Nuance’s Seattle Office: The ‘Other Guys’ from Tegic and Their Fellow Startup Vets Build A Mobile Innovation Hub
(Page 2 of 2)
finally, it seems like it’s been forever—but we have Flex T9 out there and we have the ability to put Flex T9 on a whole bunch of different devices, including television sets and all of that, because we have the whole deployment machine, the innovation machine. That’s been the fun thing for me.”
XT9 started at Tegic before the Nuance deal—it was a rebuilding of the predictive text input technology that started out in T9. The old version had been built to be extremely data-efficient, since it was going on relatively small devices. With XT9, developers could look ahead to really powerful mobile devices—and they built it to deal with all kinds of future user interfaces.
“The idea was looking toward the future when devices had more horsepower,” Bargen said. “When we started working on, for example, Trace, we had this huge, solid base underneath it of XT9. I mean, Trace is really a feature of XT9. It’s not like a new product in the sense that it’s a whole new code base. It’s sort of an input-output filter on XT9. That’s been fun, and everybody’s really enjoying that. We have a nice rivalry going with the Swype team here.”
A huge area of focus for any company trying to translate human ideas into technical tasks is language, both written and spoken. For Nuance, that effort includes field teams that work all over the world to help document and understand different languages, as well as high-powered computing that crawls the Web to understand how people are using typed language to find things and express ideas.
“One of the things that’s moved forward our language model a lot here for text input, for typing and for tracing, has been the work we’ve been doing with the speech team. Because they use much deeper language modeling and they have a huge store of samples and things that they use for language modeling,” Bargen said.
As for the future, Nuance is working on taking its technology to cars and TVs, Bargen said—a whole new level of complexities to work through, particularly when it comes to autos. That’s underscored by today’s news that Nuance has acquired SVOX, which provides speech-recognition products for cars. So how quickly will we all be talking to our cars and living-room boxes, Jetsons-style?
Bargen says there’s still a pretty big data gap in language recognition—despite all the computing power, years of experience, and PhD-level linguists studying the problems.
“I feel like there’s still a gap there in terms of having data sources where you can say, ‘I’d like to watch an action film tonight.’ Now, you can imagine you’d get a list of action films, but getting more complicated than that, you start running out of data. That’s my impression from some of the work we’ve been doing. So getting the data becomes more of the challenge,” Bargen said.
“If you said, ‘How many phone calls did I make to Uruguay last month?’ AT&T doesn’t expose that level of data to anyone, I’m assuming. So you’re stuck.”
It’s funny how dramatically things have changed in the mobile software business: People like the team from Tegic used to focus on how to shrink their product down as small as possible to use up very little room on a phone. Now, with powerful minicomputers in everyone’s pockets, they’re worried about not having enough data fast enough to perform the magic tricks they really want to do.
“In the beginning, it was very much a problem of being able to compress these lists of words and things just amazingly—to the point where I think it was like one byte a word. It was great compression. But you’re so focused on getting it into a small footprint that there was no temptation to add the frills and things. But now, you have the headroom to add the frills,” Bargen said.