Speaking With ART’s Daniel Wald

Stemming from its tradition of higher education and business acumen, Israel is at home with a large amount of the world’s advanced technologies, handwriting recognition included. And with offices in Tel Aviv and Cupertino, California, Advanced Recognition Technologies (ART) is one of many companies that has turned military technology into consumer and business-oriented solutions.

ART has developed two important products that it licenses directly to OEMs: smARTwriter, its handwriting recognition technology; and smARTspeak, a compact and highly efficient voice recognition system.

Recently, I had the opportunity to speak with Daniel Wald, General Manager of ART at the company’s Cupertino office, just around the corner from Apple Computer. Our conversation touched on a number of topics ranging from technology issues, to new platforms, to the growth of the industry. Among other things, Wald brings a refreshingly international perspective to the business of mobile computing and advanced pattern recognition.

“The first technology that we developed was handwriting recognition back in the early days of pen computing, an bout which you could probably write a book.”

Small Company Challenges

Wald received his Bachelors and Masters degree in Electrical Engineering from Syracuse University in upstate New York, joining Eastman Kodak shortly after college. Following Kodak, Wald held positions at Xerox and Memorex Corporation. With all this big company experience, Wald was ready to move on to something smaller in scale. He began: “In a big corporation, you have a lot of resources and sometimes you can get by with being just adequate. In a small company, you really need to be crafty. You have to do stuff that you never thought would be part of the job.”

Wald identified the time somewhere around his stay at Memorex when he discovered his interest in managing the business aspects of the technology. He noted: “I moved away from the electrons and more into the business side and management.” He added: “I mean there are so many smart people with so many new ideas. Some of them are practical while others are more like ‘wow, I can’t believe you did this. Now what do you do with it?’ I’m more interested with how you create something practical, solve a problem for someone, and make a business a success.”

Wald applied this principle directly to his current challenge of making sophisticated pattern recognition meaningful in the context of everyday life. He explained: “Some kind of shape or pattern recognition, such as handwriting and voice, is useful for the purpose of allowing humans to interface with small devices. Especially in the case of mobile devices, they are getting smaller and smaller and they don’t have all the facilities of a large desktop machine. Therefore you need to have alternative inputs. Also, for some kind of devices you won’t necessarily have your hands available, so voice is handy.”

Originally from Israel but in the United States for the past 25 years, Wald also sees part of his role as acting like a bridge between the two cultures and business styles. He expanded: “Having been in the US for 25 years or so, I can understand the mentality and the mind-set, along with the language. A bridge between the two cultures is really handy, because otherwise, there can be some gap there.”

Wald leverages this experience when it comes to dealing with some of ART’s customers. He revealed: “We license our technology to OEMs, and several OEMs do a lot of their manufacturing somewhere else. These are people from other cultures and they have different ways and experiences. Having dealt with this in the past, through multiple technologies, has been invaluable.”

A Sleeping Beauty?

Wald has faced other challenges as well, not the least of which has come with the relative inactivity of the early pen computing industry. He noted: “The first technology that we developed was handwriting recognition back in the early days of pen computing, an experience about which you could probably write a book. Because of various factors in the market, it went to sleep for a while. The opportunity that presented was to use the expertise of the company to build up a business as opposed to going to sleep with [the pen computing industry].”

Wald is nonetheless optimistic. He observed: “Usually when an industry goes to sleep, it’s a matter of two, three, or four years, at which point people learn from their mistakes, whether these are technology mistakes, marketing mistakes, expectation mistakes, or all of the above. Then they come back with better product introductions.”

Have things begun to turn around? Wald believes so: “I think that we’re seeing it now on handheld devices in general, and pen related devices in particular. But ART has used this time to develop another technology, speech recognition, which draws from the signal processing expertise of the core people in the company. In fact speech recognition, in general, is easier. It is more of a single dimensional effort, as opposed to handwriting recognition, which uses at least two dimensions to determine the x and y position of the pen. In addition, some people check other parameters including pressure, timing, and so forth.”

Wald continued: “So we developed a speech recognition capability which is very niche oriented. We do not do speech-to-text and we do not do dictation. Instead we are very focused on providing a personal solution, personal meaning for an individual. So we do command and control functions such as voice dialing in a cellular telephone. Our recognizer is so compact and small in terms of code size and computing requirements that it can fit in a small cheap device. It does not require heavy resources, neither in terms of memory or in processing.”

Wald summarized: “So we are very niche oriented. We can’t do everything for everyone, but what we do, we do in a highly accurate manner, and we try to package it with a very clever human interface so that it allows the technology to shine.”

A Smart Application

While handheld computers and PDAs may someday benefit from ART’s speech technology, Wald sees the more immediate candidates being the new generation of so-called “smart” cellular phones. Wald noted: “In those devices, the application that we push the hardest is voice dialing. And not necessarily a hundred numbers. What’s wrong with three numbers? What’s wrong with five numbers? A lot of people think that a hundred numbers is great. But I’ve read that even with phone systems that can store a hundred number locations, generally no more than seven are used by the average person.”

“A lot of the success of a technology depends on the psychological acceptance of the user. It has nothing to do with how the algorithm works.”

Wald elaborated: “A lot of the success of a technology depends on the psychological acceptance of the user. It has nothing to do with how the algorithm works. The technology companies that know that but fail to include it in their planning or thinking might miss in a big way. The average user does not like complicated procedures. The technical users, on the other hand, are always more tolerable, but they’re not necessarily our audience.”

“So the general idea is to fit into these small, smart devices. Conceptually, it’s a merger made in heaven. Now, on the other hand, I can’t compete with IBM’s dictation software—we’re simply not there. Instead this is a niche implementation. Since it’s user dependent, meaning you train it to your vocabulary, you don’t get a existing vocabulary or some type of library. You create it. Also, the practical vocabulary size will be somewhere around a few dozen words, perhaps a maximum of a hundred or two hundred words. If you need to have a thousand words or more, then don’t use this technology.”

Some Tech Talk

Turning to the question of technology, Wald took a few moments to explain the essential elements of the company’s smARTwriter recognizer. He began: “We are not character-based. We are not dictionary-based. We have a shape recognizer technology. That’s the major difference between us and everyone else. What does this mean? In simple English it means that you can draw a shape with a pen and it could be similar to a character in a language but it doesn’t have to be.”

“We are not character-based. We are not dictionary-based. We have a shape recognizer technology. That’s the major difference between us and everyone else.”

“It could be a smiling face, it could be a house with a roof, without a roof, multi-stroke, as many strokes as you like. I can train that shape to be a word or a character. The unique thing about this is that there are people all over the globe that have different writing habits, even within the same language. So the fact that I can recognize an arbitrary shape—I have a mathematical algorithm that can trace and recognize any shape that you can draw—gives me a much better capability to penetrate many applications and markets.”

Delving a bit deeper, Wald explained: “Our software looks like a multi-layer onion. The core of the onion is the shape recognition engine that I just explained. The exterior layers, which are not as heavy as the recognition engine are custom-tailored to the language and the application. The software is fairly smart, but consequently, we have the ability to add some unique characters like umlauts and epsilon for European characters and Greek respectively.”

“Obviously we can do the unistroke type alphabet, and we have our own unistroke model which is very flexible so people do not have to learn a new character set. Or we can also handle the popular Graffiti-type stuff because all of those are a subset of a shape recognizer.”

Being shape-based means that ART’s technology is adaptable to a number of different languages. However, adding more information to the recognizer makes it perform that much better. Wald explained: “In our case, the core recognition engine is the same for any language. We do have versions of our smARTwriter that works with the Japanese Kanji or Katakana just to show that we have the capability. However, we have not developed all languages. I mean, you go where you have business.”

“But we do have a model technology that’s called a linguistic layer which tries to get the accuracy higher in a particular language using probability weights. By going off-line and scanning thousands of pages of text in a language, you can look for repetitive combinations and assign them probability weights. We can then add that as an appendix to our smARTwriter. As you write, if the recognizer determines a few probability scores for a particular combination, the linguistic layer can preclude the ones that do not exist in the statistical database.”

Wald confirmed that ART has created such a linguistic layer for English and that it is available as part of the licensing agreement for the smARTwriter product. Wald stated: “It’ s an option that you can turn on or off through a pull-down menu, and statistically, it has increased our accuracy by between one and a half and two percentage points.”

Pen-Based Handhelds

While generally positive about stylus-equipped handhelds, Wald was also quick to point out that few have lived up to their sales expectations. He insisted: “If you look at the market, there is nothing that hit in a very big way. So there’s no machinery with a pen that sells in astonishing quantities, say in the millions of units.”

Continuing, he noted: “Of course, that’s relative to today— if something sold in a hundred thousand quantity and it had a pen with it, that would be a big coup. Having said that, at ART we try to focus on various platforms that seem to have something going for them. We created versions of smARTwriter that runs under Geoworks; we’ve created a version of smARTspeak on Magic Cap and the Sony Magic Link.”

“We have made a strong commitment and have ported our technology to Windows CE.”

“But most recently, we have turned a lot of the attention of the company to Windows CE because Microsoft is behind it and it’s a big name. We have made a strong commitment and have ported our technology to Windows CE. Some half a dozen big name companies are building Win CE devices and so, at ART, we think that it has a chance to get to a higher plateau than previous pen-based PDAs.”

To illustrate the sensitivity of the market today to past lessons, Wald noted: “People are careful not to say the wrong word. You say pen computer and that’s bad. The word PDA is bad. So I think there’s some psychology to the marketing of it, but I believe that this will bring to light the whole concept of small computers shuttling data back and forth from large computers.

On the topic of smart phones, Wald observed: “There is a whole set of new devices that have a screen on them and consequently, some of them have pens. But some of them don’t such as the Nokia 9000 Communicator. You never know. We have a different opinion obviously because we sell software that works with a pen. So I hate to see that there is no pen available.”

“On the other hand, I think that people are kind of cautious as to how they bring in new capabilities. Some want to bring it in a little at a time. I know of other companies that are larger than Nokia that are working on devices with pens, and they’ve been playing with different recognition software. So a lot of people in the industry are looking at bringing additional capabilities and features to the telephone.”

The Pen Has its Place

Joining his colleagues in the handwriting recognition field, Wald echoed the conclusion that the technology was poorly marketed from the start. He observed: “Perhaps the expectations were too high. If somebody pays, say $50, for a device and gets totally frustrated, the person will tend to forget about it. It won’t leave a big scar. But when you pay somewhere close to a thousand bucks, give or take, people don’t like to feel that they bought something that’s a boat anchor. And that’s what I think takes longer to recover.”

“People bought the early Newton thinking that they’ll do a lot of nice things and Newton, in my personal opinion, is a wonderful technology. But maybe their expectations were set was too high and they were disappointed. Today’s version of the Newton is remarkably improved in everything that it does, but it scarred the industry quite heavily.”

On the positive side, Wald noted that things do appear to be turning around. He commented: “I think that we are now at somewhat of a different stage. It seems that people are willing to take another look. I don’t think it’s like night and day—it’s a gradual change. But still, my personal opinion is that, today, the notion of someone writing a letter to their grandmother on a touch screen is not here.”

“I do think, however, that it has become very practical to go on small devices and implement an application with handwriting that will allow you to take some quick notes, or maybe go on a pager and write, ‘my flight is late, will be at the office at 11am, I’m on flight # something’, hit the send button and off it goes. Those things do not take a lot of effort and there are many feasible approaches.”

Continuing, he noted: “I think it’s important to realize that people have an affinity to gadgets and to smaller machines that are easier to handle and are not so heavy. And for them I think there is definitely renewed life to the pen. But I think that there is a higher probability for acceptance and success if one comes up with a practical use for the pen as opposed to a comprehensive replacement of the keyboard. You see that more and more vendors are coming out with machines that have both. So they’re trying to give the user an alternative which I think is a very clever way do to things—not to force the user to do things a certain way.”

Referring back to ART’s smARTspeak technology, Wald added: “And if the device has a microphone, give the user the option to use voice as a way to navigate and perhaps to perform shortcuts by using a voice command as opposed to performing manual operations.”

Where Do We Go From Here?

Notwithstanding the challenges faced by any small company in getting their technology noticed, Wald sees a lot of near-term potential in the company’s technologies and vision. But Wald was quick to note that hardware vendors also face the challenge of having to be equally innovative in their design and marketing. In addition, Wald sees much of the early action taking place in the verticals.

He observed: “I think that we are not quite there in terms of some of the machines that have a pen. For one, they do not have a very comfortable arrangement for the user to write on. Sometimes you find that you hold the pen but the palm of your hand or knuckles are pushing buttons accidentally. HP made an effort to solve these problems with the OmniGo, which flips 180-degrees and puts the keyboard on the other side. Those ergonomic-related things will make it more user friendly for handwriting.”

“I also think that honing in on vertical solutions is what is going to lift some of these segments up. There’s a big push in the medical field, for both data collection or prescription management. There are also some approaches that have been tested to collect the ink and do recognition, but not in real time. But I think that still none of the machines on the market today have all the attributes necessary to completely replace an 8 1/2 x 11″ pad and a pencil.”

On the larger consumer scale, Wald sees another challenge. “With between 130-150 million computer users in this country who all use keyboards, it seems obvious that the use of a pen to input into your computer is something specialized and may first appeal to some special class of people, but not necessarily to every class of users.”

On the issue of cursive versus printed writing, Wald sees plenty of opportunity for a fast, small, and efficient print recognizer. Plus, Wald added, some people like himself feel more comfortable with print: “It’s a matter of habit and I could envision that it may be very important to some people and it may be less important to others. Personally, I print and this is before I came to ART. So this is not an answer made to order. But I print because I found that I get pretty frustrated when I cannot read somebody’s handwriting. And I figure, if I print individual characters, then it makes my handwriting 100% readable.”

As for connecting to the Internet or corporate intranets, Wald sees that as being some time away. He concluded: “I think that connectivity of pen computers or wireless computers to the Internet and intranet is great, but I don’t necessarily see that it will provide a medium or vehicle to elevate the use of the pen much more than it is already being used.”

“Where you see the pen being used is where it’s advantageous—for somebody who performs data collection, inventory management, and so on. And if they need hands-free operation, they use speech.”

“Where you see the pen being used is where it’s advantageous—for somebody who performs data collection, inventory management, and so on. And if they need hands-free operation, they use speech. But if they need something that is extremely small and they can’t type, then they use the pen.”

Daniel Wald
Executive VP & General Manager
Advanced Recognition Technologies, Inc.
20380 Town Center Lane, Suite 175
Cupertino, CA 95014
1-408-973-9786
1-408-973-9787 (fax)
[email protected]