<scribe> scribe:ddahl
https://w3c.github.io/voiceinteraction/voice%20interaction%20drafts/paArchitecture-1-1.htm
dirk: reviews changes
jon: difference between data providers and external services?
dirk: data providers are just
data, external services provide services
... data is "how high is the Eiffel Tower" services trigger
action
jon: what is the difference
between an explicit request vs. implicit request -- I don't
care who answers the question
... external service might be the target of an explicit
request
debbie: an implicit request could also be "it's cold in here", so should be careful about terminology
jon: data providers could be implicit request
dirk: this separation makes sense but these are orthogonal to what we have here
debbie: this could be a fuzzy distinction
dirk: data provider is kind of a
service
... question-answering is a big topic, so we might want them to
be called out separately
debbie: would there be different protocols to access the three gray boxes?
dirk: data providers and external services could be the same protocol, IPA could be different
debbie: are data providers and
external services stateless?
... or could they be either stateless or statefull?
dirk: maybe data providers are
stateless
... dialog maintains state information, data provider shouldn't
need to
... example if the data service returns two options, the dialog
has to disambiguate
debbie: architecture gets complicated if more than one component maintains state
dirk: added "local services" and
"local data providers"
... "local services" could be started locally, like temperature
measurement
jon: for example, in a car, there
could be local services, like temperature, lights, media,
GPS
... and local data providers
... also mobile phone services
... also client-specific services, like windows in a car
dirk: continues review, adds
conversational context, switching contexts, reasoning, long and
short-term memory, grounding
... nlp needs access to context
... context could be endless
debbie: some might be intractable
jon: are there simpler ideas about context?
debbie: support for NLU is possibly easier than drawing on all world knowledge
dirk: should put in disclaimer
that we aren't trying to deal with every aspect of
context
... could add captions to other figures
... should questions directed at other agents be included in
context?
jon: if I have explicitly a question about Liedel, does it remember that it has to do with groceries?
dirk: this is captured by long-term memory
jon: over time the value of context will grow
dirk: this should be added
jon: implied consent for
data-sharing
... I am willing to share my personal data with Liedel because
I trust them, but maybe I don't want to share data with
Lufthansa
debbie: too many data-sharing questions can be annoying
dirk: this can be contingent on previous behaviors
jon: jon's IPA gets better at knowing about jon
debbie: we need another session
to finish going over
... all the changes
dirk: is there a dedicated component to capture previous behaviors
debbie: could call this
history
... knowledge graph could be generic or specific to me
... my personal knowledge
... next call will be January 13
This is scribe.perl Revision of Date Check for newer version at http://dev.w3.org/cvsweb/~checkout~/2002/scribe/ Guessing input format: Irssi_ISO8601_Log_Text_Format (score 1.00) Present: dirk jon debbie Found Scribe: ddahl Inferring ScribeNick: ddahl WARNING: No date found! Assuming today. (Hint: Specify the W3C IRC log URL, and the date will be determined from that.) Or specify the date like this: <dbooth> Date: 12 Sep 2002 People with action items: WARNING: IRC log location not specified! (You can ignore this warning if you do not want the generated minutes to contain a link to the original IRC log.)[End of scribe.perl diagnostic output]