question

Most people are rather annoyed when it gives values not on the list. But that [i]is[/i] what the doc says. Note, though, the doc also says it is weighted towards the ones on the list. Axiomatically that means that it is weighted against those not on the list. It may be that whatever is said has a low confidence rating to start with that gets then weighted even lower, past whatever confidence threshold they use to determine if it is worth sending or not. I've got a feature requests somewhere for them to return several interpretations with the confidence value for each, so we can make our own determination. (That's how IBM's speech to text service works on Bluemix.)

Yes I think you are right. It appears I may have to abandon custom slots to implement their requirement - what I would like (at least to start) is a switch for "send everything" even if it's not a slot. That way I could implement what I want. Then, add confidence threshold (with option of 100 for only sending enumerated values).

I take it back! I switched back to my pre-custom slots version, and the functionality of the string Slot is now different than before. Previously, if someone tried a slot that I considered invalid (not in my sample utterances), it would return the slot with the value that didn't match. Now, it returns no value if it doesn't match something - custom slot or string slot. I can't seem to tell if someone is giving me an invalid slot so I can re-prompt/act differently. Example: Alexa ask Program for directions to the zoo Old and new behavior, custom or string slots: "slots": { "SlotName": { "name": "SlotName", "value": "zoo" } } this is the problem: Alexa ask Program for directions to the cat new behavior: "slots": { "SlotName": { "name": "SlotName" } } (which is the same result if they asked for directions without the slot being filled) The response I used to get: "slots": { "SlotName": { "name": "SlotName", "value": "cat" } } I want to respond differently in the same intent if someone just asks for directions and if someone asks for directions to a cat but that doesn't seem possible anymore. Anyone else noticed this?

If you want to differentiate between those two types of utterances you can just make two distinct intents: 1) Intent1 matches "Alexa ask Program for directions" 2) Intent2 matches "Alexa ask Program for directions to {place}"