE “AI 10 to Watch”, a recipient of the Intel Rising Star Faculty Award, the Samsung AI Researcher of the Year and the NSF CAREER Award.

Subho Majumdar, Senior Scientist, Splunk

Subho is a senior scientist in the Applied ML Research team of Splunk. Before recently joining Splunk, he spent 3 years in AT&T Data Science and AI Research (erstwhile part of AT&T Bell Labs). Throughout his career, he has worked on data-driven solutions that pushed the boundaries for a variety of challenging and cross-product problems. His current focus is on trustworthy machine learning in the wild: not only proposing novel solutions to technical problems that ensure qualities such as fairness, transparency, privacy, and robustness, but also implementing them in real-world use cases.

Fei Wang, Associate Professor in Division of Health Informatics, Cornell University

Fei Wang is an Associate Professor in Division of Health Informatics, Department of Population Health Sciences, Weill Cornell Medicine, Cornell University. His major research interest is data mining, machine learning and their applications in health data science. He has published more than 250 papers on the top venues of related areas such as ICML, KDD, NIPS, CVPR, AAAI, IJCAI, JAMA Internal Medicine, Annals of Internal Medicine, Lancet Digital Health, etc. His papers have received over 19,000 citations so far with an H-index 67. His (or his students’) papers have won 8 best paper (or nomination) awards at top international conferences on data mining and medical informatics. His team won the championship of the NIPS/Kaggle Challenge on Classification of Clinically Actionable Genetic Mutations in 2017 and Parkinson's Progression Markers' Initiative data challenge organized by Michael J. Fox Foundation in 2016. Dr. Wang is the recipient of the NSF CAREER Award in 2018, as well as the inaugural research leadership award in IEEE International Conference on Health Informatics (ICHI) 2019. Dr. Wang’s Research has been supported by NSF, NIH, ONR, PCORI, MJFF, AHA, Amazon, etc. Dr. Wang is the past chair of the Knowledge Discovery and Data Mining working group in American Medical Informatics Association (AMIA). Dr. Wang is a fellow of AMIA and a Distinguished Member of ACM.