The Google team explains how it uses the search engine AI to interpret searches for COVID-19 vaccines, taking into account various variations of the name.
This is part of the first tests with a unified multitasking model to improve the search system.
Google’s AI improves search results for vaccines
We’ve all heard and even searched for information about The Google About some COVID-19 vaccines, such as AstraZeneca, Sputnik Moderna or Pfizer. However, these names, which are well known to us, may not mean anything to users looking for vaccines on their own label, given language, country or culture.
In fact, the Google team reports that there are more than 800 variations of vaccine names in over 50 languages. And that’s regardless of the number of different terms and phrases used to search for information about COVID-19 vaccines.
So that this is not a problem for the search engine, and can provide relevant results to users, Google implements the Unified Multitasking Model (MUM) technology. And what can take hundreds of hours and weeks of work, MUM can recognize you in seconds:
Using MUM, we were able to identify over 800 types of vaccine names in over 50 languages in a matter of seconds. After validating the MUM results, we applied them to a Google search so people could find timely, high-quality information about COVID-19 vaccines around the world.
This was possible thanks to one of the dynamics introduced by the unified model of multitasking. And in fact, it was one of the aspects that Google highlighted at its I/O event with the introduction of the new MUM model: knowledge transfer between languages.
And no, it’s not a simple translation:
For example, imagine reading a book; If you are a polyglot, you will be able to share the book’s main conclusions in other languages you speak, depending on your fluency, because you have an understanding of the book that does not depend on language or translation. MUM transmits knowledge across languages in the same way as this language.
That is, it can take information from sources in languages other than the original user’s query, process that content, and provide that information as a search result.
“Problem solver. Proud twitter specialist. Travel aficionado. Introvert. Coffee trailblazer. Professional zombie ninja. Extreme gamer.”