Friday, 24 March, 2023

Google uses ML to find out language translation work goof ups

<p>Google scientists, for machine translation, compile a data set of over a thousand languages. There are very few papers to utilise as samples for ML in so-called low-resource languages. Authors comb the Internet for texts in low-resource languages and use ML techniques to identify them. 13 of the 38 languages had a quality rating of less than four on a scale of 0 to six. Humans labelled 20% of the translations as &quot;nonsense/wrong language,&quot; deeming them unsuitable.<br /> &nbsp;</p>