Working Around Machine Translation’s Need for Large-Scale Training Data
Unlike some fields in machine learning, machine translation still requires large sets of training data. The solution? Creating more data when none (or not enough) exists.
*New* Slator Pro Guide: Translation AI — 10 Practical LLM Use Cases for the Language Industry
Unlike some fields in machine learning, machine translation still requires large sets of training data. The solution? Creating more data when none (or not enough) exists.
The new model, praised by Meta founder and CEO Mark Zuckerberg, is already being used to improve more than 25 billion translations daily on Facebook, Instagram, and other apps.
Meta eliminates text generation in speech to speech translation; shares code and research with the public.
Among the many things cooking at Meta: downloadable language packs for Android and iOS users to enjoy download-on-demand translation.
Meta’s VP of Internationalization, Iris Orriss, joins SlatorPod to talk about language operations at Meta, pioneering new localization approaches, and the future of the Metaverse.
In virtual reality, physical borders may be less of an issue than language barriers. As Meta hypes its Metaverse, language managers work behind the scenes to localize the space.
Meta shoots for completing world’s fastest supercomputer by mid-2022. Here’s what it can do in terms of natural language processing and translation — and what the facility looks like today.
Curating datasets, reviewing user-generated content, and liaising with locals — all in a day’s work for a linguist hired by big tech. (No computer science degree or published research required.)
To be fair, this is the first-ever multilingual model to win the international machine translation contest. But major tech companies have been exploring multilingual models for years.
As the language industry turns to speech-to-speech translation, Facebook AI partners with “AI community” Hugging Face to release speech-to-text translation models for four languages.
Zoom call add-ons planned for beta release in September 2021. Transcription available in 30 languages, and translation, in 12, by year-end 2022.
How DeepL outmaneuvered big tech, won a following, and is now rapidly moving into machine translation as a managed service for enterprise customers.
Facebook sees textless natural language processing (NLP) as rendering automatic speech translation (ASR) obsolete by working in true end-to-end fashion: from speech input to speech output.
As Facebook and Google pour their own money into research on speech translation, investors outside the language industry turn their attention to multilingual voice assistants.
New paper by Facebook, Amazon, Twitter, and University of Melbourne examines vulnerability of systems using back translation to attacks on monolingual training data.
Slator Weekly: Join over 15,800 subscribers and get the latest language industry intelligence every Friday
Tool Box: Join 10,000 subscribers for your monthly linguist technology update.
Your information will not be shared with third parties. No Spam.
This will close in 0 seconds