Google To Suggest Email Replies

send111

Google’s Inbox email app will soon not just make it quicker to reply to messages but will help you decide what to say.

A new ‘Smart Reply’ feature in the app (which runs separately from the main Gmail app and includes a variety of organizational tools) is based around the idea that often messages simply need a reply of few words. The theory is that people tend to overthink how to reply to a message and not only eat up more of their own time thinking what to write, but go into excessively elaborate detail that the other person in the conversation then has to wade through.

Of course, many users don’t see this as a problem and enjoy detailed communication with more of a careful personal touch. This is aimed more at people who deal with huge numbers of messages a day and for whom going through the Inbox feels like work. It’s also designed for the fact that users of the app are on a mobile device where lengthy typing can be a pain.

‘Smart Reply’ works by offering a small selection of possible replies (up to three in the examples Google has demonstrated) to a message, based on the content of the message.

For example, a message that (among more elaborate detail) asks if the recipient has set their vacation plans yet triggers the feature to offer the responses “No plans yet.” “I just sent them to you.” or “I’m working on them.” Tapping any of the responses automatically adds the text to a reply, at which point the user can either simply hit send or add more detail.

Google has detailed the process of building the feature, noting that it decided against simply trying to create a set of fixed rules for parsing text and deciding options. Instead it concentrated on extensive testing and refining, with the tool learning from the responses that human testers chose as the most suitable. (While the development drew heavily on a data set of real Gmail exchanges, the staff involved weren’t allowed to see the content of those messages.) That refinement will continue based on the choices of real users.

Among the main problems that needed tweaking were the system coming up with three likely responses which largely amounted to the same meaning, and the system being too keen on suggesting “I Love You” as a response suitable for almost any occasion, something that’s unlikely to be true for users powering through a busy work inbox.

That appears to be down to the writers of the real emails in the dataset being likely to use the term when replying to a loved one regardless of the wider subject matter in the exchange. That led to Google tweaking the system to put more emphasis on the contextual link between the original message and the suggested reply.


Geeks are Sexy needs YOUR help. Learn more about how YOU can support us here.