A Conversation with Lexicus’ Ronjon Nag

The handwriting business is entirely about expectations. Managed properly, some applications can bypass the issue entirely and still adequately complete the task. Other applications require a limited amount of free-form user input that is difficult to anticipate simply using lists and menus. For these applications, printed input is adequate and, in fact, may replace an existing form also requiring printing.

However most people, as a rule, do not use a single style of writing when taking notes or recording information. Instead, they use a mix of printing and cursive that changes from one word to the next, and even from one instance of a word to another. People regard this as perfectly normal and for certain types of vertical applications, highly desirable.

As pen-based computing enters the mainstream through truly accessible consumer devices, cursive handwriting recognition will become essential. This is the position in which Lexicus Corporation, developer of the Longhand cursive recognition system, and its CEO Ronjon Nag find themselves. From its roots as a small Palo Alto-based software developer to its present structure as a Motorola Division, Mr. Nag and company have been at the front lines of pen computing from the early days.

An Academic Background

Nag explained: “I received an undergraduate degree in Electrical Engineering at Birmingham and a Ph.D. in Speech Recognition at Cambridge. I then worked for a couple of small companies in the UK.”

Following university, Nag decided that it was time for something completely different from speech recognition, namely management consulting in banking strategy. He recalled: “I did that for a couple of years, and then managed to land this wonderful scholarship to come to the United States. It is called the Harkness scholarship, and it allowed me to spend two years doing whatever I wanted in the States academically.”

Nag continued: “So for the first half I went to MIT, and the second half was at Stanford. During the first half at MIT, I completed my MBA, and I did a little bit of research there in bank check handwriting recognition. And unlike my colleagues at MIT who went on to become investment bankers, venture capitalists, executives, etc., I decided to do a bit more science at Stanford.” Nag joked: “I was one of the people who skewed the numbers to the low side.”

At Stanford, Nag chose to work with one of the leading researchers in neural networks, Professor David Rumelhart. There he met his future partner, Chris Kortge, who was in the process of completing his Ph.D. It was also at Stanford that Nag became engrossed with the problem of cursive handwriting recognition.

Interestingly, by this time in 1991, neural networks had already passed its peak in popularity a couple years back. Nag agreed: “Definitely, it had passed its vogue. But, I’m not a neural networks bigot or any other kind of bigot. If any other kind of technology is there, I want to learn a bit more about it.”

A Classic Silicon Valley Start-up

What do you do when you need a job in Silicon Valley? Well, start a company, of course. Nag recalled: “Chris [Kortge] was finishing his Ph.D. and he needed a job too. So we decided to start a company. At the time we didn’t know whether what we had was sufficient to make an impact in the marketplace.”

Nag recalled that, at first, they were inspired by the intellectual challenge of the project. “The first prototype had a one thousand word vocabulary and we thought it was quite good. We had read in Byte magazine that cursive handwriting recognition was impossible, would take another ten years, and would need a Cray. We thought that this probably wasn’t correct.”

While the original prototype was developed on a Sun workstation, Nag realized that they would create the greatest impact by showing the technology on a regular Windows computer. Nag recounted: “For Christmas 1991, we didn’t have it on the Windows platform. But we wanted to have it on the Windows platform by January, because we wanted to show it to GO [Corporation].”

With the meeting with GO arranged for January, 1992, Nag and Kortge had their work cut out for them. Not unlike many technology demos, the work continued to the very last minute. Nag remembered: “We managed to get the recognizer going at 2 o’clock in the morning for a 9 o’clock morning meeting.”

Being a small company without the funds to get a proper laptop for a mobile demonstration, Nag and his team lugged a full-size 486 PC to GO’s headquarters. At GO, Nag demonstrated the software to Jerry Kaplan, Bill Campbell, and others. “We spent about four hours and, by the end of this, thought that there must be some interest here, to be here four hours and have all this people look at it and try it out.”

People at GO were impressed and, judging by their remarks, was clearly thinking about a marketable version of the technology. Nag recalled: “At a certain point they told us that a thousand words might be a bit of a toy. What they really needed was ten or twenty thousand words. We said that we could work on it, which led us to talk about some arrangement with them. But in the end nothing worked out.”

Another option for Lexicus at that time was to search for some form of outside investor, either a venture capitalist or a less active financial angel. In the end, Nag opted to bootstrap the young Lexicus. He explained: “We would license the recognizer to people, and that would give us enough money to grow the company from license fees.”

However, Lexicus continued to court important industry partners. Nag explained: “We showed it to Microsoft and, at that time, we were very careful not to go into any exclusive arrangement, or get locked into things so early on. Many of these companies, like Microsoft, finally helped us.”

“And unlike my colleagues at MIT who went on to become investment bankers, venture capitalists, executives, etc., I decided to do a bit more science at Stanford… I was one of the people who skewed the numbers to the low side.”

Nag continued: “When they launched Windows for Pen Computing, they gave us some space in their presentation launch. But again, we were fairly careful in those days, and we couldn’t really pursue very much with Microsoft other than just simple promotional and co-promotional activities.”

The “Official” Launch of Pen Computing

Old-timers in this industry will recall that this time, around April, 1992, was almost magical for the industry. With the near simultaneous official launch of Microsoft’s Windows for Pen Computing and GO’s PenPoint operating system, many predicted a very rosy future.

While Windows for Pen offered a relatively familiar (and inexpensive) development environment—something which Lexicus exploited to best advantage—PenPoint seemed to capture the imagination of some developers in a way reminiscent of the early Macintosh.

Nag concurred: “Windows is quite a convenient platform. It was easy to develop on it and didn’t require too much support. But when we came back from the Windows for Pen launch, we thought that we should also do something with PenPoint. So we were able to do something with the PenPoint launch as well, keeping most people in the industry moderately happy.”

At this point, Lexicus still consisted of the two original founders. However, within a few months, the team ballooned to three members. More significantly, important companies such as AT&T were expressing active interest in Lexicus’ technology. In fact, Nag explained that several companies began taking another look at Lexicus. “We used to often go to trade shows, with companies letting us exhibit for free because many of them needed to address the handwriting recognition issue.”

Geography also helped. Nag recalled: “Pretty soon, everybody found out about Lexicus. And what’s interesting about being in Silicon Valley is that you can really cut down on your travel time, because everyone’s in the area within a 10-15 minutes drive. And, if they’re not in the area, they visit the area and stop by with you at the same time.”

Nag summarized: “Pretty soon, anybody in this category that did any kind of due diligence would end up finding out about Lexicus.”

California Suite

In early 1992, Lexicus moved into a small suite in downtown Palo Alto, just one block from the very popular University Avenue. It’s safe to say that anyone who had the opportunity to visit Lexicus in those days still remembers the experience, with its somewhat uneven floors and near-windowless rooms.

Not too different, Nag nonetheless quickly reminded me that the building was also home to the Palo Alto Chamber of Commerce. At any rate, being in the middle of Palo Alto had many advantages. Nag described that: “it was quite useful to be near Stanford. We had all our buddies over there, and there were lots of facilities—things like photocopy shops—in the area.”

“In addition, it was actually cheap while most downtown office space is expensive. So it didn’t look as nice as what we’d normally like, but we decided that it would be appropriate for us.”

Several companies, including IBM, Sharp, Motorola, and Ricoh, paid a visit to Lexicus’ Palo Alto office with varying expectations. “I think that the American companies, they just sort of expected it because we’re in Silicon Valley. And the Japanese companies just said ‘Oh, so this is what Silicon Valley is like.'”

Nag elaborated: “Some of the local Japanese, and the American employees of some of the Japanese companies based here in Silicon Valley, often got visiting Japanese executives to see us when they visited General Magic and EO. And they’d say ‘well this is another type of start-up in Silicon Valley.'”

Whatever Lexicus lacked in office space, they made up for with strong, leading-edge technology. Nag remembered: “We were in the lucky position of picking up a fair amount of PR [public relations], where journalists would just call up. We did some proactive work, but not that much compared to most companies.”

He continued: “This was possible because there were only a couple of companies that provided commercial cursive handwriting recognition software, and we were one of them. So any company looking in this area would have to look at us.”

Motorola’s Mobile Vision

Several companies have taken a strong position on the mobile market, however few have made it as strategic to their future as Motorola. Nag described the situation at the time: “Motorola came over early in 1992. At that time, we used to sell our beta copies. In fact, in the early days, the more bugs we had the more we could charge.”

Nag explained: “The thinking was that people were really paying for the opportunity of getting a very early look. And the earlier it was, the more bugs it had and the more support it required.”

“In the early days we used to charge something in the order of $10,000 for two floppy disk, and in that way, be able to get our name out. We got a few customers like that. That, of course, gave you a license for twenty users or so. Later on, we dropped it so you could buy for just one user if you liked, for $500.”

Over time, Lexicus received visitors from no less than five Motorola divisions. However, at the time, AT&T was still the company’s main propounder. Nag stated: “AT&T decided to launch the Hobbit at COMDEX, and they wanted to show something complex running on the chip. So they wanted to evangelize our recognizer running on the Hobbit.”

Nag expanded: “Actually they wanted to evangelize the concept of personal communicators, and why you would need a Hobbit. It was supposed to be a much faster processor, and one of their arguments was that you could do things like cursive handwriting recognition. Of course, we could do it on a 386 or 486 as well, but that’s beside the point.”

Though Nag has seen several mobile computing companies either fail or suffer a dishonorable discharge from the industry, he credits this pioneering spirit for some the success of other companies such as Lexicus. “One of the reasons we were able to enter this category is that we didn’t have to invent everything. We just had to do the recognizer.”

“You had all these companies—people at GO, Microsoft, EO, and IBM—all working on the hardware and software, together with a whole range of ISVs. You didn’t have to invent the whole caboodle to make these things useful. So we were able to focus on our small part, small but incredibly difficult part.”

Nag is also convinced of the importance of Lexicus’ technology and the role it will play in the future. He noted: “Handwriting recognition will be on the front line. It will be part of a whole genre of keyboard-less input mechanisms. People will continue to demand it no matter how much manufacturers try to play it down, and continue to demand that it gets better and better.”

A Visit from Apple

The time around 1992 and 1993 also coincided with the development of Apple’s Newton PDA. Nag claims they had no intention of showing their technology to Apple, however, they received a visit from the Cupertino giant nonetheless. Nag recalled: “I don’t know what stage they were at actually, but my impression was that they had been working with ParaGraph for some time.”

Nag explained the visit as follows: “Large companies, in my experience, cannot ignore new technologies, no matter how small a company is. If you’re early enough in the development cycle, you can’t ignore investigating as part of a due diligence factor. If you want the right technology in your product, then you have to keep looking.”

He continued: “A lot of companies often have processes to catch these technologies as they come about to see how they can be used in their companies. And I guess we were in part of that process, but in the end it didn’t work out. One reason is that we were very early in our development. I think our competitor had been working on it for many years. They had many more resources.”

Nag had seen a demonstration of ParaGraph’s handwriting system at a trade show and was familiar with the company through the press. But he still didn’t know very much about his Russian competitor. Nag recalled: “After we had shown our technology in the Microsoft booth, we bumped into Stepan [Pachikov] from ParaGraph, a very nice chap. I think he said that he was quite pleased that we came on board. It’s quite hard if you’re the only company with nobody to compare your technology with. And he’s probably right. Even with only two of us, it is quite difficult to do a comparison.”

“The first prototype had a one thousand word vocabulary and we thought it was quite good. We had read in Byte magazine that cursive handwriting recognition was impossible, would take another ten years, and would need a Cray. We thought that this probably wasn’t correct.”

With little hard information to go on, Nag was nonetheless confident in what they had. In Nag’s estimation: “At that point, I thought ours was more accurate. That was my instinct and my feeling based on a small number of words. I wouldn’t ever say that as a scientist, but ours was definitely faster. However with the accuracy statement, I’m afraid I have no scientific basis.”

Nag expressed only slight disappointment about the meetings with Apple noting that: “it was an uphill battle because ParaGraph had already been ingrained in that institution.”

EO, GOing, Gone

With Lexicus’ competition firmly in the Apple camp, Nag directed his company to focus on AT&T and EO’s PenPoint platform. However, before either company was able to make any significant end roads with the new platform, AT&T pulled the plug.

Nag approached the issue philosophically. “The way I view it is that early machines obviously are going to have many more problems. I think EO had to solve three or four significant problems in computer science: power management, handwriting recognition, and sending data through cellular phones. They weren’t necessarily going to do each of these themselves, but they were expecting to somehow put these things together.”

Continuing, Nag noted: “It’s pretty tough to do that and I think that certainly people who develop similar products today don’t expect their first versions to be major hits. They serve really more like learning grounds, a learning ground for the next version. I think a lot of people learned a lot of lessons from that. Unfortunately so have other companies that are no longer here now.”

The timing of EO’s collapse had little direct effect on Lexicus. Nag recalled that: “By that time, Motorola was in heavy negotiations with us. However, I think it’s not good to have people exiting from our point of view. It’s one less platform that we could have been on.”

“Also, it is one less company that is trying to think about all these problems. You just physically need a few hundred people thinking about these problems. They may not all be in the same company, but you need to have them thinking about how to make the mobile computing paradigm work.”

Motorola Makes its Move

After having entertained several visits from various divisions within Motorola, Nag received a call around August, 1993, from the company’s corporate office asking to arrange for another visit. Nag wasn’t sure what to expect but went ahead and completed another presentation.

Nag remembered: “They immediately took a liking to us and about six weeks later, perhaps a little longer, we were able to negotiate an agreement. We talked about either an investment or acquisition and, in the end, Motorola decided to make a full-fledged acquisition, lock, stock and barrel. The rationale, and one of the things that there were only a few companies that seemed to realize the potential of these kinds of user interface technologies.

“From Motorola’s point of view, as it moves into two-way messaging and makes lots of small devices, how do you get information into these boxes? These devices will naturally lend themselves to keyboard-less technologies.”

Nag was also pleased with the structure of the arrangement. “Rather than just having us assimilated into the research group, the intention was to keep us as a separate business unit. We have marketing people and development people who are thinking about how to make new products on a day to day basis.”

For Nag, the important consideration is making sure that the technology leaves the lab and enters the marketplace in a timely manner. He noted: “There’s so much technology lying around in a company like Motorola, the real challenge is actually getting it out into products. They thought that it would be a shame to squash the entrepreneurial nature of our company and so we were set up as a business unit within the messaging sector.”

Serving Vertical Markets

While the common wisdom is that cursive handwriting recognition is perhaps the most horizontal of applications for pen-based computers, Lexicus has taken a more realistic approach as the new technology strives for maturity. “We think that our current technology, at the vertical market level, works very well—an excellent component to have an application wrapped around it.”

Nag continued: “We’re not saying it’s there for Granny to use yet, an so that’s a niche. However, if you can work with your toughest customer in an emerging market, it’s the best thing you can possibly do—one that keeps beating up on you. We look for those people that beat up on us, because they give us incredible feedback.”

“You have to learn from the marketplace, but at the same time, you want to meet the needs of a segment, a segment at a time, and eventually you’ll reach the wide masses. And if you look at the PC industry, it followed a very similar path.”

With respect to Lexicus’ current product Longhand, Nag describes it as being “positioned so that you can wrap an application around it. But it’s basically a stand-alone program that will work in Word for Windows, Write, and all those other applications.”

Life With Motorola

Being a division of Motorola presents Lexicus with some interesting opportunities. Not only is the parent a licensee of both the Newton and General Magic technologies, but it has been particularly aggressive in bringing products to market with the Marco and Envoy respectively. For Lexicus and its cursive recognizer technology, Nag characterizes development issues in the following ways: “One is processing power, another is the operating system architecture. Does the operating system actually allow you to port a recognizer to that system.”

With respect to the Newton, Nag feels that it would be difficult, if not impossible to port Longhand to the platform under the current architecture. He noted: “There are some memory issues, some issues about the existing technology already in there, and there are no publicly available APIs.” However, since Motorola is a licensee of the operating system, Nag confirmed that “the intention is to move forward and attempt to try to have a recognizer system for the Newton platform.”

There are other issues in moving the cursive technology to a new platform. According to Nag, “the core recognizer, on its own, is about a 100 KB of code and 250 KB of dictionary. That’s variable, it can be smaller if you have less than 25,000 words in the dictionary. In addition, you can have as much as you want for the user interface. On the Windows version, we use quite a lot of UI tricks and things which brings it up to about a megabyte or so.”

Nag also described the other factors that determine the success of the recognizer including the processing power, architecture, and digitizer. According to Nag, “the Newton falls short on architecture and tools. The processor’s ok for us.”

With respect to the Envoy, based on General Magic’s Magic Cap software, Nag sees a different sets of challenges. “I think the digitizing capability on the Magic Link leaves something to be desired from a handwriting recognition point of view at least.”

“Obviously, if you don’t sample enough and you lose data, it’s a bit difficult to get that data back. In a way, it sort of forces us to think about how to come up with algorithms that can deal with rougher kind of input on these platforms.”

Nag continued: “With General Magic, the tools and the architecture are not an issue really. The General Magic architecture is fine, well reasonably fine. It’s satisfactory for our purposes.” In addition, Nag is pleased with the level of support they have received from General Magic, as well as the quality of the tools and documentation.

However, he offered this caveat: “We’ve seen the PenPoint stuff and you see a very similar story with General Magic. There’s a period of time where there is instability, and where they transition to greater stability. Anybody who has developed with a first generation operating system knows that it just goes with the territory.”

“In the long term, only the high-end players survive. And in the long term, the high-end players will be able to do the low-end stuff as well. Anybody doing the low-end will be commoditized out of existence.”

The Power of Cursive

Nag is convinced that cursive handwriting is important for at least two reasons: customers will ultimately expect it, and it helps differentiate the company. “From a marketing point of view, it’s very important. There aren’t many people who can offer cursive, however, there are quite a few people who offer print or variants of print recognition. Basically, we will always do high-end technology because it’s much easier to survive if you’ve got good technology.”

He continued: “Fighting it out with one competitor is very different than fighting it out with seven to ten competitors, in terms of mind share and marketplace. So, you may be forced to do some low-end stuff, but you have to have the high-end stuff. In the long term, only the high-end players survive. And in the long term, the high-end players will be able to do the low-end stuff as well. Anybody doing the low-end will be commoditized out of existence.”

So how does Nag rate Palm Computing’s Graffiti: a low-end product or a brilliant marketing idea? Nag’s reply: “Well whatever it is, their barriers are very low. So, if it’s a great success, then we would expect, looking at the economics, that the margins will be driven down to quite a very small amount. This is because there may be a number of companies that enter that marketplace.”

At the end of the day, Nag is confident that cursive handwriting holds great promise, and may even serve as the bridge between the current generation and the much-vaulted, horizontal, consumer market. “In my view, at any given level of performance of technology, there’s somebody who will find a utility for it. And the trick is finding those early niches which can make use of the technology and attacking that niche with zeal, using the feedback from that niche to take it to the next generation.”