Google Search is a service many of us use multiple times a day, but have you ever thought about how you phrase your queries when you start to type in the search bar? If you’ve ever searched for something over and over again, changing your wording slightly each time to try and make Google understand exactly what you’re looking for, then you’ll likely be happy to hear today’s news.
As Pandu Nayak, Google fellow and VP of Search, wrote in an official blog post today, Google (the company) wants to make Google Search more effective at understanding natural language, rather than making users have to translate their questions into what they call "keyword-ese" to help the search engine figure out what to look for.
Google’s has achieved this is by using Bidirectional Encoder Representations from Transformers, shortened to BERT. BERT technology, powered by new cloud server hardware, is able to understand a given word more effectively by inferring the context from the other words in the search phrase.
In practise, Google claims that this will help Google Search better understand 1 in 10 of everything it gets asked. And given that Google processes billions of search queries a day, that’s a lot of people who can get better answers to whatever’s on their mind.
The blog highlights how small words can now drastically change what results will pop up, examples being how BERT notices the "to" in the query below and realizes the user wants to travel from Brazil, not from the US (above); or how it understands that the "no" in this search means it should display a different parking guide.
This feature currently only works for searches in English, but Google has implemented this new programming in featured snippets (the highlighted result at the top of the search page that Google thinks is the most relevant for your query) for several other languages, and promises that BERT will be applied to more languages around the world soon.
While this change might not make Google foolproof, you’ll at least be able to ask questions in a less artificial feeling way; which will be particularly handy if you’re using Google Assistant.