The battle for handwriting recognition

Zi Corp.is set to acquire the private-owned Swedish handwriting recognition developer Decuma AB for about US$ 1 million in newly issued common shares. Decuma's handwriting software us its patented Geometrical Invariant Technology, with versions for Western languages (Albanian, Danish, Dutch, English, Finnish, French, German, Icelandic, Irish, Italian, Luxembourgish, Norwegian, Portuguese, Spanish and Swedish characters as well as different writing styles in these languages), Japanese and Chinese.

Decuma's design features a single input area for writing and presenting recognized letters and for editing the text. This makes input natural and fast, and helps users focus on content rather than writing as such. Uppercase and lowercase characters, digits, punctuation marks and special symbols, as well as accented characters, can be mixed in the same input mode.

Educating the customer

Voice Buzz has a useful interview with Michael Zirngibl, head of Angel.com that provides a voice site platform. Two issues are worth quoting for the language technology business:

The first is about going beyond the ‘technology’ word:

The biggest ‘competitor’ for any player in the speech market is non-consumption. We need to start a “buzz about Voice” and get people understanding what it can do for any business. With Angel, we want the customer to think “Voice Sites” instead of “Speech Applications.” As soon as businesses start to realize that they can use voice technology to improve, enhance or extend most of the things they currently do with their website, we will be a lot closer to a “Speech Revolution.”

The second is about educating customers:

“One of the biggest challenges in the voice business remains educating customers on a broader scale.

A lot of companies in the space continue to focus their efforts on targeting a very niche group of decision-makers, thus limiting the buy-in speech applications can receive throughout an entire organization. At Angel.com we focus on the business benefits of the resulting applications, which makes us much more attractive to decision makers outside the traditional call center space. In fact, most of our best customer wins come from savvy business buyers in Marketing, HR, or other “traditionally non IVR buying” departments.

The education challenge is one we believe can also be solved in part through better distribution channels to small-to-mid size enterprises. While we hear a lot in the mainstream press about the huge deployments, this is not what the industry needs in order to gain critical mass. The industry needs a way to reach the customers that might not even be considering speech technology right now, which is why one major focus of Angel.com in 2005 will be distribution.”

The need to educate customers would appear to be true for the language technology sector in general. One notable effort is the NOTaS (Ditch Organisation for Language and Speech Technoloogy) effort under way in Holland, and pioneered by Geert Kobus of Knowledge Concepts, who died tragically earlier this year.

Multilingual mobile project

MuLiMob (careful, I couldn't get the site to open) is a new support action for the European Commission's latest IT research and technology development programme. Here's the pitch:

If you work in the wireless industry, you might be interested in the MuLiMob project. Funded by the European Commission as part of the IST programme, MuLiMob SSA aims to make the most of Europe's multilingual and multicultural diversity for the benefit of its mobile users and workers, by making information available on current and potential developments and by identifying research to be done. Benefits of participation are:
- presence of your brand/company in our reports
- exclusive access to the findings and analyses of our research
- to get connected to the Mulilob network and community

Bill Gates on innovation in speech technology

Here is part of an interview with Bill Gates on innovation.

Asked whether innovation is incremental, BG said:

What about speech recognition? In the 1960s, people did 20% accuracy speech recognition. And the things we've done over a period of years to get to 30%, 40%, 50% -- is that incremental? Our speech server absolutely is a collection of about 5k incremental improvements in speech recognition that lead to a paradigm shift of how you interact with the computer. Most breakthroughs are a compilation, where a ton of incremental improvements come together.

You probably need to subscribe to get the full article since this blog gives another quote from the 'same' interview:

"We've invested 10 years in the Tablet PC" -- Microsoft's platform for pen-computing, which has won only modest adoption. "We were in there even before the fad came along and after the fad went away. . . . We've been working on speech recognition for a long, long, long, long time!" The Speech Server project began around 1995. It too has cost close to $100 million a year.

Everything depends on how you define the beginning of a project, because MS has been doing speech recognition since 1993, and of course all that work is behind what you see in today's Speech Server. But the same technology also shows up in Office, Windows (SAPI), Voice Command for PocketPC, and lots of cool upcoming products as well.

We talk internally a lot about our tenacity, and how we just never give up. We definitely think that way on the speech team, where one of our favorite discussions is about what we think the state of the art will be 5-10 years from now. It's especially fun to talk about stuff you know we'll be around to see implemented.

Information on English/Arabic and Farsi MT

Tim Oren is looking for serious information about best-of-breed technology sources for translation between English and Arabic and Farsi to help the blogging community. He says:

I'm already aware of Language Weaver and Meaningful Machines on the Arabic to English front. Since much of the development in these areas has been funded (quite openly) by DARPA and In-Q-Tel, there's more available for the path into English than the other way, but we're going to need both, so please send in any tips. The current state-of-the-art has moved to corpus based translation, but an older rule based system would be better than nothing. Right now, this is a technology survey, we'll worry about how to get the systems integration and business deals done later.

Language Weaver translation system version 2.4

The latest version of the Language Weaver MT system features
-- User Dictionary support that enables users to create dictionaries containing customized translations for words and phrases
-- Web Services (SOAP) API enables programmers to create client applications using multiple programming languages for multiple target platforms -- includes interoperability with C++, .Net and Java
-- Translation Server may now be started as a Windows Service
-- Optional user authentication implemented using HTTP Basic Authentication via Apache
-- Overall smaller memory footprint
Pairs available: Arabic/English, Chinese/English, French/English, and Spanish/English; unidirectional languages include Somali to English and Hindi to English, with others coming soon.

Bank supports NLP-related research

State Street Corp.'s money management unit is supporting a research center at Massachusetts Institute of Technology where students learn how to apply financial technology in the business world.

Boston-based State Street's State Street Global Advisors (SSgA) unit will join five other sponsors in supporting the Laboratory for Financial Engineering at MIT's Sloan School of Management, in nearby Cambridge, Massachusetts, the company said.

At the lab, which was set up as partnership between industry and academia to promote quantitative research in financial engineering, State Street will support work that may help fund managers pick stocks and bonds more effectively.

For example, the company said it will participate in developing natural-language programs that can quickly scan databases and documents like analyst and annual reports to help money managers analyze key financial information.

Current LFE sponsors include Credit Suisse First Boston, Gifford Fong Associates, Investment Technology Group, Merrill Lynch & Co. Inc. (MER.N: Quote, Profile, Research) and Reuters Holdings PLC. (RTR.L: Quote, Profile, Research) .

Kurzweil predicts translation market

Ray Kurzweil, the U.S. technologist and self-styled visionary, spoke to the military the other day of upcoming
technology developments:

Another example is the development of instantaneous language translation devices, which Kurzweil predicted will be common on cellular telephones by the end of the decade. “Within a few years, we will be able to talk to anyone, regardless of language,” he said.

There are various speech translation projects in the pipeline. Who will be the first to bundle such a system into a commercial mobile phone service?