Google now understands more conversational search queries

Google Search has just gotten better at deciphering your sometimes conversational, sometimes awkwardly phrased queries. That's made possible by implementing a neural network-based technique for language processing called Bidirectional Encoder Representations from Transformers, or BERT, which gives Search the power to recognize the importance of word sequences. The company says it's the product's "biggest leap forward in the past five years, and one of the biggest leaps forward in the history of Search."

By being able to recognize word sequences, Search is also able to understand the context of words in the query. For instance, it's now better at understand queries wherein prepositions like "for" and "to" matter a lot to their meaning. In the past, when you search for "brazil traveler to USA need a visa," you'll see results that cater to Americans looking to travel to Brazil at the top of the pile. With BERT implemented, the first results you get are now links to information on how to get a US visa as a Brazilian.

Here's another example: pre-BERT Google returns generic results on how get a prescription filled when you type in something like "Can you get medicine for someone pharmacy." Now, Search prioritizes results with instructions on how to prescription filled for someone else, which is what you'd mean with a query like that.

At a press event announcing the big Search update, BuzzFeed News says Google VP of Search Pandu Nayak also gave this query as an example: "How old was Taylor Swift when Kanye went on stage?" Before BERT, Google would've returned a video of Kanye's infamous award-crashing rant at the MTV Video Music Awards. Now, the first result is a snippet of a BBC article with Swift's age -- it's even highlighted to make sure you see it.

Despite the massive update, Google admits that Search is still far from perfect. If you ask it "what state is south of Nebraska," Google will still surface results for a community called "South Nebraska" instead of for Kansas even with BERT. Nayak wrote in his announcement post that "language understanding remains an ongoing challenge, and it keeps [the team] motivated to continue to improve Search." For now, BERT will only be applied to English-language searches made in the US, but Google plans to expand its reach to more regions and languages over time.