To date, the most impactful change enabled by AI is making matplotlib's impossible possible.

👏 tl;dr ~ automatic quality estimation metrics for MT show *undue* preferences towards specific gender marks and words. Check the paper for a discussion on the implications and way forward! #nlp #nlproc

Our paper on gendered performance of speech models won the Social Impact Award #EMNLP2024 🎉 A huge shout-out to my co-authors @BeatriceSavoldi @DennisFucci, and @dirk_hovy :) If curious, here's the paper:

If you're attending #EMNLP2024, stop by some of @sardine_lab_it posters and talks! 👇

Excited to present xTower at #EMNLP2024! 🎉 Join us for our poster session on November 12 at 4:00 PM in Riverfront Hall (Session C). See you there! @emnlpmeeting

@BeatriceSavoldi @DennisFucci @dirk_hovy In "Metrics for What, Metrics for Whom: Assessing Actionability of Bias Evaluation Metrics in NLP," @pieterdelobelle, @debora_nozza, @ZeerakTalat, @sulin_blodgett, and I discuss requirements and implications of *Actionability* of bias metrics in NLP.

And yes, about the Lisbon bit. @sardine_lab_it and @andre_t_martins are hiring postdocs and PhD students, so take the chance to chat with any of us 🐟

In "Twists, Humps, and Pebbles: Multilingual Speech Recognition Models Exhibit Gender Performance Gaps," @BeatriceSavoldi, @DennisFucci, @dirk_hovy, and I show how speech recognition serves different gender groups differently and what to do about it. https://t.co/Z3kfKIzGrD

Hey, I'll be in Miami next week for #EMNLP2024 🐙 I'm presenting two papers (on actionability of bias metrics and gendered performance of speech recognition systems 👇) and keen to eat/drink/catch up/chat about visiting & working in Lisbon & ongoing work :) 🧵

🚀 Excited to share our work, “Analyzing Context Contributions in LLM-based Machine Translation” with @nunonmg and @andre_t_martins! We dive deep into how LLMs utilise context in MT. Come by our poster @emnlpmeeting for a deeper dive! Paper 👉 https://t.co/LZfFGw6z1w 🧵Thread ⬇️

How does one's religion affect her emotional responses? Many religions cultivate specific emotions, but do LLMs capture this phenomenon? Our #EMNLP2024 Findings paper explores how sacred emotions are represented in current LLMs!

Much of GenAI is trash, but I'm finding human-AI coding delightful. Sort of "this is what early computers must have felt like".

Apex of: addressing issues whose solution is not AI, with AI. https://t.co/T0DsVgyd2V

Super excited and grateful that our paper received the best paper award at #ACL2024 🎉 Huge thanks to my fantastic co-authors — @clara__meister, Thomas Hofmann, @vlachos_nlp, and @tpimentelms — the reviewers that recommended our paper, and the award committee #ACL2024NLP

Hey, be sure not to miss @CurriedAmanda 's presentation. You'll get 1) fantastic insights on an understudied issue in the field, 2) a cool discussion about ethics and AI (and, if you prompt her, about Swift's latest news, too).

I am so happy to share "Building Bridges: A Dataset for Evaluating Gender-Fair Machine Translation into German", joint work with @peppeatta and @anne_lauscher accepted in the findings of @aclmeeting. #acl2024 https://t.co/JHPhdwfFXP

A wonderful thing would be for Apple to share worldwide and on marketing ads that quantization and adapters -- the two inventions enabling their on-device LMs -- result from ideas/scientific validation from several academic labs.

While I'm sure many are looking forward to knowing more about Italia, here's a recommended read @igeniusai https://t.co/hdnrvofz8V

Here's how to release a "100% open source model," of which 0% of its training code, data, and architectural details are available while gating access to weights :)

Happy to share our #ACL2024 paper: "Causal Estimation of Memorisation Profiles" 🎉 Drawing from econometrics, we propose a principled and efficient method to estimate memorisation using only observational data! See 🧵 +@clara__meister, Thomas Hofmann, @vlachos_nlp, @tpimentelms

The lineup of speakers is completed with Laura Weidinger (@weidingerlaura), staff research scientist at Google DeepMind! 🚀 Check out our website for the final schedule: https://t.co/vGbH23bJu0