🎉🎉🎉🎉🎉🎉🎉🎉🎉🎉🎉🎉

Some good news that I'm excited to share today: I've received an NSF CAREER award! This award will support my group's research on robust, fair, and culturally aware commonsense reasoning in natural language. Feeling grateful to my students & the many people who have supported me!

🗣️🌏Submit your work on spoken language translation to IWSLT by April 29!
In addition to shared task papers, you can submit regular scientific papers. And we have a special highlights track to reflect on emerging trends, or share work published elsewhere with the IWSLT community

2. Proposed Solution:

Alexandra DeLucia, Kevin Duh, and I propose Anti-LM decoding objective with exp decay, motivated by the observation that LLMs are performing Bayesian inference under the hood (Xie et al., 2022).

arxiv.org/pdf/2311.08324…
Paper to appear at NAACL Findings 🙏

Very obscure question:

Can anyone confirm that the batched strided GEMM implementations in CUDA and MKL (cblas_sgemm_batch_strided) are in fact not equally powerful?

It seems the MKL version doesn't allow for strides that are smaller than full submatrix size while CUDA does?

📢 MASC-SLL Call for Papers is out! 📢
🎓 Are you a student passionate about Speech, Language, and Learning? 🗣️✨
🌟 Present your research at the Mid-Atlantic Student Colloquium on Speech, Language, and Learning (MASC-SLL) at Johns Hopkins Johns Hopkins Engineering on May 3rd! 🌟

ICYMI I'm looking for strong postdoc candidates in the space of human-AI interaction!

this year, i've been experimenting a lot with something i call 'mentorship in public.' instead of spending time and energy on 1-1 mentorship, i only invest in public forms of mentorship (e.g. public portfolio crits and q&a's).

Excited to have been a part of StarCoder2! We use Data Portraits to document the training data. One cool thing is the dataset contains lots of text about code - like pull requests, documentation, and papers⬇️

Great opportunity to read our 2021 survey on tokenization in the modern age ;)

'Between words and characters: A Brief History of Open-Vocabulary Modeling and Tokenization in NLP'
arxiv.org/abs/2112.10508

Heya, the Microsoft Translator team (more specifically me and Huda Khayrallah) is hiring for a senior researcher role.

Apply here!
jobs.careers.microsoft.com/global/en/job/…

From Armita R. Manafzadeh, a fundamental advance in our field — a true rarity in any discipline. In making this advance, she gives us an insight into the life of extinct dinosaurs of a sort that would have seemed impossible even a few years ago. Yale University Yale Peabody Museum Yale UniversityFaculty

This paper is the culmination of a research journey I embarked on nearly a decade ago – one which has changed the way I think about joints. It all started with grocery store chickens and protractors at UC Museum of Paleo, when I hoped to reconstruct locomotion from fossil bones.

For centuries, paleontologists have had to rely on intuition re whether reconstructed joint poses 'look right.' Today in Nature Communications: we created a way to formalize & test that intuition, & discovered that articular surface interactions distinguish locomotor poses in dinosaurs 🧵

Apply to work with us!

it really should not surprise me how I often I go looking for this tweet.

#EAMT2024 2nd CfP is out!📢
Check our tracks out: eamt2024.sheffield.ac.uk/conference-cal…

We also have open calls for:
- Tutorials
- Workshops
- Best Thesis

See: eamt2024.sheffield.ac.uk/conference-cal…

#NLProc #MachineTranslation

📢In an upcoming TACL paper, Sweta Agrawal evaluated a wide range of text simplification models by asking an important but overlooked question: how well do readers understand key elements of the original content after reading the simplified version?👇

📢 Exciting News! 📚
Our latest paper (w/Marine Carpuat) is out: 'Do Text Simplification Systems Preserve Meaning? A Human Evaluation via Reading Comprehension.' 📖

Presenting my poster on dataset documentation today at 5pm (#619)!

If you liked Jelani Nelson's keynote yesterday on sketching, come see how we use efficient datastructures to document datasets at crazy compression: 800 GB dataset ➡️ 27 GB 🎉🎉🎉