Google Search is getting an upgrade that promises to boost the quality of results on the search engine.
The improvement will help Google comprehend queries that contain conversational language, and use prepositions like "for" and "to," which can significantly alter a sentence's meaning. It will also help Google deduce the intent behind grammatically incorrect queries.
The improvement only affects 10 percent of the English language searches for the US market. Nevertheless, Google vice president Pandu Nayak called it "the biggest leap forward in the past five years" for the world's most popular search engine.
"With the latest advancements from our research team in the science of language understanding—made possible by machine learning—we're making a significant improvement to how we understand queries," he wrote in a Friday blog post.
In the past, Google could get confused when it came to deciphering the full meaning of certain queries. An example includes "2019 brazil traveler to usa need a visa." You'd never speak this way with another human. But people can end up inputting search queries to Google in this way to emphasize certain keywords they hope the company's algorithms will understand.
Unfortunately, this approach doesn't always work. "The word 'to' and its relationship to the other words in the query are particularly important to understanding the meaning," Nayak said. "It's about a Brazilian traveling to the US, and not the other way around. Previously, our algorithms wouldn't understand the importance of this connection, and we returned results about US citizens traveling to Brazil.
Another example includes "can you get medicine for someone pharmacy." The user is trying to find out whether you can fill a prescription for someone else, like a family member. But in the past Google would overlook the phrase "for someone," and simply return a general result on filling a medical prescription.
So to fix the problem, the company incorporated a new language processing computing model, called BERT, which is designed to specifically understand the context around sentences. It does this by processing words in relation to all other words in the same sentence, "rather than one-by-one-in order," Nayak said.
As a result, Google's search engine should be smarter at understanding the nuances around your search queries. For instance, if you type in "parking on a hill with no curb," the search engine will now prioritize the words "no curb." Meanwhile, on a search for "math practice books for adults," Google will realize the word "adult" is vital and avoid returning search results relating to "young adults."
According to Payak, the BERT technology also works on improving search for other languages, including Korean, Hindi, and Portuguese. Still, don't expect it to be flawless.
"Even with BERT, we don't always get it right. If you search for 'what state is south of Nebraska,' BERT's best guess is a community called 'South Nebraska,'" he said, adding: "Language understanding remains an ongoing challenge, and it keeps us motivated to continue to improve Search."