Google Home Follows Conversational Recipes

Following a cooking recipe is a simplified kind of conversation.   The assumed script is something that has been standardized over time. There are ingredient list and quantities,  cooking instructions,  Serving instructions.  Then there are are comments/assumptions on the script.  Potential substitutions, alternative methods, techniques, visuals,  ...  And then there are interactions of various sorts, which depends on how experienced a cook you are.  But it all starts with  a few simple lists.  At the end perhaps some helpful hints, like serving quantities. So this is a good place to start thinking about useful conversations and their context.

Google Home just added a 'skill' for cooking. You can ask it for a recipe via search,  or find one of your via search and send it to the Google Home App.  It interprets it, and then you can start following the recipe. So it really supports recipe browsing.

Amazon Echo has had a similar recipe reciting capability for sometime.  It has a far shorter list of recipes derived from Allrecipes, but those alone are quite complete.  Amazon does not support easy recipe browsing.

In each system you no longer need to load the skill.  Just ask some questions that imply you are interested in cooking recipes.   And off you go.  More details of the experiment below:



I ran both against some very simple examples, like 'Baked Chicken' to get a feeling for how the interface worked.   Both worked fairly well.  Whats most annoying in such a system is when it misinterprets s very simple command and you have to repeat.  You move forward in the list by saying 'Next' and its easy to repeat a step.    Assuming you are not following in the App, it is easy to get lost. I was  was surprised that Google more often lost track of what 'mode' I was in, and I had to start adding the trigger word for each command, while Amazon kept it straight once I was in list mode.  A considerable negative.  I also tried to foul up each interface by adding additional words like 'Next Step Please',   But they both dealt with that.

Overall the Amazon approach was cleaner and easier to follow.  I saw no evidence that there was any subtle natural language or remembered context thing going on. No real conversation occurred, that is I was never asked anything clarifying, like we would expect in a human conversation.  Just indications that it had not understood.

This is a good place to start when thinking about conversations.


OK Google, Lets Start Cooking.

Some video examples:

https://support.google.com/googlehome/answer/7309433?utm_source=EMAIL_H_M_1_B__100066322_10033935_201704281101__us__learn_more_en&utm_campaign=100066322_10033935_201704281101&utm_medium=EMAIL_H_M_1_B&utm_term=learn_more_en&utm_content=us


https://www.youtube.com/watch?v=GdVGynGWQ2Y&feature=youtu.be&utm_source=EMAIL_H_M_1_B__100066322_10033935_201704281101__us__video_en&utm_campaign=100066322_10033935_201704281101&utm_medium=EMAIL_H_M_1_B&utm_term=video_en&utm_content=us