Analyzing and interpreting neural networks for NLP

Main page

All plenary sessions will be held in zoom, where we will livestream the keynote and oral presentations, that are followed by a live Q&A. Questions can be asked in rocket chat during the presentations. Poster sessions will be held in gather town. The links to the zoom session, gather.town space and the rocketchat channel can be found on the EMNLP page of the workshop (only accessible with conference registration).

Summary of the program (Time indications: Punta Cana)

Session	Time	Location
Opening remarks	4:00 - 4:15	Zoom
Keynote speaker 1 – Anna Rogers	4:15 - 5:00	Zoom
Break
Oral presentations 1	5:15 - 6:00	Zoom
Break
Demo presentation	6:15 - 6:30	Zoom
Poster session 1	6:30 - 8:00	gather.town room K-N
Break
Keynote speaker 1 – Anna Rogers	8:15 - 9:00	Zoom
Break
Keynote speaker 2 – Roger Levy	11:00 - 11:45	Zoom
Break
Oral presentations 2	12:00 - 13:00	Zoom
Break
Keynote speaker 3 – Idan Blank	13:15 - 14:00	Zoom
Awards and closing remarks	14:00 - 14:20	Zoom
Demo presentation - repetition	14:25 - 14:40	Zoom
Break
Poster session 2	14:30 - 16:00	gather.town room K-N
Break
Keynote speaker 2 – Roger Levy	19:00 - 19:45	Zoom
Break
Poster session 3	20:00 - 21:30	gather.town room K-N
Oral session 3	21:30 - 22:45	Zoom
Break
Keynote speaker 3 – Idan Blank	23:00 - 23:45	Zoom

Oral presentation session 1 (Time indications: Punta Cana)

5:15 - 5:27 The elephant in the interpretability room: Why use attention as explanation when we have saliency methods?. Jasmijn Bastings and Katja Filippova .
5:27 - 5:39 BERTs of a feather do not generalize together: Large variability in generalization across models with similar test set performance. R. Thomas Mccoy, Junghyun Min and Tal Linzen.
5:39 - 5:51 Evaluating Attribution Methods using White-Box LSTMs. Yiding Hao.
5:51 - 6:00 Live Q&A with all paper authors.

Oral presentation session 2 (Time indications: Punta Cana)

12:00 - 12:12 What Happens To BERT Embeddings During Fine-tuning?. Amil Merchant, Elahe Rahimtoroghi, Ellie Pavlick and Ian Tenney.
12:12 - 12:24 The elephant in the interpretability room: Why use attention as explanation when we have saliency methods?. Jasmijn Bastings and Katja Filippova (repetition).
12:24 - 12:36 Dissecting Lottery Ticket Transformers: Structural and Behavioral Study of Sparse Neural Machine Translation. Rajiv Movva and Jason Zhao.
12:36 - 12:48 The EOS Decision and Length Extrapolation. Benjamin Newman, John Hewitt, Percy Liang and Christopher D. Manning.
12:48 - 13:00 Live Q&A With all paper authors.

Oral presentation session 3 (Time indications: Punta Cana)

21:30 - 21:42 Evaluating Attribution Methods using White-Box LSTMs. Yiding Hao (repetition).
21:42 - 21:54 The EOS Decision and Length Extrapolation. Benjamin Newman, John Hewitt, Percy Liang and Christopher D. Manning (repetition).
21:54 - 22:06 BERTs of a feather do not generalize together: Large variability in generalization across models with similar test set performance. R. Thomas Mccoy, Junghyun Min and Tal Linzen (repetition).
22:06 - 22:18 Dissecting Lottery Ticket Transformers: Structural and Behavioral Study of Sparse Neural Machine Translation. Rajiv Movva and Jason Zhao (repetition).
22:18 - 22:30 What Happens To BERT Embeddings During Fine-tuning?. Amil Merchant, Elahe Rahimtoroghi, Ellie Pavlick and Ian Tenney (repetition).
22:30 - 22:45 Live Q&A With all paper authors.

Poster session 1

Archival papers

BERTering RAMS: What and How Much does BERT Already Know About Event Arguments? - A Study on the RAMS Dataset Varun Gangal and Eduard Hovy.
Exploring Neural Entity Representations for Semantic Information Andrew Runge and Eduard Hovy.
Structured Self-Attention Weights Encodes Semantics in Sentiment Analysis Zhengxuan Wu, Thanh-Son Nguyen and Desmond Ong.
This is a BERT. Now there are several of them. Can they generalize to novel words? Coleman Haley.
Attacking Semantic Similarity: Generating Second-Order NLP Adversarial Examples John Morris.
Discovering the Compositional Structure of Vector Representations with Role Learning Networks Paul Soulos, R. Thomas Mccoy, Tal Linzen and Paul Smolensky.
Examining the rhetorical capacities of neural language models Zining Zhu, Chuer Pan, Mohamed Abdalla and Frank Rudzicz.
Linguistically-Informed Transformations (LIT): A Method for Automatically Generating Contrast Sets Chuanrong Li, Lin Shengshuo, Zeyu Liu, Xinyi Wu, Xuhui Zhou and Shane Steinert-Threlkeld.
Unsupervised Distillation of Syntactic Information from Contextualized Word Representations Shauli Ravfogel, Yanai Elazar, Jacob Goldberger and Yoav Goldberg.
Investigating Novel Verb Learning in BERT: Selectional Preference Classes and Alternation-Based Syntactic Generalization Tristan Thrush, Ethan Wilcox and Roger Levy.
Neural Natural Language Inference Models Partially Embed Theories of Lexical Entailment and Negation Atticus Geiger, Kyle Richardson and Christopher Potts.
Probing for Multilingual Numerical Understanding in Transformer-Based Language Models Devin Johnson, Denise Mak, Andrew Barker and Lexi Loessberg-Zahl.
Do Language Embeddings capture Scales? Xikun Zhang, Deepak Ramachandran, Ian Tenney, Yanai Elazar and Dan Roth.
Searching for a Search Method: Benchmarking Search Algorithms for Generating NLP Adversarial Examples Jin Yong Yoo, John Morris, Eli Lifland and Yanjun Qi.

Extended abstracts

Inductive Biases of Seq2seq Learners under Minimal Stimuli. Eugene Kharitonov and Rahma Chaabouni.
How Much Does RoBERTa Know About Quantifiers? An Assessment via Natural Language Inference. Authors: Cedegao Zhang.
Interpreting Neural Networks with Topic Models: Evidence from “Predicting In-game Actions From Interviews of NBA Players”. Amir Feder, Nadav Oved and Roi Reichart.
BERT’s Adaptability to Small Data. Héctor Vázquez Martínez and Annika Heuser.
The Linear Geometry of Contextualized Word Representations. Evan Hernandez and Jacob Andreas.
CausaLM: Causal Model Explanation Through Counterfactual Language Models. Amir Feder, Nadav Oved, Uri Shalit and Roi Reichart.
Analyzing saliency in neural content scoring models for science explanations. Brian Riordan, Sarah Bichler, Allison Bradford and Marcia Linn.
Mighty Morpho-Probing Models. Naomi Tachikawa Shapiro, Amandalynne Paullada and Shane Steinert-Threlkeld.

Findings papers

Improving Text Understanding via Deep Syntax-Semantics Communication. Hao Fei, Yafeng Ren and Donghong Ji.
Exploring BERT’s sensitivity to lexical cues using tests from semantic priming. Kanishka Misra, Allyson Ettinger and Julia Rayz.
Corpora Evaluation and System Bias Detection in Multi-document Summarization. Alvin Dey, Tanya Chowdhury, Yash Kumar and Tanmoy Chakraborty.
On the Language Neutrality of Pre-trained Multilingual Representations. Jindřich Libovický, Rudolf Rosa and Alexander Fraser.
How Can Self-Attention Networks Recognize Dyck-n Languages?. Javid Ebrahimi, Dhruv Gelda and Wei Zhang.
Why and when should you pool? Analyzing Pooling in Recurrent Architectures. Pratyush Maini, Keshav Kolluru, Danish Pruthi and Mausam.
Rethinking Self-Attention: Towards Interpretability in Neural Parsing. Khalil Mrini, Franck Dernoncourt, Quan Hung Tran, Trung Bui, Walter Chang and Ndapa Nakashole.
Reducing Sentiment Bias in Language Models via Counterfactual Evaluation. Po-Sen Huang, Huan Zhang, Ray Jiang, Robert Stanforth, Johannes Welbl, Jack Rae, Vishal Maini, Dani Yogatama, Pushmeet Kohli.
Towards Controllable Biases in Language Generation. Emily Sheng, Kai-Wei Chang, Prem Natarajan and Nanyun Peng.
Leakage-Adjusted Simulatability: Can Models Generate Non-Trivial Explanations of Their Behavior in Natural Language?. Peter Hase, Shiyue Zhang, Harry Xie and Mohit Bansal.
Event-Related Bias Removal for Real-time Disaster Events. Salvador Medina Maza, Evangelia Spiliopoulou, Eduard Hovy and Alexander Hauptmann.
Undersensitivity in Neural Reading Comprehension. Johannes Welbl, Pasquale Minervini, Max Bartolo, Pontus Stenetorp and Sebastian Riedel.
Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs. Ana Marasović, Chandra Bhagavatula, Jae sung Park, Ronan Le Bras, Noah A. Smith and Yejin Choi.
Optimizing Word Segmentation for Downstream Task. Tatsuya Hiraoka, Sho Takase, Kei Uchiumi, Atsushi Keyaki and Naoaki Okazaki.
Assessing Robustness of Text Classification through Maximal Safe Radius Computation. Emanuele La Malfa, Min Wu, Luca Laurenti, Benjie Wang, Anthony Hartshorn and Marta Kwiatkowska.
Evaluating Factuality in Generation with Dependency-level Entailment. Tanya Goyal and Greg Durrett.
Weakly- and Semi-supervised Evidence Extraction. Danish Pruthi, Bhuwan Dhingra, Graham Neubig and Zachary C. Lipton.
On the Sub-Layer Functionalities of Transformer Decoder Yilin Yang, Longyue Wang, Shuming Shi, Prasad Tadepalli, Stefan Lee and Zhaopeng Tu.

Poster session 2

Archival papers

The Explanation Game: Towards Prediction Explainability through Sparse Communication Marcos Treviso and André F. T. Martins.
Defining Explanation in an AI Context Tejaswani Verma, Christoph Lingenfelder and Dietrich Klakow .
Controlling the Imprint of Passivization and Negation in Contextualized Representations Hande Celikkanat, Sami Virpioja, Jörg Tiedemann and Marianna Apidianaki .
How does BERT capture semantics? A closer look at polysemous words David Yenicelik, Florian Schmidt and Yannic Kilcher .
Leveraging Extracted Model Adversaries for Improved Black Box Attacks. Naveen Jafer Nizar and Ari Kobren.
Attacking Semantic Similarity: Generating Second-Order NLP Adversarial Examples John Morris.
BERTnesia: Investigating the capture and forgetting of knowledge in BERT Jaspreet Singh, Jonas Wallat and Avishek Anand.
Linguistically-Informed Transformations (LIT): A Method for Automatically Generating Contrast Sets Chuanrong Li, Lin Shengshuo, Zeyu Liu, Xinyi Wu, Xuhui Zhou and Shane Steinert-Threlkeld.
Latent Tree Learning with Ordered Neurons: What Parses Does It Produce? Yian Zhang.
Emergent Language Generalization and Acquisition Speed are not tied to Compositionality. Eugene Kharitonov and Marco Baroni.
Neural Natural Language Inference Models Partially Embed Theories of Lexical Entailment and Negation Atticus Geiger, Kyle Richardson and Christopher Potts.
It’s not Greek to mBERT: Inducing Word-Level Translations from Multilingual BERT. Hila Gonen, Shauli Ravfogel, Yanai Elazar and Yoav Goldberg.
On the Interplay Between Fine-tuning and Sentence-Level Probing for Linguistic Knowledge in Pre-Trained Transformers. Marius Mosbach, Anna Khokhlova, Michael A. Hedderich and Dietrich Klakow.
Unsupervised Evaluation for Question Answering with Transformers. Lukas Muttenthaler, Isabelle Augenstein and Johannes Bjerva.
Do Language Embeddings capture Scales? Xikun Zhang, Deepak Ramachandran, Ian Tenney, Yanai Elazar and Dan Roth.
Searching for a Search Method: Benchmarking Search Algorithms for Generating NLP Adversarial Examples Jin Yong Yoo, John Morris, Eli Lifland and Yanjun Qi.

Extended abstracts

How Much Does RoBERTa Know About Quantifiers? An Assessment via Natural Language Inference. Authors: Cedegao Zhang.
Inductive Biases of Seq2seq Learners under Minimal Stimuli. Eugene Kharitonov and Rahma Chaabouni.
Interpreting Neural Networks with Topic Models: Evidence from “Predicting In-game Actions From Interviews of NBA Players”. Amir Feder, Nadav Oved and Roi Reichart.
The Linear Geometry of Contextualized Word Representations. Evan Hernandez and Jacob Andreas.
Mighty Morpho-Probing Models. Naomi Tachikawa Shapiro, Amandalynne Paullada and Shane Steinert-Threlkeld.
CausaLM: Causal Model Explanation Through Counterfactual Language Models. Amir Feder, Nadav Oved, Uri Shalit and Roi Reichart.
Analyzing saliency in neural content scoring models for science explanations. Brian Riordan, Sarah Bichler, Allison Bradford and Marcia Linn.
Analyzing Neural Machine Translation Trained Using Targeted Part-Of-Speech. Subhadarshi Panda.

Findings papers

Improving Text Understanding via Deep Syntax-Semantics Communication. Hao Fei, Yafeng Ren and Donghong Ji.
On the Language Neutrality of Pre-trained Multilingual Representations. Jindřich Libovický, Rudolf Rosa and Alexander Fraser.
Corpora Evaluation and System Bias Detection in Multi-document Summarization. Alvin Dey, Tanya Chowdhury, Yash Kumar and Tanmoy Chakraborty.
Interpretable Entity Representations through Large-Scale Typing. Yasumasa Onoe and Greg Durrett.
Be Different to Be Better! A Benchmark to Leverage the Complementarity of Language and Vision. Sandro Pezzelle, Claudio Greco, Greta Gandolfi, Eleonora Gualdoni and Raffaella Bernardi.
NLP Service APIs and Models for Efficient Registration of New Clients. Sahil Shah, Vihari Piratla, Soumen Chakrabarti and Sunita Sarawagi.
Rethinking Self-Attention: Towards Interpretability in Neural Parsing. Khalil Mrini, Franck Dernoncourt, Quan Hung Tran, Trung Bui, Walter Chang and Ndapa Nakashole.
Reducing Sentiment Bias in Language Models via Counterfactual Evaluation. Po-Sen Huang, Huan Zhang, Ray Jiang, Robert Stanforth, Johannes Welbl, Jack Rae, Vishal Maini, Dani Yogatama, Pushmeet Kohli.
Undersensitivity in Neural Reading Comprehension. Johannes Welbl, Pasquale Minervini, Max Bartolo, Pontus Stenetorp and Sebastian Riedel.
Assessing Robustness of Text Classification through Maximal Safe Radius Computation. Emanuele La Malfa, Min Wu, Luca Laurenti, Benjie Wang, Anthony Hartshorn and Marta Kwiatkowska.
Universal Dependencies According to BERT: Both More Specific andMore General. Tomasz Limisiewicz, David Mareček and Rudolf Rosa.
LSTMS Compose—and Learn—Bottom-Up. Naomi Saphra and Adam Lopez.
What’s so special about BERT’s layers? A closer look at the NLP pipeline in monolingual and multilingual models. Wietse de Vries, Andreas van Cranenburgh and Malvina Nissim.

Poster session 3

Archival papers

BERTnesia: Investigating the capture and forgetting of knowledge in BERT Jaspreet Singh, Jonas Wallat and Avishek Anand.
Probing for Multilingual Numerical Understanding in Transformer-Based Language Models Devin Johnson, Denise Mak, Andrew Barker and Lexi Loessberg-Zahl.
Examining the rhetorical capacities of neural language models Zining Zhu, Chuer Pan, Mohamed Abdalla and Frank Rudzicz.
On the Interplay Between Fine-tuning and Sentence-Level Probing for Linguistic Knowledge in Pre-Trained Transformers Marius Mosbach, Anna Khokhlova, Michael A. Hedderich and Dietrich Klakow.
Emergent Language Generalization and Acquisition Speed are not tied to Compositionality Eugene Kharitonov and Marco Baroni.
Unsupervised Distillation of Syntactic Information from Contextualized Word Representations Shauli Ravfogel, Yanai Elazar, Jacob Goldberger and Yoav Goldberg.
It’s not Greek to mBERT: Inducing Word-Level Translations from Multilingual BERT Hila Gonen, Shauli Ravfogel, Yanai Elazar and Yoav Goldberg.
Investigating Novel Verb Learning in BERT: Selectional Preference Classes and Alternation-Based Syntactic Generalization Tristan Thrush, Ethan Wilcox and Roger Levy.
Discovering the Compositional Structure of Vector Representations with Role Learning Networks Paul Soulos, R. Thomas Mccoy, Tal Linzen and Paul Smolensky.
Unsupervised Evaluation for Question Answering with Transformers Lukas Muttenthaler, Isabelle Augenstein and Johannes Bjerva.
This is a BERT. Now there are several of them. Can they generalize to novel words? Coleman Haley.
Leveraging Extracted Model Adversaries for Improved Black Box Attacks Naveen Jafer Nizar and Ari Kobren.
Latent Tree Learning with Ordered Neurons: What Parses Does It Produce? Yian Zhang.
BERTering RAMS: What and How Much does BERT Already Know About Event Arguments? - A Study on the RAMS Dataset Varun Gangal and Eduard Hovy.
The Explanation Game: Towards Prediction Explainability through Sparse Communication Marcos Treviso and André F. T. Martins.
Controlling the Imprint of Passivization and Negation in Contextualized Representations Hande Celikkanat, Sami Virpioja, Jörg Tiedemann and Marianna Apidianaki.
How does BERT capture semantics? A closer look at polysemous words David Yenicelik, Florian Schmidt and Yannic Kilcher.
Exploring Neural Entity Representations for Semantic Information Andrew Runge and Eduard Hovy.
Structured Self-Attention Weights Encodes Semantics in Sentiment Analysis Zhengxuan Wu, Thanh-Son Nguyen and Desmond Ong.
Defining Explanation in an AI Context Tejaswani Verma, Christoph Lingenfelder and Dietrich Klakow.

Extended abstracts

Analyzing Neural Machine Translation Trained Using Targeted Part-Of-Speech. Subhadarshi Panda.
BERT’s Adaptability to Small Data. Héctor Vázquez Martínez and Annika Heuser.

Findings papers

Optimizing Word Segmentation for Downstream Task. Tatsuya Hiraoka, Sho Takase, Kei Uchiumi, Atsushi Keyaki and Naoaki Okazaki.
Universal Dependencies According to BERT: Both More Specific andMore General. Tomasz Limisiewicz, David Mareček and Rudolf Rosa.
Natural Language Rationales with Full-Stack Visual Reasoning: From Pixels to Semantic Frames to Commonsense Graphs. Ana Marasović, Chandra Bhagavatula, Jae sung Park, Ronan Le Bras, Noah A. Smith and Yejin Choi.
Evaluating Factuality in Generation with Dependency-level Entailment. Tanya Goyal and Greg Durrett.
What’s so special about BERT’s layers? A closer look at the NLP pipeline in monolingual and multilingual models. Wietse de Vries, Andreas van Cranenburgh and Malvina Nissim.
On the Sub-Layer Functionalities of Transformer Decoder Yilin Yang, Longyue Wang, Shuming Shi, Prasad Tadepalli, Stefan Lee and Zhaopeng Tu.
Leakage-Adjusted Simulatability: Can Models Generate Non-Trivial Explanations of Their Behavior in Natural Language?. Peter Hase, Shiyue Zhang, Harry Xie and Mohit Bansal.
Interpretable Entity Representations through Large-Scale Typing. Yasumasa Onoe and Greg Durrett.
Towards Controllable Biases in Language Generation. Emily Sheng, Kai-Wei Chang, Prem Natarajan and Nanyun Peng.
Be Different to Be Better! A Benchmark to Leverage the Complementarity of Language and Vision. Sandro Pezzelle, Claudio Greco, Greta Gandolfi, Eleonora Gualdoni and Raffaella Bernardi.
How Can Self-Attention Networks Recognize Dyck-n Languages?. Javid Ebrahimi, Dhruv Gelda and Wei Zhang.
NLP Service APIs and Models for Efficient Registration of New Clients. Sahil Shah, Vihari Piratla, Soumen Chakrabarti and Sunita Sarawagi.
Why and when should you pool? Analyzing Pooling in Recurrent Architectures. Pratyush Maini, Keshav Kolluru, Danish Pruthi and Mausam.
Exploring BERT’s sensitivity to lexical cues using tests from semantic priming. Kanishka Misra, Allyson Ettinger and Julia Rayz.
LSTMS Compose—and Learn—Bottom-Up. Naomi Saphra and Adam Lopez.