Lightnews — Scholar-powered news

Reposted by Dirk Hovy

UKP Lab @ukplab.bsky.social · 1d

🔗 𝗥𝗲𝗹𝗮𝘁𝗲𝗱 𝗿𝗲𝘀𝗼𝘂𝗿𝗰𝗲𝘀
𝗔𝗥𝗥 𝗗𝗮𝘁𝗮 𝗖𝗼𝗹𝗹𝗲𝗰𝘁𝗶𝗼𝗻: arr-data.aclweb.org

𝗗𝗮𝗴𝘀𝘁𝘂𝗵𝗹 𝗦𝗲𝗺𝗶𝗻𝗮𝗿 𝗼𝗻 𝗣𝗲𝗲𝗿 𝗥𝗲𝘃𝗶𝗲𝘄: www.dagstuhl.de/en/seminars/...

(5/5)

ACL Rolling Review Data Collection (ARR-DC)

Collecting and curating a large-scale dataset of peer reviews and associated metadata from the ACL community.

arr-data.aclweb.org

1 2

Reposted by Dirk Hovy

UKP Lab @ukplab.bsky.social · 1d

🚀 𝗟𝗮𝘁𝗲𝘀𝘁 𝗣𝗲𝗲𝗿 𝗥𝗲𝘃𝗶𝗲𝘄 𝗗𝗮𝘁𝗮𝘀𝗲𝘁 𝗥𝗲𝗹𝗲𝗮𝘀𝗲 𝗳𝗿𝗼𝗺 𝗔𝗥𝗥 𝟮𝟬𝟮𝟱!
tudatalib.ulb.tu-darmstadt.de/handle/tudat...

📊 𝗡𝗲𝘄𝗹𝘆 𝗮𝗱𝗱𝗲𝗱 𝗔𝗖𝗟 𝟮𝟬𝟮𝟱 𝗱𝗮𝘁𝗮:
✅ 𝟮𝗸 papers
✅ 𝟮𝗸 reviews
✅ 𝟴𝟰𝟵 meta-reviews
✅ 𝟭.𝟱𝗸 papers with rebuttals

(1/🧵)

Illustration of a woman wearing a graduation cap and a lab coat, holding a magnifying glass and examining a document. A mechanical parrot with gears and circuits is perched on her shoulder.

1 1 2

Reposted by Dirk Hovy

Flor Plaza @florplaza.bsky.social · 2d

📢 Are you interested in a PhD in #NLProc to study and improve how AI model emotions and social signals?

🚨Exciting news:🚨 I’m hiring a PhD candidate at LIACS,
@unileiden.bsky.social.

📍 Leiden, The Netherlands
📅 Deadline: 17 Nov 2025

👉 Position details and application link: tinyurl.com/5x5v6zsa

PhD Candidate in Emotionally and Socially Aware Natural Language Processing

The Faculty of Science and the Leiden Institute of Advanced Computer Science (LIACS) are looking for a:PhD Candidate in Emotionally and Socially Aware Natural Language Processing (1.0fte)Project descr...

tinyurl.com

8 9

Dirk Hovy @dirkhovy.bsky.social · 1d

One of the most important things said about AI/NLP/ML recently. Could not agree more

Manoel Horta Ribeiro @manoelhortaribeiro.bsky.social · 3d

Computer Science is no longer just about building systems or proving theorems--it's about observation and experiments.

In my latest blog post, I argue it’s time we had our own "Econometrics," a discipline devoted to empirical rigor.

doomscrollingbabel.manoel.xyz/p/the-missin...

5

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 9d

#MemoryModay #NLProc 'Universal Joy: A Data Set and Results for Classifying Emotions Across Languages' by Lamprinidis et al. (2021) explores how AI research affects our planet. Tech can be green too! #SustainableTech

Universal Joy A Data Set and Results for Classifying Emotions Across Languages

Sotiris Lamprinidis, Federico Bianchi, Daniel Hardt, Dirk Hovy. Proceedings of the Eleventh Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis. 2021.

aclanthology.org

2 3

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 6d

#TBT #NLProc Explore 'Wisdom of Instruction-Tuned LLM Crowds' by Plaza et al. LLM labels outperform single models in tasks & languages. But few-shot can't top zero-shot. Supervised models rule.

Wisdom of Instruction-Tuned Language Model Crowds. Exploring Model Label Variation

Flor Miriam Plaza-del-Arco, Debora Nozza, Dirk Hovy. Proceedings of the 3rd Workshop on Perspectivist Approaches to NLP (NLPerspectives) @ LREC-COLING 2024. 2024.

aclanthology.org

2 2

Reposted by Dirk Hovy

EMNLP @emnlpmeeting.bsky.social · 5d

🚨 Important reminder that the deadline for authors to register for #EMNLP2025 is TODAY, October 3rd! For everyone else, the early registration deadline is just a few days away (October 6th)! 🚨

1 1 3

Reposted by Dirk Hovy

Culture & Computation Lab @cucolab.bsky.social · 9d

Watch out for the next talk by @dirkhovy.bsky.social. It approaches the question of the human condition in a digitally enriched lifeworld from a new perspective, considering the role of „AI“ in particular.

Join us on site or via Webex, 9.10.2025, 16:00 CEST.

Infos: cucolab.uni.lu/blog/convers...

3 3

Reposted by Dirk Hovy

Tanise Ceron @taniseceron.bsky.social · 9d

📣 New Preprint!
Have you ever wondered what the political content in LLM's training data is? What are the political opinions expressed? What is the proportion of left- vs right-leaning documents in the pre- and post-training data? Do they correlate with the political biases reflected in models?

2 14 44

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 12d

MilaNLPers at CLiC-it 2025 presenting "Probing Feminist Representations: A Study of Bias in LLMs and Word Embeddings"

Check the paper at clic2025.unica.it/wp-content/u...

#NLProc #clicit25

4 17

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 16d

#MemoryModay #NLProc 'Benchmarking Post-Hoc Interpretability Approaches for Transformer-based Misogyny Detection' - Attanasio et al. Explores reliability of interpretability in hate speech detection.

Benchmarking Post-Hoc Interpretability Approaches for Transformer-based Misogyny Detection

Giuseppe Attanasio, Debora Nozza, Eliana Pastor, Dirk Hovy. Proceedings of NLP Power! The First Workshop on Efficient Benchmarking in NLP. 2022.

aclanthology.org

2 4

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 13d

#TBT #NLProc 'Classist Tools: Social Class Correlates with Performance in NLP' by Curry et al. (2024) explores AI's hidden energy problem, and how machine learning impacts environmental sustainability. Tech can be green! #CleanTech

arxiv.org

2 3

Reposted by Dirk Hovy

Anne Lauscher @a-lauscher.bsky.social · 15d

🚨 Are you looking for a PhD in #NLProc dealing with #LLMs?
🎉 Good news: I am hiring! 🎉
The position is part of the “Contested Climate Futures" project. 🌱🌍 You will focus on developing next-generation AI methods🤖 to analyze climate-related concepts in content—including texts, images, and videos.

1 14 22

Reposted by Dirk Hovy

Johannes B. Gruber @jbgruber.bsky.social · 17d

Just wanted to share this Google Scholar trick: I often have the problem that I want to find papers using certain computational methods, but specifically in my own field (for lit reviews).

You can do that by limiting the search to certain sources. My (imperfect) collection in the alt text.

"BERT" AND "multilingual" source:"Digital Journalism" OR source:"Digital Journalism" OR source:"International Journal of Press/Politics" OR source:"Journal of Communication" OR source:"New Media and Society" OR source:"Communication Methods and Measures" OR source:"Communication Research" OR source:"Journal of Computer-Mediated Communication" OR source:"Big Data and Society" OR source:"Political Communication" OR source:"Social Media and Society" OR source:"Computational Communication Research"

4 17

Reposted by Dirk Hovy

EMNLP @emnlpmeeting.bsky.social · Sep 2

#EMNLP2025 is offering Virtual Registration Subsidies for those who would otherwise be unable to attend.

Note that these are only available for participants who are NOT registering any paper.

To apply, please read the details here, and fill out the linked form: 2025.emnlp.org/calls/virtua...

Call for EMNLP 2025 Virtual Registration Subsidies

Official website for the 2025 Conference on Empirical Methods in Natural Language Processing

2025.emnlp.org

1 3 5

Dirk Hovy @dirkhovy.bsky.social · 19d

Hi @gavina.bsky.social. Sorry to hear about that. We are aware of the issues, and have put out some information that hopefully helps:
2025.emnlp.org/visa/

EMNLP 2025 Visa Requirements & Invitation Letters

EMNLP 2025 Visa

2025.emnlp.org

Reposted by Dirk Hovy

EMNLP @emnlpmeeting.bsky.social · 20d

To everyone in the #EMNLP2025 community, we are aware of the issue regarding delays in responses to visa requests, and are actively working to find a viable solution to the visa invitation letter issue.

1 2 2

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 20d

#TBT #NLProc 'Geolocation with Attention-Based Multitask Learning Models' by Tommaso Fornaciari, @dirkhovy.bsky.social (2019) reveals how online political talks can become one-sided. Breaking out of our bubbles! #SocialMedia

Geolocation with Attention-Based Multitask Learning Models

Tommaso Fornaciari, Dirk Hovy. Proceedings of the 5th Workshop on Noisy User-generated Text (W-NUT 2019). 2019.

aclanthology.org

2 4

Dirk Hovy @dirkhovy.bsky.social · 21d

Absolutely. LLMs just scale this problme up to a new degree. The work is an attempt to induce a bit of caution in the current enthusiastic adoption of these tools.

1 4

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 27d

#TBT #NLProc 'MilaNLP @ WASSA: Does BERT Feel Sad When You Cry?' by Fornaciari et al. (2021) indicates that emotion and empathy are not related tasks for prediction.

aclanthology.org

2 2

Reposted by Dirk Hovy

EMNLP @emnlpmeeting.bsky.social · 22d

🪶 #EMNLP2025 is currently seeking applications for Birds-of-a-Feather and Affinity Group Events, as additional networking opportunities for EMNLP participants. 🪶

Please see this page and submit your application before 26 September, 2025 11:59pm (AoE): 2025.emnlp.org/calls/bof/

Call for Birds-of-a-Feather and Affinity Group Events

Official website for the 2025 Conference on Empirical Methods in Natural Language Processing

2025.emnlp.org

1 3 4

Reposted by Dirk Hovy

MilaNLP Lab @milanlp.bsky.social · 23d

#MemoryModay #NLProc 'Dense Node Representation for Geolocation' by Fornaciari & @dirkhovy.bsky.social reveals efficient geolocation methods using node2vec & doc2vec models. Greater network size, less parameters. /publication/2019_m2v/2019_m2v

Dense Node Representation for Geolocation

Tommaso Fornaciari, Dirk Hovy. Proceedings of the 5th Workshop on Noisy User-generated Text (W-NUT 2019). 2019.

aclanthology.org

2 3

Reposted by Dirk Hovy

Johannes B. Gruber @jbgruber.bsky.social · 23d

If you feel uneasy using LLMs for data annotation, you are right (if not, you should). It offers new chances for research that is difficult with traditional #NLP/#textasdata methods, but the risk of false conclusions is high!

Experiment + *evidence-based* mitigation strategies in this preprint 👇

Joachim Baumann @joachimbaumann.bsky.social · 26d

🚨 New paper alert 🚨 Using LLMs as data annotators, you can produce any scientific result you want. We call this **LLM Hacking**.

Paper: arxiv.org/pdf/2509.08825

$We present our new preprint titled "Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation". We quantify LLM hacking risk through systematic replication of 37 diverse computational social science annotation tasks. For these tasks, we use a combined set of 2,361 realistic hypotheses that researchers might test using these annotations. Then, we collect 13 million LLM annotations across plausible LLM configurations. These annotations feed into 1.4 million regressions testing the hypotheses. For a hypothesis with no true effect (ground truth $p > 0.05$), different LLM configurations yield conflicting conclusions. Checkmarks indicate correct statistical conclusions matching ground truth; crosses indicate LLM hacking -- incorrect conclusions due to annotation errors. Across all experiments, LLM hacking occurs in 31-50\% of cases even with highly capable models. Since minor configuration changes can flip scientific conclusions, from correct to incorrect, LLM hacking can be exploited to present anything as statistically significant.$

1 4 22

Reposted by Dirk Hovy

Joachim Baumann @joachimbaumann.bsky.social · 26d

Why this matters: LLM hacking affects any field using AI for data analysis–not just computational social science!

Please check out our preprint, we'd be happy to receive your feedback!

#LLMHacking #SocialScience #ResearchIntegrity #Reproducibility #DataAnnotation #NLP #OpenScience #Statistics

1 1 8

Dirk Hovy @dirkhovy.bsky.social · 26d

Comprehensive preprint out now, check it out (feedback welcome!).

Joachim Baumann @joachimbaumann.bsky.social · 26d

🚨 New paper alert 🚨 Using LLMs as data annotators, you can produce any scientific result you want. We call this **LLM Hacking**.

Paper: arxiv.org/pdf/2509.08825

$We present our new preprint titled "Large Language Model Hacking: Quantifying the Hidden Risks of Using LLMs for Text Annotation". We quantify LLM hacking risk through systematic replication of 37 diverse computational social science annotation tasks. For these tasks, we use a combined set of 2,361 realistic hypotheses that researchers might test using these annotations. Then, we collect 13 million LLM annotations across plausible LLM configurations. These annotations feed into 1.4 million regressions testing the hypotheses. For a hypothesis with no true effect (ground truth $p > 0.05$), different LLM configurations yield conflicting conclusions. Checkmarks indicate correct statistical conclusions matching ground truth; crosses indicate LLM hacking -- incorrect conclusions due to annotation errors. Across all experiments, LLM hacking occurs in 31-50\% of cases even with highly capable models. Since minor configuration changes can flip scientific conclusions, from correct to incorrect, LLM hacking can be exploited to present anything as statistically significant.$

1 3