NAACL-HLT 2021

Increasing Financial Accessibility in NAACL

2021-06-18T00:00:00+00:00

Overview

The D&I Committee at NAACL 2021 strived to increase participation by providing subsidies which could help offset financial burdens to contribute towards reducing financial burdens on participants. Following in the footsteps of previous D&I Committee efforts, the subsidies offered by the committee contributed towards arranging, publicizing, and awarding financial subsidies to offset conference costs. The subsidies covered several requests:

For broadband or VPN costs - contributing towards more accessibility for the conference across several regions as well as providing privacy options for participants who required it
For registration - contributing towards more reachability especially for participants from developing countries
For caregiving and disability support - contributing towards providing a better experience for participants who require caregiving support as well as navigation for the virtual format

As with previous iterations of virtual conferences, subsidies could focus on coverage of conference registration and ACL membership fees. A positive consequence of the shift to a virtual conference format is the potential for participation by community members who would have been unable to travel to the physical conference location. To take advantage of this change, when soliciting applications for financial subsidies, the subcommittee encouraged applications from all interested community members regardless of whether the prospective attendee had an accepted paper in the conference. About 60% applicants did not have an accepted paper at the conference.

D&I Subsidies

The application for the D&I subsidy went online on April 20, 2021, with a deadline on May 6, 2021. The form was publicized through Twitter in addition to a blog post on the official website. The D&I chairs also shared the form via social media, along with organizations such as Queer in AI, Black in AI, LatinX in AI, AfricaNLP, and Masakhane. The form was kept open until the conference date, however requests made after the early registration deadline were considered on an emergency case-by-case basis. In total, we received over 210 applications. Over 70% applicants said they would not have registered for NAACL without subsidy.

The subsidy forms were heavily advertised in AfricaNLP, Masakhane, Queer in AI, Black in AI and LatinX in AI lists. All AmericasNLP and Student Research Workshop (SRW) accepted authors were strongly encouraged to apply for subsidies and the subsidy form link was added in the acceptance emails.

Figure 1: Number of applicants grouped across different affinity groups.

In terms of countries of residence of these applicants, the most strongly represented countries were the United States (23%) and India (14%). In total, 38 countries were represented in the applications we received. In terms of regions, US and Canada were the most represented (28%), along with South Asia (19%).

Figure 2: Number of applicants grouped across different geographical regions.

In selecting subsidy recipients, the subcommittee considered a number of factors – for example, whether the applicant was a student (57% graduate + 19% undergraduate students), was attending a *ACL conference for the first time (59% first-time attendees), came from an underrepresented geographical region (details mentioned above), or identified as a member of a minoritized group (details mentioned below).

The D&I awarded grants to 194 applicants, each receiving coverage for some combination of the conference registration fee and ACL membership fee (147), volunteership (33), and internet bandwidth/VPN/caregiving/disability support subsidy (14). The grants were given to almost all applicants who applied before the due date and had complete and correct information. We were selective for applicants who applied after the due date. Some of the awardees opted for volunteership, instead of D&I awards. In total, we awarded $13450 towards registration and membership fees, and offered another 14 accessibility + bandwidth + VPN grants $1100 (excluding fund-transfer fees) for subsidies.

Figure 3: Number of subsidy applicants across different countries. Darker shades denote more subsidy recipients.

Why are people applying for subsidies?

We found some patterns among applicants as to the reasons why they were applying for the subsidies. Some of these are listed below:

In academia, many undergraduate and masters student authors reported that they did not have access to other funding, and were not being supported by grants, fellowships or their universities to attend and present at the conference.
For researchers from developing regions, like Latin America, Africa and Asia, the power parity results in the need to apply for subsidies. Many folks from these regions reported that registration fees consisted of >50% of their monthly salaries, which is a significant sum.
Many folks also listed the financial crisis due to COVID-19 being a significant contributing factor in the need to apply for subsidies.

How can we do better in the future?

We recommend future conferences to continue prioritizing outreach to researchers from underrepresented regions to better bridge the gaps and make our community more inclusive. We also recommend making sure that the forms are available across regions, and are well publicized.

We also asked various affinity groups on how this scheme can be better in future. We talked to Diana Galvan and Luciana Benotti, two of the core organizers of LatinX in AI. Luciana is also a NAACL executive board member. They recommended:

Latin Americans are a largely underrepresented community at NAACL, while NAACL chapter serves the whole continent. At NAACL 2021, 71 Latin Americans registered in comparison to over 1600 from the United States and Canada. As of March 2021, ACL has 2970 NAACL chapter members, 2922 are from the United States and Canada, while 48 members are from Latin America. In contrast, 64% of the people in the American continent live in Latin America. This underrepresentation happens in a context in which reduced fees and virtual format of the conferences made them more inclusive. And also the D&I Committee at NAACL 2021 strived to increase participation of underrepresented communities. As in person conferences start to come back, the few Latin Americans present today in NAACL will face higher economic barriers. In order to promote the participation of this community, as the NAACL constitution suggests, I have the following recommendations for future NAACL diversity and inclusion efforts.

First, registration should be drastically reduced or waived for people affiliated with a Latin American university. Second, the NAACL Regional Americas Fund should consider funding more proposals, and should strive to also support those it cannot fund by promoting them in a transparent way. Third, there should be regular and dedicated funding for sponsoring in person conference attendance. Finally, remote participation at conferences should always be an option. Funding could come from the D&I sponsors.

The challenges that Latin American countries must face are often overlooked. Montoya and Rivas (2019) report salaries for engineers and researchers in the United States that go from $100K and up while in Latin America they start from $10K a year. Also unemployment rates are higher than 40% in some Latin American countries . Moreover, research funds have been cut in recent years. Finally, deaths for Covid in Latin America are more than one million in May 2021, that is more than 3 times the amount in India with half the population. In order to make NAACL really diverse, it is imperative to bear in mind the unequal conditions that political instability, economic collapse and health inequalities create for Latin American researchers.

Pranav A, Luca Soldaini and Sabine Weber, the core organizers from Queer in AI recommended:

LGBTQIA+ individuals are marginalized in many research fields due to, among other things, economic hardship and ostracization due to homophobia and transphobia; in particular, individuals who reside in countries with laws that severely undermine the rights of queer people face extreme hardship. Therefore, the *CL community should strive in implementing measures that not only allow current LGBTQ researchers to attend, but also open our conferences to queer individuals who might be interested in NLP research, but otherwise have no opportunity to get involved due to lack of support.

First, we recommend future D&I chairs to significantly ramp up advertising of D&I grants, both in terms of expanding advertising efforts and opening up applications months, rather than weeks, in advance. Ideally, D&I funding should be awarded at paper submission time, so that queer and other margizalized researchers can plan their submissions knowing that they would be able to attend the conference if their work is accepted.

Second, conference organizers should strive to make their event as queer-friendly as possible. Queer in AI recently released a guide on how to make virtual conferences as accessible for LGBTQIA+ individuals as possible; equal care should be placed in organizing physical and hybrid conferences once those are an option again.

Hady Elshar, one of the core organizers from Masakhane recommended:

There should be full support for event organizers and speakers. The other idea could be establishing an online process to allow donations and facilitate sponsorship: a very easy way to support people is to allow transferable tickets (through discount coupons maybe) . This allows somebody to buy for e.g. 10 tickets of *CL conference, get an immediate receipt (for tax returns) and send them as a gift to students or workshop organizers. The current system doesn’t allow donations and for sponsorship it is an invoice based which has an extremely long overload on event organizers.

Concluding remarks

The NAACL 2021 D&I committee thanks the Diversity and Inclusion: Champion sponsors, Google Research and Microsoft, that enabled us to provide financial subsidies. In the future, as the world adjusts again to the new normal, we recommend maintaining as well as increasing the accessibility considerations that have come to light during the COVID-19 pandemic.

Best Demo Award

2021-06-09T00:00:00+00:00

Congratulations to the recipients of the Best Demo Award at NAACL 2021!

COVID-19 Literature Knowledge Graph Construction and Drug Repurposing Report Generation
Qingyun Wang, Manling Li, Xuan Wang, Nikolaus Parulian, Guangxing Han, Jiawei Ma, Jingxuan Tu, Ying Lin, Ranran Haoran Zhang, Weili Liu, Aabhas Chauhan, Yingjun Guan, Bangzheng Li, Ruisong Li, Xiangchen Song, Yi Fung, Heng Ji, Jiawei Han, Shih-Fu Chang, James Pustejovsky, Jasmine Rah, David Liem, Ahmed ELsayed, Martha Palmer, Clare Voss, Cynthia Schneider and Boyan Onyshkevych

The paper will be presented on Wednesday June 9th 9:00-10:20am PDT. Please see the conference schedule for more details and hope to see you there!

Best Demo Award Committee

Vittorio Castelli - IBM Research AI
Matt Gardner - Allen Institute for Artificial Intelligence
Mona Diab - Facebook AI Research

We’d like to specially thank the Best Demo Paper Award committee who carefully reviewed all the papers, their reviews and pre-recorded demo videos to come to the final judgement.

Best Industry Paper Award

2021-06-08T00:00:00+00:00

Congratulations to the recipients of the Best Industry Paper Award at NAACL 2021 Industry Track!

Query2Prod2Vec: Grounded Word Embeddings for eCommerce
Federico Bianchi, Jacopo Tagliabue, Bingqing Yu

The paper will be presented on Wed June 9th at the end of IND6-Oral session (18:20–19:40 PDT). Please see the conference schedule for more details, and hope to see you there!

Careers in NLP Panel

2021-06-05T00:00:00+00:00

Name: Careers in NLP, a panel in the Industry Track of NAACL HLT 2021

Time: Tuesday June 8 at 1pm PDT, 4pm EDT.

Duration: 60 minutes

The questions are based on the survey results from this form. Additional questions will be taken from the audience if time permits.

Moderator

Philip Resnik

Philip Resnik is Professor of Linguistics at the University of Maryland, with a joint appointment at UMD’s Institute for Advanced Computer Studies and an affiliate appointment in CS. He was named an ACL Fellow in 2020. He earned his Ph.D. at the University of Pennsylvania in 1993, and joined the University of Maryland faculty in 1996. Philip’s industry experience prior to entering academia includes time at BBN, IBM T.J. Watson Research Center, and Sun Microsystems Laboratories. Alongside academia he has been a technical co-founder for CodeRyte (NLP for electronic health records, acquired by 3M in 2012), and an advisor to Converseon (social strategy and analytics), FiscalNote (machine learning and analytics for government relations), and SoloSegment (web site search and content optimization). He was recently appointed to the Scientific and Technical Advisory Board of the Coleridge Initiative, a non-profit working with governments to ensure that data are more effectively used for public decision-making. His current work focuses on computational social science, NLP for mental health, and the cognitive computational neuroscience of language.

Panelists

Isabelle Augenstein

Isabelle Augenstein is an associate professor at the University of Copenhagen, Department of Computer Science, where she heads the Copenhagen Natural Language Understanding research group as well as the Natural Language Processing section. She also co-heads the research team at CheckStep Ltd, a content moderation start-up. Her main research interests are fact checking, low-resource learning and explainability.

Mona Diab

Mona Diab is a researcher at Facebook AI and a professor at George Washington Unviersity. She conducts research in Statistical Natural Language Processing (NLP) is a rapidly growing, exciting field of research in artificial intelligence and computer science. Interdisciplinarity is inherent to NLP, drawing on the fields of computer algorithms, software engineering, statistics, machine learning, linguistics, pragmatics, information technology, etc. In NLP, researchers model language and its use, and build both analytical models and predictive ones. In Professor Diab’s NLP lab, they address problems in social media processing, building robust enabling technologies such as syntactic and semantic processing tools for written texts in different languages, information extraction tools for large data, multilingual processing, machine translation, and computational sociolinguistic processing. Professor Diab has a special interest in Arabic NLP, where the emphasis has been on investigating Arabic dialect processing where there are very few available automated resources.

Jimmy Lin

Jimmy Lin is a Professor and the David R. Cheriton Chair in the David R. Cheriton School of Computer Science at the University of Waterloo. Previously, he was at the University of Maryland. He completed my Ph.D. in Electrical Engineering and Computer Science at MIT in 2004. From 2010-2012, He spent an extended sabbatical at Twitter working on services designed to identify relevant content to users (search, recommendation, etc.) and analytics infrastructure to support data science (Hadoop tools, machine learning libraries, etc.). He has also worked for Cloudera. His research aims to build tools that help users make sense of large amounts of data. He work at the intersection of information retrieval, natural language processing, and databases, with a focus on large-scale distributed algorithms and infrastructure for data analytics.

Sebastian Ruder

Sebastian Ruder is a research scientist in the Language team at DeepMind, London. He completed his PhD in Natural Language Processing and Deep Learning at the Insight Research Centre for Data Analytics, while working as a research scientist at Dublin-based text analytics startup AYLIEN. Previously, he studied Computational Linguistics at the University of Heidelberg, Germany and at Trinity College, Dublin. He is interested in transfer and multilingual learning and making ML and NLP more accessible.

Program Chairs Comments on Ethics Review Process

2021-06-03T00:00:00+00:00

Slides (PDF): Takeaways from Ethics Review by Program Co-Chairs

The process followed for the NAACL 2021 ethics review was outlined by the Ethics Chairs in an earlier post. Comments from the ethics reviewers and track chairs were reviewed by the Program Chairs when making final acceptance decisions on the papers. As a result, 7 papers were accepted conditionally, which means that there were certain conditions imposed on the paper in the form of questions or requirements to be fulfilled for the paper to be accepted to the conference. For the majority of conditionally accepted papers, these included clarifications of the process the studies underwent for human subjects research, institutional review board protocols established for the studies, and discussion of whether and how the data collection procedures were coordinated with and complied with the rules outlined in terms of service for the websites and organizations that owned the data. As a result, 6 conditionally accepted papers were accepted into the conference, and one paper was rejected.

Our takeaway from this process is as follows: While we think the ethics review process is highly valuable, we found it very difficult to execute well in its current form. We believe that ethics review should be done as much as possible prior to paper submission, as described below. A conference program committee is not the best group to ensure that the reported work complies with ethical standards.

The reasons are two-fold. First, making last-minute ethical decisions about specific studies so late in the game, when the study is completed and written up, is less than ideal. A much better way to do it would be when the study is designed and considered, and this is normally precisely when the investigators set up human subjects research protocols with their institutional review boards (IRBs). Setting up a study protocol approved by an IRB may take up to several months and often involves back-and-forth, with clarifications from the authors, suggestions from the IRB to reframe the research or correct the protocol appropriately. This is simply impossible (and also too late) to do in the context of ethics review in the timeframe of a conference. Further, IRBs are typically composed of diverse individuals with complementary expertise from different fields, and as such, already are designed for such a review process. This is why we believe the burden of ethics review should be shifted back to this existing and well thought-out mechanism of ensuring that the research is conducted ethically.

No doubt, IRBs have to be educated about some of the subtleties of AI work and its implications for society at large. This should be done by all of us who work in the NLP field and understand these issues.

As far as compliance of the authors with the terms of service and other considerations related to data copyright and use ethics, we felt that we lacked the expertise and bandwidth to address such questions properly, and as a result, we believe that this might be much better handled by legal departments in the authors’ institutions.

To sum up, because of the difficulties involved in making ethical judgments, we felt that many of the questions we were trying to address during the ethics review process were better left to the properly composed multi-disciplinary IRBs and, as far as terms of service compliance is concerned, institutional legal departments. If an author reports that a human subjects study protocol and compliance with service terms is approved by their institution, that should provide sufficient guarantees for the ethics review process done by the conference.

This, of course, raises the question of what to do for unaffiliated authors or in cases when an author’s institution does not have an IRB and/or a legal department that can help address such issues. This is a difficult question to resolve, and perhaps one of the solutions might be to establish a standing review board associated with the ACL which can review such cases at author’s request.

Paper Review Process

2021-06-03T00:00:00+00:00

For the review process at NAACL-HLT 2021, we largely adopted the best practices of other recent virtual conferences, while modifying the format somewhat to treat all papers equally, as described in more detail below. Our paper review process followed the recent trend of a hierarchical organization, with senior area chairs (SACs) that organized coherent research tracks and area chairs (ACs) who shepherded smaller batches of papers within each track.

NAACL-HLT 2021 received 1797 submissions–a record for our conference! We accepted 477 papers, including 350 long and 127 short, for an overall acceptance rate of 26%. The acceptance rate for long papers was higher than short papers (28% vs. 23%), although this gap was smaller than in other recent conferences at least in part due to minor but explicit rebalancing done the the PC chairs. The following graphs show how these numbers break down in comparison to previous years.

From the accepted papers, and based on the nominations from SACs and review by the best paper committee, we selected the best papers in the long and short paper categories, as well as a small number of outstanding papers in each category. The best paper selection process is described here. NAACL-HLT 2021 will also feature 18 papers that were published at Transactions of the Association for Computational Linguistics(TACL) and 4 papers from the journal of Computational Linguistics(CL).

When designing the call for papers, we made an effort to balance the size of the tracks. Recent NLP conferences have had many tracks that received well over 200 submissions, making them mini-conferences of their own. To make the senior area chairs’ jobs more manageable, we split the machine learning track by areas (into “Classification and Structured Prediction Models” and “Language Modeling and Sequence to Sequence Models”) and separated Machine Translation from Multilingual. We also wrote a brief guide to authors, including descriptions of areas, to help them decide which track was most appropriate for their work. Overall, we felt that this did make the organization more manageable and that other tracks, including NLP applications, could possibly benefit from further splitting in future conferences. Otherwise, we followed recent traditions in track selection, including keeping some of the smaller, more recent additions (e.g. Green NLP and Ethics). The following graph shows a list of the paper submission and acceptance numbers per track.

We also had a special theme for the conference, which we called “New Challenges in NLP: Tasks, Methods, Positions.” This theme was selected to recognize that we have made significant progress in NLP over the last five years, and that the community could benefit from thinking about the new problems and upcoming challenges we should focus on next. Despite the general applicability of the unsupervised pre-training/fine-tuning paradigm, many problems are still very challenging for current models. At the same time, given the recent progress, there are likely broad new classes of problems that can now be studied for the first time. Hence, the special theme targeted papers focusing on: “What tasks or capabilities should we focus on next?” and “What new classes of models should we be investigating?” We envisioned papers falling into this theme including (but not limited to) (1) empirical and dataset papers that propose new challenges that bring us closer to human-level language understanding and generation, and (2) position papers framing an important direction or highlighting an understudied research problem.

We recruited reviewers through a centralized process, designed to minimize workload for senior area chairs (SACs) without sacrificing review quality. We collected a list of likely qualified reviewers based on the reviewer and author pool of other recent NLP conferences. These candidates were invited to sign up to review, and were required to fill out a profile that allowed us to better assess their potential area fit and experience levels. We were fortunate enough to have enough volunteers to not need everyone, and were able to bias the final selection towards more senior reviewers, although many junior and first time reviewers remained in the pool. The final program committee included 54 SACs, 267 ACs, and 1941 committee members. We greatly appreciate the incredible amount of work they all did, and also thank all of the volunteers who were not selected in the end.

There was one part of the review process that we did this year, which we would likely not repeat in hindsight. We coordinated the review process so that authors could see their reviews and withdraw their paper to resubmit to ACL 2021 if they were not happy with the quality. Hundreds of papers were withdrawn, and anecdotal evidence suggests that many of them would have been accepted if they had been left in the pool. We hope that initiatives like the ACL Rolling Review will provide more robust solutions to these types of problems in the future, and otherwise suspect overlapping review periods will just become the norm as conferences continue to grow.

Finally, for reference, here is a full table with all the per-track paper acceptance statistics.

Area\Number of	Long Paper Submissions	Long Papers Accepted	Short Paper Submissions	Short Papers Accepted	All Papers Submissions	All Papers Accepted
Computational Social Science and Social Media	44	11	17	3	61	14
Dialogue and Interactive Systems	90	30	27	5	117	35
Discourse and Pragmatics	19	7	14	3	33	10
Ethics, Bias, and Fairness	15	4	19	6	34	10
Green NLP	8	5	4	3	12	8
Language Generation	73	20	18	1	91	21
Information Extraction	95	15	27	1	122	16
Information Retrieval and Text Mining	38	7	22	7	60	14
Interpretability and Analysis of Models for NLP	70	16	40	12	110	28
Language Grounding to Vision, Robotics and Beyond	45	13	17	8	62	21
Language Resources and Evaluation	53	26	20	2	73	28
Linguistic Theories, Cognitive Modeling and Psycholinguistics	13	3	6	2	19	5
Machine Learning for NLP: Classification and Structured Prediction Models	84	21	41	8	125	29
Machine Learning for NLP: Language Modeling and Sequence to Sequence Models	58	11	27	8	85	19
Machine Translation	60	19	35	9	95	28
Multilinguality	34	10	18	4	52	14
NLP Applications	109	35	53	9	162	44
Phonology, Morphology and Word Segmentation	10	5	6	2	16	7
Question Answering	68	20	31	7	99	27
Semantics: Lexical Semantics	29	9	10	1	39	10
Semantics: Sentence-level Semantics and Textual Inference	66	20	24	8	90	28
Sentiment Analysis and Stylistic Analysis	56	15	16	2	72	17
Speech	10	3	11	2	21	5
Summarization	53	22	22	5	75	27
Syntax: Tagging, Chunking, and Parsing	20	7	9	2	29	9
Special Theme	34	12	10	3	44	15
Total	1254	366	544	123	1798	489

Startups in NLP: Meet the Panel

2021-06-03T00:00:00+00:00

Name: Startups in NLP, a panel in the Industry Track of NAACL HLT 2021

Time: Monday June 7 at 1pm PDT, 4pm EDT.

Duration: 60 minutes

Questions will be taken from the audience

Moderator

Alon Lavie

Alon Lavie is the VP of Language Technologies at Unbabel, a leading translation technology and services company, and a Consulting (adjunct) Professor at the Language Technologies Institute at Carnegie Mellon University. At Unbabel, Alon leads and manages the company’s US AI lab based in Pittsburgh and provides strategic AI leadership company-wide. Prior to joining Unbabel, Alon was a senior manager at Amazon, where he led and managed the Amazon Machine Translation R&D group in Pittsburgh. Prior to that, Alon co-founded “Safaba Translation Solutions”, a Machine Translation technology start-up company, where he was Chairman of the Board, President and CTO. Safaba was acquired by Amazon in 2015. From 1995 to 2015, Alon was a Research Professor at the Language Technologies Institute at Carnegie Mellon University. His main research interests and activities focus on Machine Translation evaluation (the METEOR and COMET metrics), translation Quality Estimation, Machine Translation adaptation approaches with and without human feedback, and methods for multi-engine MT system combination. Alon served as the President of the International Association for Machine Translation (IAMT) (2013-2015). Prior to that, he was president of the Association for Machine Translation in the Americas (AMTA) (2008-2012), and was General Chair of the AMTA 2010 and 2012 conferences. He is a member of the Association for Computational Linguistics (ACL), where he was president of SIGParse - ACL’s special interest group on parsing (2008-2013).

Panelists

Apoorv Agarwal

Apoorv Agarwal is the Chief Executive Officer and founder of Text IQ, a leading AI company whose products are used by some of the world’s largest enterprises and organizations to identify sensitive information hiding in their unstructured data. Text IQ’s AI is used by its customers to identify unconscious bias within their organization, protect privileged and confidential information, and elevate their privacy programs.

In May this year, Apoorv secured a successful exit by selling Text IQ to Relativity, a $3.6 billion unicorn, which is the legal and compliance platform of choice for the vast majority of Fortune 500 companies and top law firms. By joining hands with Relativity, Apoorv and his team at Text IQ intend to realize their shared vision of bringing the benefits of AI to the world’s largest companies so they can finally discover the truth in their data.

With a natural curiosity towards finding ways to use AI to solve organizational and business problems, Apoorv has dedicated a significant part of his early business and academic career to research and development. During his PhD in Computer Science at Columbia University, he developed the socio-linguistic hypergraph (SLH), which eventually became the core technology behind Text IQ’s products—the same technology that is now being used by some of the world’s most prominent and heavily regulated companies, including 5 out of the top 10 life sciences companies and 4 out of the top 10 banks.

Before Text IQ, Apoorv was a member of the IBM Watson core team, where he developed two patents and received the prestigious IBM Ph.D. Fellowship award.

Spence Green

Spence Green is the CEO of Lilt, which he co-founded with John DeNero. He graduated from Stanford University with a PhD in computer science under the direction of Chris Manning and Jeff Heer. His work is on machine translation and mixed-initiative systems.

Nasrin Mostafazadeh

Nasrin Mostafazadeh is Co-founder of Verneek, a new AI startup that is striving to enable anyone to make data-informed decisions without needing to have any technical background. Before Verneek, Nasrin held senior research positions at AI startups BenevolentAI and Elemental Cognition and was earlier at Microsoft Research and Google. She received her PhD at the University of Rochester working at the conversational interaction and dialogue research group, with her PhD work focused on commonsense reasoning through the lens of story understanding. She has started lines of research that push AI toward deeper understanding and showing basic common sense, with applications ranging from storytelling to vision & language. She has been a keynote speaker, chair, organizer, and program committee member at different AI venues. Nasrin was named to Forbes’ 30 Under 30 in Science 2019 for her work in AI.

Kieran Snyder

Kieran Snyder is the CEO and Co-Founder of Textio, the augmented writing platform. For anything you write, Textio tells you ahead of time who’s going to respond based on the language you’ve used. Textio’s augmented writing engine is designed to attach to any large text corpus with outcomes to find the patterns that work. Prior to founding Textio, Kieran held product leadership roles at Microsoft and Amazon. Kieran has a PhD in linguistics from the University of Pennsylvania. Her work has appeared in Fortune, Re/code, Slate, and the Washington Post.

Best Paper Awards

2021-06-02T00:00:00+00:00

Congratulations to the recipients of the best paper awards at NAACL 2021! We awarded best long paper, best short paper, and best thematic paper awards, as well as outstanding paper awards for both short and long papers. These papers will be presented in a plenary session from 11:40-1:10 PDT on Wed June 9th. Please see the conference schedule for more details, and hope to see you there!

Best Long Paper

Video-aided Unsupervised Grammar Induction

Songyang Zhang, Linfeng Song, Lifeng Jin, Kun Xu, Dong Yu and Jiebo Luo

Outstanding Long Papers

Unifying Cross-Lingual Semantic Role Labeling with Heterogeneous Linguistic Resources

Simone Conia, Andrea Bacciu and Roberto Navigli

It's Not Just Size That Matters: Small Language Models Are Also Few-Shot Learners

Timo Schick and Hinrich Schütze

Best Short Paper

Learning How to Ask: Querying LMs with Mixtures of Soft Prompts

Guanghui Qin and Jason Eisner

Outstanding Short Paper

How many data points is a prompt worth?

Teven Le Scao and Alexander Rush

Best Thematic Paper

Preregistering NLP research

Emiel van Miltenburg, Chris van der Lee and Emiel Krahmer

Best Paper Selection Process

These papers have been selected from the pool of 1797 total submissions to NAACL 2021. Candidates were nominated by Senior Area Chairs and reviewed by the Best Paper Committee. Each committee member ranked a subset of the papers and the PC co-chairs considered all of the recommendations to make the final decisions. Special thanks to our distinguished Best Paper Committee: Isabelle Augenstein, Marco Baroni, Barbara Di Eugenio, Jacob Eisenstein, Hanna Hajishirzi, Omer Levy, Jessy Li, Yang Liu, Chris Quirk, and Bonnie Webber.

NAACL Mentoring Sessions

2021-06-02T00:00:00+00:00

Goals: The NAACL 2021 Diversity & Inclusion committee has taken on the responsibility for organizing mentoring at the conference. Our goal is that people with less experience can get advice from community members with more experience. The focus is mainly on graduate and undergraduate students, as well as early career researchers.

What are the Mentoring Sessions? We have organized two types of mentoring sessions. One type is the matched mentoring sessions where the mentors and mentees are matched according to their interests and topics. In addition, we are organizing four open mentoring sessions in order to make mentoring accessible to everyone, including those who were unable to sign up on time. These sessions will be open to all conference attendees and have no set topics. Multiple mentors will be assigned to these sessions to enable effective discussion on various topics. The group mentoring sessions are organized to span different time zones to offer students and early-career researchers a chance to meet and get advice from senior researchers on various topics.

How to join the mentoring sessions? Instructions for joining the sessions will be posted on the virtual conference site.

What is the format of the mentoring sessions? The matched mentoring sessions are structured as 1-hr virtual roundtable discussion/Q&A sessions which will be held on Zoom and deliberately kept small to maximize interaction, with at most 15 members in each group. These sessions require pre-registration and are accessible only to those who signed up.

When are the mentoring sessions scheduled for? The matched mentoring sessions are scheduled according to the preferences of the mentors and mentees who signed up for this resource.

The open sessions are currently planned on:

Monday, Jun 7, 2 pm PDT
Tuesday, Jun 8, 7 am PDT
Tuesday, Jun 8, 2 pm PDT
Wednesday, Jun 9, 2 pm PDT

What are the potential topics of discussion?

Some potential topics of discussion include:

Diversity and Inclusion in NLP
Finding a great mentor/advisor
“Managing up” (working effectively with your advisor/supervisor)
Mental Health
Surviving graduate school
Work-life balance
Learning to draw boundaries/how to say no

Various career related questions:

Long-term career planning
Changing career direction to NLP
Choosing between academia and industry
Building your professional identity
Whether to do a PhD, and how to apply for and choose a PhD program
Getting a faculty position

As well as research advice like:

Navigating research problems/directions in NLP
Establishing collaborations and networking
Effective communication in research
Becoming a research leader

What are the expectations of the mentor? (or How to make the mentoring most effective?)

Mentors are expected to:

Manage floor time: make sure each participant gets a turn to speak and interact
Make the session engaging: Set the right tone of the discussion that encourages mentees to speak up and share their questions and experiences.
Keep discussion focused: While it is ok to discuss a broad range of topics, a themed discussion may be more effective.
Set expectations: are mentees allowed to follow up with you? What types of contact are appropriate?

For open mentoring sessions where there are multiple mentors present: We suggest that the mentors for the open mentoring sessions coordinate amongst themselves about how they would like to conduct their sessions. For example, they could decide if they would like to introduce certain topics and make opening remarks before opening up the floor to questions. They could also choose to present something to the audience, have a panel-style discussion, share experiences, or generally just go with the flow.

What are the expectations of the mentee (or How to make the most of your time with the mentor?):

Set goals: What are your goals and outcomes from this mentoring session? Come prepared with a list of questions that you’d like the mentor to address.
Be specific: Asking for specific feedback is more likely to result in advice that you can act or work on. This includes identifying action items that you could work on in the short-term as well as long-term.
After the mentoring session - check with the mentor if they are ok with follow up via email or other channels.

Code of Conduct: All participants are expected to abide by code of conduct rules which would entail pledging to make participation a harassment-free experience for all, regardless of age, body size, visible or invisible disability, ethnicity, gender identity and expression, level of experience, education, socio-economic status, nationality, personal appearance, race, religion or sexual identity and orientation. Instances of unacceptable behavior should be reported to PCC or by emailing the D&I committee. All complaints will be promptly and fairly investigated.

How did you (D&I Committee) prepare this list of recommendations?

We use the following as our references:

Emelo, R. (2011), ‘‘Group mentoring: rapid multiplication of learning’’, Industrial and Commercial Training, Vol. 43 No. 3, pp. 136-45.

Bynum, Y. P. (2015). The power of informal mentoring. Education, 136(1), 69-73.

Conference Structure

2021-06-01T00:00:00+00:00

The NAACL 2021 conference is fast approaching, and we wanted to give you a quick overview of the conference format and schedule.

NAACL 2021 will be held as a completely virtual conference, in response to the restrictions imposed by the COVID-19 pandemic. We are using the Underline platform to host the conference events online, including pre-recorded videos, interactive presentations and live poster sessions. A walk-through video for navigating NAACL 2021 using the Underline is available here.

There will be the total of 499 presentations at the main conference, which includes:

350 long papers
127 short papers
18 papers published in the Transactions of the Association for Computational Linguistics (TACL), and
4 papers from the journal of Computational Linguistics (CL).

There are also 39 industry track papers, 17 system demonstrations, and 22 Student Research Workshop (SRW) papers, which will be presented in sessions parallel with the main track.

Following recent successful online events, our virtual format for the main conference includes both interactive talk sessions and posters in Gather.Town. A 12-minute video will also be available for long papers and a 7-minute video for short papers, to be viewed at any time during or after the conference.

This year, for the first time, there will be no distinction between oral and poster papers—all accepted papers will be presented in the interactive format.

The papers will be grouped into 80-minute interactive sessions by subject area, with 5–6 papers per session. Each session will consist of two segments: The first segment (approximately 50 minutes) will be held in a Zoom conference room and will consist of a brief oral pitch for each paper (5 minutes for long papers, 3 minutes for short papers), followed by a live Q&A (3 minutes per paper).

The second segment of each session will take place in Gather.Town and will have poster presentations for each paper.

The goal of this format is to allow the audience to engage into an in-depth discussion with the authors if desired. This is a new model that likely will not scale to much larger conferences, but will hopefully allow for rich interactions at the conference. This format only requires each author to be present for a single session.

For the oral pitches for in the first segments, we advise authors to think of these as brief summaries of each paper. The authors can use any format they prefer; they may want to use a reduced set of slides, play a part of their video, or use their poster as a prop. The idea was to have a “reminder” pitch before the Q&A for each paper. Then the poster segment of each session can be used by the folks interested in each paper for longer discussions with the authors.

The industry track and SRW sessions will follow the same format, while system demonstrations have pre-recorded videos, posters, and longer (80 minutes) Gather.Town sessions where the presenters could also demonstrate their systems interactively.

In addition to paper sessions, there will be keynote presentations (recorded or live), followed by live Q&A, best paper awards session, a business meeting, opening and closing sessions, as well as mentoring sessions, social events and sponsor booth events.

The conference will have a morning and an evening session block to account for the difference across time zones, with one keynote talk and 3 paper sessions per block, for a total of 6 paper sessions per day. The blocks are timed as follows:

The morning block will run from 8 am Pacific Daylight (UTC-7) / 11 am Eastern Daylight (UTC-4) to 1 pm Pacific / 4 pm Eastern.
The evening block will run from 4 pm Pacific / 7 pm Eastern to 9 pm Pacific / 12 am Eastern.

These time blocks were chosen to maximize access to different portions of the program for people located in different time zones. The interactive sessions within each block were then arranged to be in the normal waking hours for the contact author of each paper as much as possible.

Finally, the schedule also includes a business meeting from 11:40–1pm PDT on June 8th and a special plenary session for the best papers from 11:40–1:20pm PDT on June 9th.

Here is a high-level program overview:

Outside of the main conference days, there are 6 tutorials on June 6, which will be presented in 4-hour-long live sessions, and 22 workshops on June 6, 10, and 11. Most workshops include both live video-calls and Gather.Town poster sessions.

The D&I committee also coordinated a full schedule of events, including 7 affinity groups socials, 5 D&I sessions, and several thematic events: Queer in AI and Mexican school mentorship sessions, 2 panels on getting into NLP research, and tours that will virtually explore Mexico City and its linguistic diversity. Further, the program includes 17 Birds-of-a-Feather sessions, 30 small group mentoring meetings, and 4 open large-group mentoring meetings.

Participants are also welcome to stop by the virtual Gather.Town lounge at any time during the conference and have the opportunity to run into colleagues and visit sponsor spaces. Areas designed for specific research topics will be included. Attendees can also network using the Whova conference app.