W3C

– DRAFT –
Voice Interaction

23 February 2022

Attendees

Present
debbie, dirk, jim, jon, noreen, tobias, ulrike
Regrets
-
Chair
debbie
Scribe
ddahl

Meeting minutes

interfaces document

https://w3c.github.io/voiceinteraction/voice%20interaction%20drafts/paInterfaces/paInterfaces.htm

dirk: changes to interfaces document
… changes to rename API providers as External, new version of architecture
… External Data, Services, IPA providers
… architecture is version 1.3
… also added descriptions to UML notation, added some links to explanation of UML

dirk: looking at sequence diagram
… added small internal call in case you don't need to access external services
… or in case internal information supplements the external services
… the internal "processDialogInput" is not optional
… or it should be "processDialogInputs"

ulrike: what's the difference between "processInput()" and "processDialogInput"

dirk: "processInput" could trigger some internal "processDialogInput"

dirk: should add some more descriptive text
… "processInput" has two paths, local or external
… both should be combined to determine the next dialog step
… as a result, you may call an external service

jim: change "opt" to "optional"?

dirk: this is also standardized in UML
… the "opt service call" is also part of "processInput()"

ulrike: "processDialogInput" seems more like "process semantics"

dirk: finally, "deliverResponse" and play back audio
… "callService" is just a remote method call

dirk: should I explain that this is one cycle

tobias: there could be some indication that this could be a loop

debbie: it would be useful to have captions on the figures

dirk: it makes it clearer if everything has a return
… tried to reduce the amount of interfaces we need to describe
… could use "return value" instead of "clientResponse"

debbie: return value is implicit in the client input call

ulrike: is there something in the standard that would make this clearer?

dirk: also changed table in interface client input and external client input

debbie: do we have any way to report errors?

dirk: could be in multimodal output or we could gloss over error case

debbie: some kind of status

dirk: have "call result details" in a service call, or should that be reflected in client interaction

debbie: I think we should add that

dirk: success indicator plus optionally more detailed description

debbie: the user doesn't always need to know

dirk: the call to the Dialog might also fail
… then the Client needs to know
… will add something to table in Section 4.3
… the next interface is External Client Input
… return value is some semantic interpretation
… arrays should be list, need to consider multi-intents

dirk: nbest intents can be included in multiple interpretations
… for multi-intents don't cover
… yet

tobias: what about a group of utterances with a similar meaning

dirk: many different utterances can be mapped to an intent

debbie: we should provide for multi-intents

tobias: we might want to think about how other AI's annotate confidence
… can we standardize confidence

jim: confidence in AI's is not well-defined, they're just estimates

dirk: still need to talk about External Service Call

Minutes manually created (not a transcript), formatted by scribe.perl version 185 (Thu Dec 2 18:51:55 2021 UTC).