Artificial Intelligence (AI) has already taken over the world, offering innovative solutions in almost every aspect of life. In a recent development in this field, search engine giant Google has come up with an ambitious project: Building a single AI language model that will support not one but 1000 most spoken languages of the world.&nbsp;&nbsp;



However, 1000 languages can’t be supported overnight. That is why, Google has taken its first step towards the goal by developing an AI language model with 400 languages. It is called Universal Speech Model (USM). Although it’s is just a prototype, this too has earned the status of “the largest language coverage seen in a speech model today.” Experts suggest that Google has always been ahead of others in coming up with language models, but this current development has led to the formation of both powerful and multifunctional “large language models” or LLMs and promises to bring such cutting-edge solutions to the market as soon as possible.&nbsp;&nbsp;



In fact, Google has already integrated the Universal Speech Model into its major language-based products such as Google Search, Google Translate and Google Assistant. The tech giant is hopeful that this trail attempt will help them understand about the in-depth systems operation, how the model can further be improved and promoted from 400 to 1000 language support in less than a year.&nbsp;&nbsp;



Interestingly, Google itself has revealed some of the prominent flaws from its test runs. Some of them are ejecting discriminatory and harmful social biases such as racism and xenophobia, as well as failure to parse languages as per human sensitivity and requirement. However, despite these root problems, Google’s “1000 Languages Global Initiative” has been deemed as necessary and potent with the ability to perform many tasks such as basic language generation (faster than OpenAI’s GPT-3), easy translation (influential than Meta’s ‘No Language Left Behind’ or NLLB system) and above all being a single system with great many languages spoken across the world. Interestingly, Meta’s NLLB (unveiled in July 2022) is an AI-based advanced open-source language model that can conduct high-quality direct translations between 200 languages (including translations for 55 African languages, that is a first). OpenAI’s GPT-3, on the other hand, is an autoregressive language model that generates human-like text through deep learning methods.&nbsp;&nbsp;



Next up, Google is planning on implementing the 400 language AI model into YouTube captions. The goal is to make information more accessible while also expanding its language portfolio. Additionally, the tech major also plans to add 24 more languages to the Google Translate platform alongside enabling voice typing for 9 African languages on Gboard.&nbsp;&nbsp;

Topics

Most Popular

loading loading loading loading loading loading

loading loading loading loading loading loading

loading loading loading loading loading loading

loading loading loading loading loading loading

loading loading loading loading loading loading

loading loading loading loading loading loading