4

I am building an Alexa app that needs to be able to process answers to a question. I have an SkipIntent intent that has sample utterances to skip a question.

I want to build an AnswerIntent that can take answers that can be anything and process them against the correct answer. I tried using an Amazon.LITERAL type for this with a few samples as such (from this question: How to accept the Free form text as input to Amazon Skill Kit?):

AnswerIntent {bottle|Answer}
AnswerIntent is it {bottle|Answer}
AnswerIntent is it a {bottle|Answer}
AnswerIntent is it an {bottle|Answer}
AnswerIntent a {bottle|Answer}
AnswerIntent an {bottle|Answer}
AnswerIntent {pillow|Answer}
AnswerIntent is it {pillow|Answer}
AnswerIntent is it a {pillow|Answer}
AnswerIntent is it an {pillow|Answer}
AnswerIntent a {pillow|Answer}
AnswerIntent an {pillow|Answer}

This actually works if I prepend the answer with "is it" or one of the other defined prefixes, but it doesn't get the "answer only" piece. It seems to get confused with my SkipIntent which is defined as:

SkipIntent i don't know
SkipIntent don't know
SkipIntent skip
SkipIntent i don't know that
SkipIntent who knows
SkipIntent i don't know this question
SkipIntent i don't know that one
SkipIntent dunno

Am I defining the AnswerIntent correctly? If not, is there a better way to catch an infinite amount of possibilities? Amazon seems to not like the LITERAL method, so I would be open to a better way.

Community
  • 1
  • 1
rhlsthrm
  • 769
  • 1
  • 12
  • 23

2 Answers2

9

Amazon's Alexa service is not designed for dictation. This has been the consistent response from the Developer Evangelists. So you will not be able to give the exact customer experience you want: being able take any sort of input on its own.

You are going beyond Alexa's design specifications by 'tricking' it into accepting a 'generic slot'. Plenty of people have done this, but it will never perform well, as you have found.

Specifically, for your use case, trying a generic slot on its own is particularly bad. Internally Alexa uses the interaction model to build up a tree of possible sentences that might be recognized. It then takes the input sounds, and matches them against the tree. Whichever branch ends up with the highest confidence is the branch that is selected.

When you add a generic slot, you add a node into the tree that matches almost anything, it competes with branches that mach specific values. E.g. if Alexa has the choice between "X", "Y" and "", it is usually going to pick "". This is why your a generic slot on its own is swallowing the other answers.

If you want a skill that performs with a high quality, you should seek a design that does not use generic slots. You might, for example, use multiple choice in a Question-and-Answer type skill. Or pick questions that have one of a specific set of answers, like colors or US States.

If you are just doing a demo, then, sure, use generic slots. You can do enough takes to get the recording to look good.

Joseph Jaquinta
  • 2,118
  • 17
  • 15
  • 1
    Thanks for the answer. I was afraid this was the case, but as you said, this is what the design specifications are. – rhlsthrm Jun 09 '16 at 04:41
  • The generic design does not super work well as well :) being able to pass arbitrary queries to the logic opens up possibilities to understand even more than just a simple intent with parameters and maintain code in one place instead of spread the logic across your own implementation and Alexa skills. – Macilias Oct 01 '17 at 10:43
  • I meant the original/intended design, not the "generic", sorry. – Macilias Oct 01 '17 at 12:57
1

As of January 2017, Amazon seems to be fine with AMAZON.LITERAL slots again:

Based on developer feedback, the AMAZON.LITERAL slot is not being removed as previously described. You can continue to submit new and updated English (US) skills with AMAZON.LITERAL. However, in many cases, custom slot types provide better accuracy than AMAZON.LITERAL, so we recommend that you consider migrating to custom slot types if possible.

Source: https://developer.amazon.com/public/solutions/alexa/alexa-skills-kit/docs/alexa-skills-kit-interaction-model-reference#literal

julien_c
  • 4,942
  • 5
  • 39
  • 54
  • and here: https://developer.amazon.com/de/blogs/post/tx3ihsfqsuf3rqp/why-a-custom-slot-is-the-literal-solution – Macilias Oct 01 '17 at 10:40