Amazon fallback intent in Newbies var o/p. HELP!

Bugs 🐞
#1

I am staring off here.
Why should something so simple as:
Creating a variable to hold User speech .
Placing that var in a Capture block.
Linking a Speech block with that var.
Adding random speech to the Capture block var…

and getting {Amazon fallback intent}… output in the Speech block var ???

No matter the Input type selected or no Input type selected

Driven crazy by this. I am sure that beginners have encountered this countless times.
Nothing found on why.

HELP!!

#2

what type of slot did you choose in Capture block?

fallback intent means user’s utterance does not mach your slot type like Alexa asking the name of animal, but users answering date of today. Caputure block is easy but too simple, so might not work with unexpected users’ utterances.

I recommend you should use Choice Block. Choice block is a little more complicated than Capture block and you need to know about conversatioan concepts such as intent, sample utterances and slot. but more flexible for variety of users’ utterances, get a slot value to variable, and can treat this type of unexpected utterances with else port.

see here:

https://docs.voiceflow.com/#/blocks/intents-slots

#3

I am obliged to you for your time and patience.
I have little if any experience of coding. Essentially, Hello World in Basic and C years ago and that from passing and rushed curiosity.
Could you bear with my ignorance a little while longer.

My understanding of a slot is that it will categorise a var iable so that Alexa may be “primed” to expect a particular type of content within that variable.

The slots available to me are: custom, airport, artist,colour,country

Using any or none of these resulted in the “fallback intent” issue.

I can see that the ambiguity that is introduced by a “spoken alphanumeruic” or utterance, as it were, necessitates a slot type categorisation at least to aid the Alexa AI in narrowing down from the infinite possible range of utterances.

Also, if I test the Prototype on laptop then it works. A spoken entry to the Capture variable is recognised and read back correctly in the following linked Speak block!

So, it is clear then that when uploaded to the Alexa device there is a problem. The Alexa AI needs more.

There must be some way that Alexa can deal with un-delimited spoken alphanumeric or utterance. The Alexa Reminder skill has “What do you want to be reminded of” prompting an un-delimited utterance.

My understanding of Choice and Intent and Slots and Utterances is that collectively they are ways of segmenting the possible range of utterances (infinite). Creating Intents delimits the subsection of the possible (very narrowly). Each Intent given typical Utterances and the particular intent further helped in identification by creating additional typically relevant Slot variables.

These, nevertheless, could go nowhere near to accomodating the possible range of utterances, no matter how many of them might be created.

Any idea how the unlimited range of utterances are accomodated in the Reminder skill?

I notice Fallback as an existing Intent in the Choice block. Perhaps if I discover what Fallback delimits then I might be on the right track!

Again, I appreciate you patience.

Cheers

#4

will just answer some essential points.

  • You should always use test simulartor on Alexa Developer Console, not Voiceflow’s Prototype testing. Prototype testing is good for simple logic testing, not for NLU.
  • You can use SearchQuery type slot to capture what user said freely. But there’s some limitations.
    • Choice Block supports SearchQuery slot type, but Capture Block does not support it.
    • SearchQuery slot type requires carrier phrases after/before searchQuery slot. You cannot use searchQuery slot sorely. ex) “I like {searchQuerySlot}” or “{searchQuerySlot}, please” are OK, but “{searchQuery}” is not OK.
    • SearchQuery slot type just captures everything user said. That means you will need to parse it as text.
  • Capturing Alphanumeric word like product id or serial number is one of current limitations of Alexa’s NLU. Basically, Alexa’s NLU is based on big data and statistical guessing. Generally, alphanumeric word is like a combination of random lettters, does not have contexts, and there are so many variations. so NLU doesn’t work well.

Fallback means that utterance by users does not match any sample utterances you added in Choice Block and it will trigger Fallback Intent without choosing it in Choice Block. Using Fallback Intent in Choice Block means you put sample utterances which DOES NOT MATCH intentionally. I think it’s not good idea, especially if you are new.