Like people, Google software should be able to decipher the meaning of queries even if they are unclear or seem to have little meaning.
Google said last Friday that it had “made its biggest leap” in years in its search engine algorithm by unusually revealing details about its secret formula.
The technology giant uses machine learning to improve its understanding of everyday English. Google said on Friday that last year it had developed a technique for analyzing “natural language”.
This tool is based on a technology called “BERT” (Bidirectional Encoder Representations from Transformers), which is able to understand the different words of a search within a sentence and not isolate them.
Like humans, Google’s software must decipher the meaning of queries, even if they are unclear or seem to have little meaning.
“We are significantly improving our understanding of the queries, which is our biggest breakthrough in five years and one of the most significant advances in the history of (online) search,” said Pandu Nayak, vice president of search at Google.
Some of the models developed by Bert are so complex that they require ultra-powerful processors designed specifically for the “cloud” (remote computing), says Google.
“By applying the “BERT” models to both the ranking and the snippets that appear in the results, we can better help you find useful information,” says Pandu Nayak.
“In terms of ranking results, the technology will help the search engine better understand one in 10 English searches in the United States,” Nayak said.
He cited the example of a query in English that BERT now understands better “2019 Brazil traveler to USA need visa”. Previously, the algorithms did not understand some adverbs and could suggest results for U.S. citizens who want to go to Brazil. “With “Bert”, the search engine is able to capture the nuances,” he says.
Google said it wanted to extend these improvements to other languages and regions “over time”.
