For last week's @milanlp.bsky.social seminar, we had the pleasure to host Hongyu Chen. She presented her upcoming EMNLP paper "Contrast Comes at a Cost: Prompting Large Language Models for Perception-Centered Gendered Style Transfer". #NLProc

🚨 [Token][ization] Paper Alert 🚨 Tokenization is a wildly understudied area of language modeling despite it having effects across all of NLP. Over the past ~8 months, 32 (!) tokenizer researchers put together the most comprehensive survey of the field. Check it out!

🚂 The early registration for #INLG2026 departs September 15th. Standard fares after that. Utrecht, Oct 17–21. 2026.inlgmeeting.org/registration...

Check out this batch of papers headed straight to Budapest! #EMNLP2026 #NLProc #NLP #ComputationalLinguistics

🚂 127 #INLG2026 papers are on track for Utrecht! A sample: "Justify your prompts!", "Self-Verification is All You Need To Pass The Japanese Bar Examination", and live commentary generation for Age of Empires II. Congrats to all authors. Utrecht, Oct 17–21. 2026.inlgmeeting.org/accepted-pap...

Nathan Lambert arguing for RLHF against the haters saying it's just style transfer -- youtu.be/o6l6tJQgUg4 on slide 50 around 35 minute mark

3/3 papers accepted at #EMNLP2026! Thanks to great co-authors. More soon...

Style is important quote n of ongoing [...] the most immediate problem to be solved in the attack on sociolinguistic structure is the quantification of the dimension of style. — William Labov (1972)

While I'm posting for #IC2S2, I want to share my new lab and website with you all! Our lab is called "Computational Analysis of Text and Society" (CATS) and I finally updated the website! Check it out: cats-group.github.io

🆕 Check out our lineup of workshops! Both research and position papers are welcome. 2026.inlgmeeting.org/workshops-tu...

Our latest research (led by Sadie Barlow, with Dr Edoardo Manino) found LambdaG doesn’t need a calibration dataset for well-calibrated likelihood ratios, just a simple correction! This simplifies analysis and makes it the only authorship verification method without calibration. Paper on arXiv now.

LambdaG: A simplified approach to Authorship Verification

In our latest research, led by Sadie Barlow and with Dr Edoardo Manino, we discovered a very important property of LambdaG: it does not require a calibration dataset to produce well calibrated likelihood ratios. A simple correction is sufficient. This means that the analysis is significantly simpler to implement and explain. LambdaG is therefore also the only existing authorship verification method that does not require a calibration analysis to produce likelihood ratios. The paper, currently under review, is available on arXiv here:

andreanini.com

❗️Direct submission deadline for #INLG2026 has been extended to July 18th (AoE). Be sure to proof-read your work, and good luck! #CfP #NLProc

The 2nd Call for Papers for #INLG2026 is out! Two ways to submit: 📄 Direct submissions, archival and not, by 15 July: openreview.net/group?id=acl... 💍 ARR commitments by 5 August: openreview.net/group?id=acl... ——— 📍Utrecht, NL — Oct 17–21, 2026, right before EMNLP 2026.inlgmeeting.org/calls.html

Our heartfelt congratulations to Eduardo Calò, who successfully defended his PhD thesis titled "Automatically Expressing the Meaning of Logical Formulae in Natural Language" on 17 June at the Academiegebouw.

The Call for Papers for #INLG2026 is out! 🗓️ Submit by July 15 (AoE) 💍 ARR commit by August 5 🆕 Squibs welcomed (raising an issue without needing to solve it) 🆕 Non-archival track for work in progress ——— 📍Utrecht, NL — Oct 17–21, 2026, right before EMNLP 2026.inlgmeeting.org/calls.html

New paper w/ UK AISI: Millions of people now use AI to help them write and communicate. In three experiments (14k participants, 3m+ human ratings) we show that AI writing assistance systematically distorts writer personas – their perceived beliefs, personality, and identity. 🧵

A dream come true. A conference in a train museum. What will the breaks be like? Choo Choo! 🚂🚂

INLG 2026@inlg.bsky.social · 6mo ago

🚂 Choo choo! Next stop: Utrecht Maliebaan Part train station, part railway museum, and the future home of #INLG2026. A 1874 station packed with 80+ historic trains, reached via its very own train from Utrecht Centraal. All aboard for October! Call for papers and more updates coming soon

Would you realize if the book you were reading was AI? What if it was humanized to remove AI-speak? We find that even without using stylistic cues (e.g., word choice or sentence structure) narrative choices alone give AI fiction away!

Patients ask LLMs medical questions — but how they phrase it matters more than it should. Our new preprint explores how different phrasings of patient health questions can lead to inconsistent conclusions, even with the same evidence. [1/6] Full Paper: arxiv.org/abs/2604.05051

📢 The workshop on Insights from negative results will be back at EMNLP'26! Your most-insightful failures can be submitted in 4 pages by June 25. It's also possible to commit short papers reviewed through ARR. insights-workshop.github.io/2026/cfp

2026 Call for Papers

Workshop on Insights from Negative Results in NLP

insights-workshop.github.io

Yay. Finally a conference next to my house! Hope to see many of the #NLProc #NLP community there 🤗

INLG 2026@inlg.bsky.social · 6mo ago

📍It's official! #INLG2026 is coming to Utrecht, Netherlands, Oct 17-21! Hosted with support from Utrecht University and the local NLP community. Follow us here and check 2026.inlgmeeting.org for updates -- hope to see you there!

Survey-style tests developed for humans may not predict how LLMs actually behave. Our #EACL2026 paper shows they can even be misleading when measuring racism and sexism! Check out the paper 👇🏼

Jana Jung@janajung.bsky.social · 7mo ago

Are you using survey-style questionnaires designed for humans to measure characteristics of LLMs? In our #EACL2026 paper, we evaluate both the reliability and validity of such tests and found that their scores do not reflect real-world model behavior. In fact, they can be deceptive! 🧵1/3

🥁🥁🥁 Newly out from us today in Science Advances: “Biased AI Writing Assistants Shift Users’ Attitudes on Societal Issues”. Large Language Models are providing users with autocomplete writing suggestions on many platforms. Could these suggestions shift users’ own attitudes? (spoiler: YES) (1/7)

Can LLMs figure out who you are from your anonymous posts? From a handful of comments, LLMs can infer where you live, what you do, and your interests; then search for you on the web. New 📄 w/ @SimonLermenAI, @joshua_swans, @AerniMichael, Nicholas Carlini, @florian_tramer 🧵

our paper on data mixing for LMs is out! while building Olmo 3, we saw gaps between data mixing literature and real practice 🐠choosing proxy size, # runs, sampling, regression, constraints.. 🐟data shifts during LM dev: can we reuse past experiments? Olmix tackles them all!

Ai2@ai2.bsky.social · 8mo ago

Data mixing – determining how much web text, code, math, etc., you need for LM development – is a first-order lever on model quality. Introducing Olmix: a framework for configuring mixing methods at the start of dev & efficiently updating as data changes throughout. 🧵

If you're attending #EMNLP2025, we'll be presenting virtually in Gather Session 1 on Nov 5 at 4pm PT. Come say hello! w/ the wonderful: @mellymeldubs.bsky.social Anna Preus, @mariaa.bsky.social Paper: arxiv.org/abs/2510.16713 Code/Data: github.com/darthbhyrava/wisp Dash: poetry.darthbhyrava.com

What if a single model could recognize an author's writing style no matter what language they wrote in? 🌍✍️ Our new #EMNLP2025 paper explores multilingual authorship representation, showing how training across 36 languages can sharpen stylistic signals and reduce topic bias. 👇🧵

New opinion paper out with Esther Ploeger (Aalborg University): We Need to Measure Data Diversity in NLP — Better and Broader at #EMNLP2025 (main) aclanthology.org/2025.emnlp-m...

We Need to Measure Data Diversity in NLP — Better and Broader

Dong Nguyen, Esther Ploeger. Proceedings of the 2025 Conference on Empirical Methods in Natural Language Processing. 2025.

aclanthology.org

Lot's of exciting work on linguistic style this year at #EMNLP2025 #EMNLP! Including work on machine-text detection, authorship representation and more 🧵 with anthology links below 📣 with an open call to everyone to add style work that's missing

I have a new blog post about the so-called “tokenizer-free” approach to language modeling and why it’s not tokenizer-free at all. I also talk about why people hate tokenizers so much!