Translating human languages via artificial intelligence techniques is the new paradigm of technology and now Google is also working on this aspect of technology. Some of the top spoken languages the world has already been translated by Google, and now it aims to polish its translations techniques with artificial intelligence.
Now Google AI team is working on a new and advanced translating model that would minimize gender-based bias when translating from other languages.
“We have made significant progress since our initial launch by increasing the quality of gender-based translations & also expanding it to 4 more language pairs. We are committed to further improve gender bias in Google Translate and plan to extend this work to document level translation as well, ” Google Research senior software engineer Melvin Johnson said recently.
There were some flaws in olden techniques of translation, e.g. the system was unable to translate Turkish to English and more than 40% of queries of translation didn’t work with Neural Machine Translation.
The old system of translation was not working adequately with gender-based translations, In the first step, the system did translations and when the translations were done successfully, the system used a rewriter that used AI and it produced alternative translations. For training the rewriter, the system divided the whole data into training and testing. More than a million English sentences were used, feminine and masculine data were also used to train the AI.
In the next step, training from both sides was merged into a single transformer based sequence to sequence and in the end, all the results obtained from this technique had 97% accuracy.
Image source link