🔥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

Code at GitHub: github.com/ymoslem/Adapti…

Now, my IWSLT paper is available on the ACL Anthology: Leveraging Synthetic Audio Data for End-to-End Low-Resource #Speech #Translation Congratulations to all the colleagues presenting at ACL 2024 or colocated workshops. Keep up the good work! 🎉 #NLProc aclanthology.org/2024.iwslt-1.3…

The Minitron models by NVIDIA are very impressive. nvda.ws/3YSUkwR I confirmed that they are also very robust to quantization. I'll publish all my results and the quantized models, including quantized versions of Llama 3.1 Minitron 4B, on Monday!

My feeling is that any linguist who works on language/translation technology should consider learning Python. It is not acceptable anymore to not be able to write an evaluation script or build a simple demo for one’s work.

Anna Rogers One good reference I highly recommend is the video by Prof Andrew Ng. Some articles summarised it nicely like this one. forecastegy.com/posts/read-mac…

I would like to do more research on speech. If anyone has an idea or a current project, I will be happy to help. Feel free to reach out.

🥳 SEACrowd Catalogue has been accepted to EMNLP 2024 main! Amazing collaborative work by 60+ co-authors This is also a great milestone for our SEA NLP community! Paper: arxiv.org/abs/2406.10118 Our catalog: seacrowd.github.io/seacrowd-catal…

Thanks to the authors, especially Holy Lovenia Samuel Cahyawijaya for leading the project

SEACrowd's publication has been accepted at #EMNLP2024! 🚀 "SEACrowd: A Multilingual Multimodal Data Hub and Benchmark Suite for Southeast Asian Languages" (arxiv.org/pdf/2406.10118) This is a major leap for AI research in SEA, and we owe it to our amazing community of 100+! 💪

Hi Colleagues! Anyone tried Rotary Embeddings for MT, especially of longer texts? According to the paper, it should help, but the reported score improvement is fractional. Any practical advice?

Update: A couple of us Google Brain / DeepMind researchers have left to work on something new. Some observations: * Off-the shelf AI libraries (e.g. RAG and agents) do not work well enough for high-quality products. * In-house research is essential to get the higher

Hiring a postgraduate Research Intern - #MT and #LLMs - paid and fully-remote. #nlproc linkedin.com/posts/yasmin-m…

If you need a sensei who gives good guidance, try this.

Let's ask ChatGPT or Llama this question.

I might be missing something, but what's the purpose of fine-tuning a 7B LLM for "zero-shot" translation if we can achieve a better quality with a 300M encoder-decoder MT model? The whole idea of using LLMs is real-time adaptivity and incorporating more context while translating.

SEACrowd is an excellent work, currently presented at #EMNLP2024 aclanthology.org/2024.emnlp-mai…

Interspeech URGENT 2025 Challenge. The task of this challenge is to build a speech enhancement system to adaptively handle input speech with different distortions and different input formats (e.g., sampling frequencies) in different acoustic environments. urgent-challenge.github.io/urgent2025/