Google has created a dataset to better train virtual assistants. The dataset contains over 550,000 conversations between humans and virtual assistants in English, French, German, Hindi, Japanese, and Spanish. It includes conversations in which the user had to revise their instructions after the virtual assistant misunderstood the initial request, conversations with repeated or filler words, and conversations in which the user switches languages while saying an instruction, as well as users’ lists, notes, and contacts to provide context for each conversation.
Image credit: Flickr user Marc Wathieu