Workshop on
NLP for Positive Impact 2022

Recorded Videos 2022

• Opening Remark (Video)

• Introductory Talk - Zhijing Jin: "What Is NLP for Social Good?" (Video, Slides)

• Keynote Talk - Prof Preslav Nakov: "Fighting the Global Social Media Infodemic: from Fake News to Harmful Content" (Video)

• Keynote Talk - Dr. Mike Bailey (Meta): "The Role of Social Networks in Economic Mobility" (Video)

• Keynote Talk - Prof Milind Tambe (Harvard): "AI for social impact: Results from deployments for public health and conservation" (Video, Slides)

• Keynote Talk - Prof Rada Mihalcea (University of Michigan): "NLP4ALL: Recentering NLP Around ALL People" (Video, Slides)

• Keynote Talk - Prof Sam Bowman (NYU & Anthropic AI): "What's the deal with AI safety?" (Video, Slides)

• Panel - Prof Luis Chiruzzo, Tara Chklovski, Prof Dora Demszky, Prof Rada Mihalcea: "What is NLP for Social Good and how to achieve it? -- Taking NLPxEducation as an example" (Video)

• Closing Remark (Video)

Part 1: Physical + Hybrid

Physical Location: Capital Suite 21A (see this room map), Abu Dhabi National Exhibition Centre (ADNEC)

Virtual Location: Check out the zoom link on our Underline event

Time zone: Abu Dhabi time

Tip 💡: You can directly import our calendar here, or use the Google Calendar snippet below.

• 1pm Abu Dhabi time: 

  • Opening Remark (slides)

  • Introductory Talk: "What Is NLP for Social Good?" (slides)

• 2pm - 3pm: Invited Talk by Preslav Nakov

• 3pm - 4pm: Lightning Talks (To fit for the form, perhaps most speakers will be physical; a hybrid audience is welcome, as we will livestream the lightning talks)

  • Topics: What you are doing for NLP for Social Good? Or anything you want to discuss.

  • Default: 2-min lightning talk (welcome to come to the stage impromptu)

  • If you have lots of materials: feel free to take 5 min

  • Links: Julia's Talk

• 4pm - 5pm: Paper Presentation Session 1

  • We will first play the 1-min Teaser for each paper of Session 1

  • Then, move to:

  • Poster Presentations:

    • Physical Poster Room = Mezzanine and Hall (see the room map)

    • Virtual = Gather Town (link)

  • Papers in this session:

    • Securely Capturing People's Interactions with Voice Assistants at Home: A Bespoke Tool for Ethical Data Collection

    • A Dataset of Sustainable Diet Arguments on Twitter

    • Towards Countering Essentialism through Social Bias Reasoning

    • ClimaBench: A Benchmark Dataset For Climate Change Text Understanding in English

    • Enhancing Crisis-Related Tweet Classification with Entity-Masked Language Modeling and Multi-Task Learning

    • Misinformation Detection in the Wild: News Source Classification as a Proxy for Non-article Texts

    • Modelling Persuasion through Misuse of Rhetorical Appeals

    • Breaking through Inequality of Information Acquisition among Social Classes: A Modest Effort on Measuring "Fun"

    • Identifying Condescending Language: A Tale of Two Distinct Phenomena?

    • Hate-CLIPper: Multimodal Hateful Meme Classification based on Cross-modal Interaction of CLIP Features

    • Participatory Translations of Oshiwambo: Towards Culture Preservation with Language Technology

    • (Findings) Beyond Model Interpretability: On the Faithfulness and Adversarial Robustness of Contrastive Textual Explanations

    • (Findings) HARALD: Augmenting Hate Speech Data Sets with Real Data

    • (Findings) Challenges and Opportunities in Information Manipulation Detection: An Examination of Wartime Russian Media

    • (Findings) Gender Bias in Meta-Embeddings

    • (Findings) Fair NLP Models with Differentially Private Text Encoders

    • (Findings) Mitigating Covertly Unsafe Text within Natural Language Systems

    • (Findings) Logical Fallacy Detection

Part 2: Virtual Only

Virtual Location: Check out the zoom link on our Underline event. No physical rooms will be set up.

Time zone: Eastern Time (ET)

• 8:30am - 11am ET: Invited Talks:

  • 8:30-9am: Talk by Mike Bailey (Meta): "The Role of Social Networks in Economic Mobility"

  • 9-9:30am: Talk by Milind Tambe (Harvard): "AI for social impact: Results from deployments for public health and conservation" (slides)

  • 9:30-9:45am: Live Q&A by Mike & Milind

  • 9:45-10:15am: Talk by Rada Mihalcea (University of Michigan): "NLP4ALL: Recentering NLP Around ALL People" (slides)

  • 10:15-10:45am: Talk by Sam Bowman (NYU & Anthropic AI): "What's the deal with AI safety?" (slides)

  • 10:45-11am: Live Q&A by Rada & Sam

• 11-11:15am: Break

• 11:15-12:00: Panel: "What is NLP for Social Good and how to achieve it? -- Taking NLPxEducation as an example"

  • Prof Luis Chiruzzo, Professor at Universidad de la República, Uruguay (work on NLP tools to support teachers in rural areas)

  • Tara Chklovski, CEO of Technovation (actively involved in educating children in underserved communities)

  • Prof Dora Demszky, Assistant Professor in Education Data Science at Stanford (researching on NLP and Education)

  • Prof Rada Mihalcea, Professor at the University of Michigan (NLP for social good, with applications in mental health, fake news, etc.)

• 12 noon - 1:30pm: Interaction Session & Paper Presentation Session 2

  • 12:00-12:10: 1-min Teaser for each paper of Session 2 

(Now, instead of directly scattering out to Gather Town, we first finish all the "main stage" activities, and then leave an open-end for gather town at the end)

• • 12:10-12:20: (To encourage online participation) Zoom: Break-Out Rooms of 5 people/room -- to level up the participation for virtual conferences:

    • 5-min Break-Out Rooms for ice-breaking: Introduce yourself (who you are and what's our interest in NLP4SG). If you are the author of the papers, you can also talk more about the stories behind your paper.

    • 5-min Break-Out Rooms to gather ideas for "novel new tasks/frameworks for NLP4SG"

  • 12:20-12:30: Announce the Best Papers & Outstanding Papers

Best Paper 🎉: 

• • • Enhancing Crisis-Related Tweet Classification with Entity-Masked Language Modeling and Multi-Task Learning (Philipp Seeberger and Korbinian Riedhammer)

Orals (A-Z):

• • • A Dataset of Sustainable Diet Arguments on Twitter (Marcus Astrup Hansen  and Daniel Hershcovich)

    • ClimaBench: A Benchmark Dataset For Climate Change Text Understanding in English (Tanmay Laud, Daniel Spokoyny, Tom Corringham and Taylor Berg-kirkpatrick)

    • Impacts of Low Socio-economic Status on Educational Outcomes: A Narrative Based Analysis (Motti Kelbessa, Ilyas Jamil and Labiba Jahan)

    • Modelling Persuasion through Misuse of Rhetorical Appeals (Amalie Brogaard Pauli, Leon Derczynski and Ira Assent)

    • Participatory Translations of Oshiwambo: Towards Culture Preservation with Language Technology (Wilhelmina Nekoto, Julia Kreutzer, Jenalea Rajab, Millicent Ochieng and Jade Abbott)

    • Towards Countering Essentialism through Social Bias Reasoning (Emily Allaway, Nina Taneja, Sarah-jane Leslie and Maarten Sap)

  • Then, move to:

  • 12:30-13:30: Gather Town (link): Posters of all papers of Session 2

Papers in this session:

• • • A unified framework for cross-domain and cross-task learning of mental health conditions

    • Critical Perspectives: A Benchmark Revealing Pitfalls in PerspectiveAPI

    • Impacts of Low Socio-economic Status on Educational Outcomes: A Narrative Based Analysis

    • Using NLP to Support English Teaching in Rural Schools

    • "Am I Answering My Job Interview Questions Right?'': A NLP Approach to Predict Degree of Explanation in Job Interview Responses

    • Generate Me a Bedtime Story: Leveraging Natural Language Processing for Early Vocabulary Enhancement

    • BELA: Bot for English Language Acquisition

    • Transformers-Based Approach for a Sustainability Term-Based Sentiment Analysis (STBSA)

    • Understanding COVID-19 Vaccine Campaign on Facebook using Minimal Supervision

    • (Findings) Conditional Contrastive Learning for Fair Text Classification

    • (Findings) Handling and Presenting Harmful Text

    • (Findings) Don't Just Clean It, Proxy Clean It: Mitigating Bias by Proxy in Pre-Trained Models

  • [End of the Workshop]

Design Principle of the Events

Our workshop aims to foster interdisciplinary conversations:

• Keynote talks and panels that include both NLP researchers and non-NLP people working on social good

• Paper poster sessions to show a variety of NLP and social good research

• Supply vs. Need conversation: Invite people with data & application needs to propose tasks. And invite NLP researchers to take the tasks and do projects accordingly. 

  • E.g., a lightning talk session where each interested person give a 2-min pitch of their work/needs/resources.

Free-form discussions among the participants.