Virtual Assistants Discussions at SpeechTEK

Hello,

We had two discussions at SpeechTEK today on standards for virtual
assistants -- an informal morning discussion and a Birds of a Feather lunch.
The goal was to get feedback on our current ideas for possible standards and
get some new ideas from the SpeechTEK community. I think we had good
participation, about 20 people altogether, including several current
participants.

We reviewed some of our current ideas, for example: 

1. A mechanism for registration of virtual assistants, along with metadata
describing the assistant. This metadata could include, for example,
capabilities of the assistant, a phone number, and an associated website.
This would support discovery of virtual assistants, either by other virtual
assistants, or by users.

2. Protocols for communication between virtual assistants

3. semantics for common concepts (like time)

4. a format for ngram language models

5. a dialog markup language that could be used for declarative descriptions
of dialogs. 

 

The first idea seemed to be of the most interest.

 

New ideas:

1. standard support for lower-level turn taking, separating prompting from
recognition

2. support for common ways of describing users

3. change name and scope of group to include text-based interaction

4. prepare a document on best practices for virtual agent interaction design
(this could be a joint effort with AVIxD (Association of Voice User
Interface Designers)

5. a dialog annotation format for training statistical dialog models

6. support for omnichannel interaction (for example, a format for an
enterprise-wide common dictionary)

7. keeping track of people as they move through a customer journey

8. reusable SCXML-based dialog components

 

Next steps would include group participants preparing white papers on topics
of interest (they could be these topics or other topics that aren't listed
here), circulating them to the group for comment, and publishing them as W3C
Notes.

 

Debbie