Several Google results, such as Search and Assistant, are already available in India in several local languages. The company is now turning to new AI to make more of their offerings accessible to Signal language speakers – in particular, they use a technology called MuRIL.
At its landmark event today, the Big G unveiled a new language model called Multilingual Productions for Indian Languages (MuRIL). This is the first model to support interaction between 16 different Signal languages.
That includes Asamese, Bengali, English, Gujarati, Hindi, Kannada, Kashmiri, Malayalam, Marathi, Nepali, Oriya, Punjabi, Sanskrit, Sindhi, Tamil, Telugu, and Urdu.
While MuRIL is based on Google’s own BERT (Bidirectional Encoder Production Models from Transformers), researchers say it is more effective for Indian languages.
Partha Talukdar, a researcher at Google India, said the new model better understands the context of statements in local languages.
For example, understood the previous model the following Hindi statement as a negative emotion: a statement in Hindi “Accha hua account bandh ho gaya” (It is good that the account was closed). However, the new model correctly predicts that the statement is positive.
Users in India often use their English keyboard to type in local languages - like the phrase above. For that, researchers have introduced support for finding translations into other languages while using the Roman script.
Google is making this model accessible for other researchers and beginners to use.
Currently, MuRIL is not rooted in any Google results. However, based on feedback from researchers and programmers, it aims to incorporate this model into its future offerings for greater accuracy.
You can learn more and check out the MuRIL code here.
Published December 17, 2020 – 06:04 PM UTC