News and Views on Tibet

Microsoft’s translation app includes Tibetan language

Share on facebook
Share on google
Share on twitter
Microsoft's translation app can now translate Tibetan language (Graphics/Tenzin Lekhden)

By Tenzin Lekhden

DHARAMSHALA, Dec. 3: Microsoft’s Artificial Intelligence is now able to translate Tibetan from other languages, but with some caveats. The tech giant’s translation software hit the 100 plus languages milestone last October after adding 12 new languages, which are natively spoken by 84.6 million people, Microsoft writes in a blog post. The milestone comes five years after Google’s translation software hit the same milestone, which is now capable of translating 109 languages, which is 6 more than Microsoft.

The addition of 12 new languages and dialects, Microsoft claims, makes “information in text and documents accessible to 5.66 billion people worldwide.” This achievement is made possible by recent advances in AI technology, allowing Microsoft, Google, and other software companies to grow their language libraries, even with low-resource and endangered languages. Microsoft, in a blog, explained that they use a multilingual AI model called Z-code to train different yet related languages with a shared model. This method allows the various language models to learn from each other and reduces the data requirement.

Buddha Kyab, translator and children’s literature writer, on the Tibetan translation of Microsoft’s software, told Phayul that “the software is capable of conveying a rough meaning of the text given as an input, which could be practical in given cases. But gives a literal translation of words absent of contextual meaning.” In a layman’s test that we conducted, the software is perfectly capable of translating literal sentences but the accuracy takes a dip as sentences get more complex and contextualised. 

Microsoft’s Azure cognitive services provide a variety of services for languages, such as text translation, auto language detection, and custom translator, but for Tibetan, only text translation is available at the time of writing this article. Audio and voice recognition features are also not available. 

Compared to Google’s Tibetan translation software, which outputs hit-or-miss sentences, Microsoft’s software, on the other hand, outputs consistent and comprehensible sentences upon given the same inputs. Google’s Tibetan translation software is still in beta version and asks for a variety of user inputs like evaluating beta translations, validating translation for accuracy, and translating words and phrases. The software in beta has till date 45,000 contributions from various individuals.

Leave a Reply

Your email address will not be published. Required fields are marked *