Alexa skill doesn't interpret some stuff correctly

275 views Asked by At

Basically, my skill has two intents, let's say "intent1" and "intent2".

Here's my issue. When I type a phrase that doesn't have a valid intent, instead of going to the help menu like (I think?) it should, it just goes right to intent2 or intent1. I just don't get it. UDPATE: it appears that it's just picking whichever it thinks is the closest no matter how far-fetched. Is there any way to have it just go to the help menu when it's not a match?

I'm a newbie to Alexa development, so please do keep that in mind when responding. My skill is largely based on the Space Geek sample.

If more code is needed to answer the question, just let me know and I'll add it.

Thanks so much!

Update:

Intent Schema:

{
  "intents": [
    {
      "intent": "GetNewDogThoughtIntent"
    },
     {
      "intent": "GetNewCatThoughtIntent"
    },
    {
      "intent": "AMAZON.HelpIntent"
    },
    {
      "intent": "AMAZON.StopIntent"
    },
    {
      "intent": "AMAZON.CancelIntent"
    }
  ]
 }

Sample Utterances:

GetNewDogThoughtIntent  What is my dog thinking
GetNewDogThoughtIntent  What's my dog thinking
GetNewDogThoughtIntent  What my dog is thinking
GetNewCatThoughtIntent  What is my cat thinking
GetNewCatThoughtIntent  What's my cat thinking
GetNewCatThoughtIntent  What my cat is thinking
2

There are 2 answers

0
Branch On BEST ANSWER

Turns out, according to Amazon, this is just how Alexa works. They said they're considering changing it but no sort of eta as of now.

Thanks to everyone who answered!

2
Jim Rush On

Speech recognition layers are usually optimized to match the closest speech vs excluding (no match) on low confidence as it is more likely, in general, that somebody is saying something intended than something random. I've found that Alexa takes this concept fairly far, as you've found.

If you have specific types of phrases that you think are going to be commonly spoken by your users that you will not support, I recommend creating one or more intents to catch those phrases and route them into your error/help message.