Naturally Conversational Demo on Google Assistant

Google Assistant to Get Follow Up Mode and Compound Requests

One of the demonstrations at the Google I/O developer conference today involved two new Google Assistant features that will make user interactions more conversational and more efficient. Scott Huffman, VP of Google Assistant engineering demonstrated these features by asking about a basketball game, information about the game, to start a popcorn maker and to set a reminder.

The first part of the demonstration shows what Google calls Continued Conversation. It is similar to Amazon’s recently announced follow-up mode for Alexa. Once you wake up a Google Assistant device, you can continue speaking and asking for additional information or engaging in other interactions without the need to use the, “Hey Google,” wake word. You can see this in action in the video demonstration below.

In the full demonstration you can also see how this fits into the concept of natural conversation. Mr. Huffman speaks to the assistant as if it is a person. He even includes information that you might convey in human conversation, but wouldn’t normally consider when speaking to a voice assistant. After Google Assistant answers his question about the game time, he says, “great, it will be fun to watch tonight.” He then asks Google to remind him to find his Kevin Durant jersey. So, he added some extraneous information between two commands and Google Assistant knew which information to ignore as extraneous conversation.

Multiple Actions = Compound Queries

Another feature highlight was called Multiple Actions which many people refer to as compound queries. This enables you to ask multiple questions and for Assistant to understand to them as different requests. Mr. Huffman’s compound query involved asking who was the Governor of California when Kevin Durant was drafted. It answered the name of the Governor, the year and the Seattle Supersonics as the team. These are both within the same domain in terms of both requiring interaction with Google’s Knowledge Graph. So, we haven’t seen cross domain Multiple Actions which involve gathering information and making a purchase for example. However, this is a good first step. The demonstration even showed an example where he used “and” to signify more information in a single request as opposed to indicating separate requests.

Mr. Huffman said the naturally conversational features will be arriving on Google Assistant in the “coming weeks,” but Multiple Actions are “coming soon.” I’m not sure how to decode that other than to say you should expect to see these new conversational features in 2018.

Alexa and Google Assistant Battle for Smart Home Leadership, Apple and Cortana Barely Register

Video: Alexa Context Maintenance in Follow-Up Mode Delivers Improved Voice UX