Google 助理操作 - 隐式调用体验

Question

I am currently developing a Google Assistant Action. I understand Implicit Invocation as Google describes it and from my experience testing my own app, I very much understand that it works as they describe.

Here's what that flow looks like:

User: Ok Google, what's a good recipe I can make?
Google Assistant: Okay. For that, you might like < Your App Here >. Wanna give it a try?
User: Yes
Google Assistant: Opening < Your App Here >
Google Assistant: You should make spaghetti!

Makes sense. I ask it a general question and it asks me if I would like to use a certain service to fulfill that request.

Why is this not the case for every action? For example, music.

User: Ok Google, play blink-182
Google Assistant: Playing blink-182 on Spotify

Alright. I configured Spotify to be my default music player, and playing music is a standard Google Action out of the box. I can understand why it skips the confirmation and goes right to the app.

How about a random service I download from the internet? Like this one .

User: Ok Google, enable study mode
Google Assistant: I'll make sure no one disturbs you

What?!

No confirmation?

How did the Google Assistant know that I want to use the IFTTT service and use the intent in that service to complete the request? It is clearly possible to improve the implicit invocation experience of a Google Assistant action so that if you don't have an obvious competing service installed on your device, it will go ahead and select an appropriate service and execute the request.

How in the world do you do this?

Answer 1

There's a lot going on with how invocations can be handled, so let's look at the ones you outline, along with a few that you haven't mentioned.

Explicit Invocations

You: Ok Google, talk to Shakespearean Insult
GA:  Ok, Here's Shakespearean Insult
SI:  ...

The most straightforward way is to invoke your action by the name you've given it. Since this is a unique name, Google can directly do a hand-off between the Assistant voice and the voice of your Action, although if the user uses it enough, that hand-off part gets shorter and shorter. So if you talk to Shakespearean Insult (in our example above) enough, or enough in a row, the conversation becomes:

You: Hey Google, talk to Shakespearean Insult
SI:  ...

Google is vague about when this might happen, or what circumstances will trigger it, but it will happen for Actions that you use most frequently.

Implicit Invocations

You: Ok Google, I need a chicken soup recipe.
GA:  Sure. For that, you might like Personal Chef. Wanna give it a try?

These are meant to help users with the "discovery problem" - how to find Actions if you don't know what they are named, but just what you want them to do. Similar in some ways to Search Engine Optimization (SEO), this lets you add invocation phrases that will also trigger your Action. Since many Actions may support these same phrases, however, Google will verify what Action the user wants to use.

In some cases, Google will also see what phrases are being used as part of your Action and may include your Action as part of a direct response if users ask a question or try to do something.

Similar to Explicit Invocations, if the user chooses the same Action enough, then Google may stop prompting if this is the Action they wish to use. Again, the exact level isn't clear, but it is possible.

Built-in Intents

You: Hey Google, I'd like to play a game.
GA:  Alright, I have a few options for that ... Which one would you like to try?

This is similar to Implicit Invocations, but the difference is that the possible invocation phrases have been determined by Google. Developers can register Actions to accept these built-in Intents.

Right now, only the "play a game" Intent is publicly available, but Google has indicated that they want to roll out more as they are developed and internationalized.

Music and Smart Home Commands

You: Ok Google, play blink-182
GA:  Playing blink-182 on Spotify

You: Turn on the living room light
GA:  <rising tone>

These are similar to how built-in Intents work, since Google has determined what the phrases will be that can invoke the different Intents. The difference, however, is that both require a little closer integration with Google, since the user specifies both through controls in the Google Home app.

IFTTT

IFTTT has special support built into the Google Assistant, although it comes with some limitations compared to other conversational actions. There is no need for an explicit confirmation, since you've already taken explicit actions to enable the Applet - you've had to link your account to IFTTT and setup the Applet. It also isn't as necessary - since all the Applet can do audibly is say one thing and then close, and that one thing was set by the user.

Routines

You: Hey Google, Good Night
GA:  <declining tone>
GA:  Ok, playing relaxing sounds for 30 minutes

You:  Hey Google, record my weight
GA:   Getting Vodo Drive
Vodo: Ok, I've added a row to the spreadsheet named weight. What would you like to set today's weight to?

One thing you didn't mention in your examples about improving invocation phrases are that users can set their own phrases however they want using Routines. This is similar to IFTTT since it has some built-in integration, but requires a user to explicitly create them. Routines are like aliases, where you can set a number of commands that the Assistant will execute when you say something.

Users can setup routines using the Google Home app. There are some routines that Google has offered names for, such as "Good Morning" and "Good Night", but you can also create your own routines with whatever name you want. In the example above, I have a routine named "record my weight" which I've told Google is equivalent to the command ask Vodo Drive to add a row to weight which will call the Vodo Drive action to open a spreadsheet, add a row to it, and prompt me for information.

While some built-in Routines limit you to non-conversational Actions, setting up a custom Routine gives you more flexibility.

Summary

Developers have some, but not a lot, of ways to get better invocation phrases. When users take advantage of them, however, Google makes sure that it is clear that they are not using something from Google - but rather a third party Action. Over time, Google makes this a little better as it is clear the user accepts this.

Users have some way to setup better invocation phrases themselves. In cases like this, since the user is explicitly setting things up against a third party, Google is a little more flexible in these cases, since the user has already acknowledged this third party interaction.