<scribe> Scribe: Kaz
dahl: mentions there is a CG on voice interaction
bruno: any APIs available?
dahl: Amazon provides methods to access
david: you can
dahl: there was some standard format
in 90s
... very constrained model
... but now N-gram LM is available
... vocabularies
david: 3 years ago LM for financial service was nightmare
bruno: tools to fill sentences available these days
dahl: there are various viewpoints for virtual asistant
david: you'll lose context very
quickly
... very disconnected experience
mark: sometimes there is no
response
... depending on the design
david: very narrow experience
... maybe you get
dahl: Alexa is an app on mobile
sabria: have been working on
ARIA
... speech recognition is difficult
bruno: defining some keywords, e.g., "where am I", would be useful
david: sounds like accessibility discussion
lisa: there is no standard way to handle keypads, etc.
Flipchart:
+ Best practices in voice interaction design
+ AVIxD
dahl: will find a link and put it on the summary
bruno: it's kind of becoming
standards
... keywords for virtual assistant
lisa: semantics
... and GPS
... next turn left or right
... terms make sense to users
Flipchart:
+ Customize vocabulary
+ screen reader techiques
json: there is a best practice
... interest to standardize the grammar
... other areas to standardize
dahl: there is no standardization
about avators yet
... don't know if there is possibility
... several markup languages done not by W3C
... high-level concept like smile
... BML (Behavior Markup Language), etc.
lisa: robots are one of the early adapter
dahl: what's the difference betwen virtual assistants and robots
mark: Siri, etc., are user-driven
Flipchart:
+ Robot assistants
david: push model initiated by user
mark: we have modes for announcement, e.g., polite, in WAI-ARIA
dahl: guideline for safety, security and privacy for cars
Flipchart:
+ Safety, privacy, Security
jason: standard formats representing information
david: Eriza had a model on conversation
dahl: Eriza appeared in 1963
david: is there any chance to consider context capture?
dahl: there are many opportunities these days
david: context includes location, motion, ...
dahl: other people as well
sabrina: there is something in the semantic web area
dahl: some of tools have ideas of
"context"
... context is very useful
Flipchart:
+ adaptation standards
david: emotion action, etc.
... not so complex on the abstraction layer
lisa: using RDF tripples
sabrina: semantic
lisa: if you want to
Flipchart:
+ context
dahl: concept vs context
... getting back to the topic of virtual assista
david: some push model might be useful
dahl: we're getting out of time
... good discussion on virtual assistants
... there is a CG named voice CG
lisa: voice authentication would be possible
dahl: users may get benefit from a virtual assistant for accessibility purposes
Flipchart:
+ accessibility
mark: some kind of language for
experiments would be interesting
... interesting to see how tolerant a virtual assistant would
be
... Amazon recently released a virtual assistant service whith
listens to the user
jason: error tolerance is another point
dahl: use Dragon speech which uses
deep learning and it works well with my accent
... Microsoft also does deep learning
... there are not standard but conventional machine learning
systems used by researchers
[adjourned ]