Issues with the discussion

Configuration
1

Hello

I have a strange behavior via the discussion tab:

Capture d’écran 2020-11-16 à 13.12.54
Capture d’écran 2020-11-16 à 13.12.541576×968 73.1 KB

2

Does a ceiling turn off? :smiley:

I’m not sure about that :face_with_raised_eyebrow:

3

To my mind, the model does not recognize the command « Turn off the kitchen ceiling Â» because of the spelling mistake on the verb « Ă©teindre Â» which in the imperative is normally « Ă©teins Â» with an S :grin:

Screenshot 2020-11-16 at 14.40.05
Screenshot 2020-11-16 at 14.40.05698×164 13.1 KB

In the model, the command is:

{
    "label": "light.turn-off",
    "questions": ["Turn off the light in the %room%", "Turn off in the %room%"],
    "answers": ["Extinction des feux...", "I'm turning off the light!"]
}

I don’t remember how much tolerance there is for a spelling mistake in a word in the lib we use (nlp.js), there is tolerance for the general syntax of the sentence but I don’t remember for the « word by word Â» part.

Given that in your sentence the only word that makes sense is that word and it is misspelled, of course the model does not understand
 :smiley: But I agree that it would be cool if the model were more flexible with errors.

I created an issue to keep the issue in mind, but it’s very very low priority, we have millions of things before that :slight_smile:

4

Thanks for pointing out my grammar

But it does solve the issue. Lack of flexibility, all that
 :stuck_out_tongue:

5

Another issue

Capture d’écran 2020-11-16 à 14.03.38
Capture d’écran 2020-11-16 à 14.03.381546×920 73.6 KB

First order, the light called piano is not physically turned on (and therefore not in the UI) but Gladys says yes.

Second order: it works.

6

This is a real bug, not very complicated to fix, I created an issue:

7

Oh, how nice of him! :wink:

8

I had the same experience as you @Hamtaro, it’s true, we should relax the tolerance on typos. :wink:

9

Shouldn’t we enrich the sentences we can write in Gladys to have something nicer?
Because we often have « I don’t understand. Â»

Like: hello, hi, how are you,

Basic things for now and to be enriched over time.

10

That’s where it happens

11

I opened a PR on GitHub, but you need to tell me if what I did is good.
I could add more sentences later.

12

Don’t forget English :wink:

13

Yes, of course.
I will check for the French version if @pierre-gilles accepts my additions, I will do the English version.

14

What I find a bit disappointing is that just as everyone is getting used to talking to the Google Assistant
 here we are with an « assistant Â» that doesn’t really understand what we’re saying.

Silly me, I’d like to be able to say:
« turn on the light in the bedroom Â»
« turn on the bedroom light Â»
« turn on the light in the bedroom Â»
« turn on the bedroom bulb Â»

And for all of that to have just one result
 that the damn bulb gets turned on lol

I might be imagining things, but is this feasible?

15

In my opinion, it’s feasible, you just need to put the different turns of phrase in « questions Â» to get the desired result.

For example when you tell it:
« questions Â»: [« Turn on the light in the %room% Â», « Turn on in the %room% Â»],
« answers Â»: [« Turning on
 Â», « I’m turning on the light! Â»]

16

And it is the case! :stuck_out_tongue:

17

Oh, then I didn’t take the right example, because when I tested the conversation a bit, I got too (too many?) answers: I don’t understand

18

The issue is the spelling of the trigger, not specifically the articles.
And it’s annoying when you get a rejection for a « Ă©teint Â» instead of « Ă©teins Â»â€Š

19

I am adding sentences to the chat.
I have added lines with jokes that I will enrich over time.
What would you like to ask Gladys?

  • Shi fu mi
  • A number between 0 and 10
  • 

20

Heads or tails!
It’s always handy not to make a decision :slight_smile: