Google Docs Deploys Machine Learning AI To Fix Your Gnarly Grammatical Errors

Grammar
Most word processing programs will go through your work and at least perform a spellcheck, looking for misspelled words that you do not want appearing on your final document. Others look for mistakes in grammar. Google is implementing the latter into Docs as part of a sweeping G Suite update, and what is interesting about Google's implementation is that it uses machine learning to identify grammatical errors.

"It can be tricky at times to catch things like spelling and grammatical errors that inadvertently change the meaning of a sentence. That’s why we’re introducing grammar suggestions in Docs. To solve grammar corrections, we use a unique machine translation-based approach to recognize errors and suggest corrections on the fly," Google states in a blog post.

Google says its artificial intelligence scheme can identify several different types of corrections, everything from simple grammatical rules like when to use articles in a sentence, to more complicated rules of grammar that people often trip over. This can include things like subordinate clauses and how to use them correctly.

Since Google is employing machine learning to the task, the implementation will improve as time goes on, based on the types of errors its algorithm sees. "Machine learning will help improve this capability over time to detect trickier grammar issues. And because it’s built natively in Docs, it’s highly secure and reliable," Google says.

David Thacker, Google's VP for G Suite product management, talked about the use of machine learning ahead of the announcement. He likened it to the approach that Google takes with language translation. Instead of translating French to English, however, Google's AI is able to take improper English and translate it into proper English.

If you want to try it out, you can right now—grammar suggestions in Docs is available today in Google's Early Adopter Program.

Via:  Google
Show comments blog comments powered by Disqus