🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

We are planning to incorporate significant changes in the upcoming WMT 2021 metrics task. We want to share our plans with the community and give everyone the chance to request additional changes. We plan to change the following when compared the previous WMT metric tasks: (1/n)

Joint work with Melvin Johnson Aditya Siddhant Orhan Firat Graham Neubig

Our approach is to replace the standard (subword-segmented text) embeddings with "visual text" embeddings, computed from a rendered image of the raw text that is sliced into overlapping, horizontal windows, akin to ASR.

Happy to announce our most recent paper: Experts, Errors, and Context: A Large-Scale Study of Human Evaluation for Machine Translation arxiv.org/abs/2104.14478 - We re-evaluated the outputs from WMT20 using in-context MQM annotations from professional translators.

Why should you, as an NLP researcher, work on signed languages? Our upcoming #ACL2021NLP paper (w/ Amit Moryossef @jahochcam (((ل()(ل() 'yoav))))👾 Malihe Alikhani) is a call-to-action for the NLP community to include signed languages, and explains how to do so 🤟[1/7] arxiv.org/abs/2105.05222

One important thing I'd like everyone that is using NLP to know is that when someone releases a wonderful new model that supports 100 languages, that doesn't mean that it works on 100 languages.

Nice paper from @KocmiTom et al at Microsoft recommending the use of #COMET to evaluate MT over metrics like BLEU: arxiv.org/pdf/2107.10821… github.com/Unbabel/COMET

To jump on the hype of great new metric papers: Consider submitting to this year's WMT21 metric task: statmt.org/wmt21/metrics-…

Summer was for work! Now #ParaCrawl v9 corpora are done and again bigger than the previous ones!🤩 Extrinsic evaluation through MT almost finished and, according to old BLEU and new COMET, the quality of the MT output improves! 🥳 We will share corpora and more results soon!🕑

Check out this talk next week!! Eleftheria Briakou Elizabeth Salesky

This. Is. Amazing 🤩 congrats Shruti Rijhwani !!

We are thrilled to announce that Unbabel is #acquiring @Lingo24! 🎉 Learn how our combined forces will provide more consistent, high-quality #multilingual customer experiences across marketing and customer service channels. hubs.li/Q010XTsl0

#CX is changing rapidly, so our CEO Vasco Pedro shares his top 4 predictions for #CustomerExperience in 2022, on CustomerThink. 🎱 Empathy at scale, the merge of #customerservice and #marketing, and #ResponsibleAI are all going to be key this year.👇 customerthink.com/4-predictions-…

Working on a model but not sure if your random variables should be discrete or continuous? Try something in-between! Check out our new #ICLR2022 paper for the information theory of mixed variables. Bonus: Gaussian-sparsemax (the sparsemax counterpart of the logistic-normal)!

Happy to announce the WMT22 Metrics Shared Task: We are (again) planning to incorporate significant changes like (a) better human evaluation and (b) 3 new subtasks! (1/n)

Check out our #NAACL2022 paper where we put together MT and QE and compare MBR and reranking! Part of the MAIA and QUARTZ projects Unbabel CMU Portugal

Unbabel at #EAMT2022 presenting the little metric that could, highlighting the need for smaller, more efficient models

Day 3! A final day with many interesting talks, a keynote by Laura Rossi, and another poster session. Unbabel starting with a talk on MT in customer support and the impact of agent and user-generated content.

Our paper Cometinho just won the best paper award at EAMT2022! aclanthology.org/2022.eamt-1.9.…

You wanna learn how to prevent hallucinations? Then read the new work from Nuno M. Guerreiro Lena Voita and Andre Martins! You will be pleased 😉