Intent Builder vocab files extracting from remainder portion of utterance

I am seeing an issue that I have not noticed before with the following behaviour.
I have a skill that is built using the intent parser with the following vocab files. DeviceKeyword.voc, LightKeyword.voc, and OnKeyword.voc. In the OnKeyword.voc I simply have the word “on”. I have noticed that when I am looking to process the remainder of the utternance that it has extracted all instances of the vocab files.
eg. “Turn the room light on in 30 seconds” will return a remainder of “Turn in 30 secds”
I would expect it to return “Turn the in 30 seconds”
Is this a feature or flaw :wink:

the utterance is normalized, written numbers become digits, contractions are expanded, and article (“the”, “a”, “an”) are removed, keep this in mind when making your voc files! if they have the/a/an they will never trigger

1 Like

That makes some sense but what i don’t understand is why the word ‘on’ is being detected based on my onkeyword.voc and it is also catching and removing ‘on’ in the word secONds. Therefore the remainder becomes ‘secds’.

the on dissapearing in seconds is a bug, the utterance remainder is replacing every instance and does not account for when its part of a bigger word

fixed in this PR

2 Likes

Thanks for looking into it and the pull request.

Great detective work, @Jarbas_Ai

When could i expect this to get pushed to my device?

Hi there @pcwii, I will ping @forslund to see if we can get this into 18.8.6 this week.

It is already pulled in, so just a matter of waiting for the next version bump. (Or do a git pull dev yourself now)

Yes, thanks to this report and @Jarbas_Ai’s fast fix we’ll have it included in the next release (on thursday)

2 Likes