Intelligent Input

Back to Contents of Issue: March 2001

Typing messages on tiny cellphone handsets is ... painful. We look at ventures racing to provide the best AI solutions for the Japanese language.
Note: This article contains Japanese Characters.

by Paul Murphy

Venture Spotlight
Company Tegic Communications
Director (Japan) Ray Tsuchiyama
Location Seattle
Phone +813-5326-3356 (Tokyo branch)
Ownership Wholly owned by AOL
Founded 1995
Employees 6 (in Japan); 100 (all offices)
Products T9 intelligent input system; Advanced IM software
Partners Symbian, Pixo, Optimay, others
IN JAPAN, TO BE told that you push the buttons on your mobile phone like a high school girl is a compliment indeed. It implies a speed and dexterity in handling the minuscule keyboard to send text messages that those of us with trotter-like fingers can only dream about.

Size is important when it comes to mobile phone keypads, with their limited number of buttons and tiny dimensions; many claim these hinder rapid text input. The problem is made more tiresome in some languages. Inputting the word "child" on an English handset takes 12 key presses, while inputting the three syllables of its Japanese counterpart (‚±‚Ç‚à -- kodomo) takes 15. It is in more ways than one a key challenge. The problem apart from fat fingers is that people are lazy. Some studies show that for every extra key a user needs to press to use a function, the possibility of them using that function drops by half.

But a few companies reckon they have solved the convenience problem. Based on artificial intelligence (AI) technology, the three firms highlighted here have developed what they call intelligent text input, allowing people to press fewer than one phone button per letter (or, in Japanese, per syllable). Drawing on a massive word database, phones and other portable devices equipped with intelligent text input can literally guess which word you are trying to type. To take an English example, intelligent input-- enabled handsets need just five key presses to produce the word "hello," as opposed to 13 using the traditional multi-tap entry system. There is now a race on to be the first to bring Japanese-language intelligent input technology for mobile phones to the market, and Slangsoft, Tegic, and Zi are three of the leading players.

Smart, but not too swift. A look at a keyboard shows the difficulty with Japanese. The 1-key selects ‚  (a) and what follows: i, u, e, and o. The 2-key selects ‚© (ka) and ki, ku, ke, and ko, et cetera. On this Nokia phone, the #-key cycles between the two phonetic character sets, English, and numbers, while the "down" key (top) allows for the selection of Chinese characters.
First, however, a note of caution. As intelligent text input in Japanese for mobile phones is not yet on the market, only the companies making the technologies really know how well the software works. It is, says Kevin Williams, an analyst at consultancy IDC, "almost impossible" to verify companies' claims for the software. "The numbers you get from marketing divisions are pretty much useless," he says, adding, "They should be taken with a grain of salt."

One of the competitors is Seattle-based Tegic Communications, which offers the T9 Text Input system. Tegic is also working on instant messaging (IM) that would allow mobile phones and computers to link up. The firm, which opened a Tokyo office in January last year, is something of a split personality. Chizuru Tominaga, a marketing manager, hands out two business cards, one bearing the Tegic logo and one carrying the masthead of America Online, which swallowed up Tegic for an undisclosed sum in 1999. Tegic is now pushing an updated version of Japanese T9 to mobile phone manufacturers after its earlier version was rejected as simply "not good enough," says a candid Tominaga.

One of T9's new features is a customizable dictionary. For example, when Tominaga inputs the syllables of her name in four key presses, T9 searches its database of words and gives tamanegi (‚½‚Ü‚Ë‚¬ -- onion), which is the most frequently used word made up of the syllables produced by the four keys. It takes another few key presses to bring up Tominaga. The new T9 allows Tominaga to replace general word frequency with a personal frequency, ensuring that her mobile phone will in the future come up with her name before the word for onion.

Venture Spotlight
Company Zi Corp.
Acct. Manager (Japan) Mark Shepard
Location Calgary
Phone +813-5776-6061 (Tokyo branch)
Ownership Public (Nasdaq US; Toronto Stock Exchange)
Founded 1987, as Cancom Ventures; renamed Zi Corp. in 1997
Employees 3 (in Japan); 234 (all offices)
Products eZiText intelligent text input; Bluetooth & VoIP solutions
Partners Wavecom, Symbian, Pixo Dejima
The onion example shows why Japan is a laggard when it comes to offering intelligent text input for cellphones or other portable devices. The basic problem is the complexity of the Japanese language. First, there are two types of input -- the phonetic kana and the Chinese ideogram kanji characters. But perhaps more important from an intelligent input viewpoint is that, like Turkish, another latecomer to the T9 stable of 16 languages, Japanese has a large number of possible word combinations. In English T9, when you press the ghi-key (the 4-key) followed by the abc-key (the 2-key), the database can immediately discard words beginning with "hb" or "gc" -- because there are none. Not so with Japanese. Cycle through the syllables ka, ki, ku, ke, ko (by pressing the 2-key), followed by sa, shi, su, se, so (the 3-key), and there are hundreds of possible words that can be generated by the combination of these syllables.

Calgary-based Zi Corporation offers eZiText, which is similar to T9 in most respects, except that it offers word completion -- when typing in "tomorrow" on a handset enabled with eZiText the last three letters of the word will be finished automatically. Zi, which has set up Japanese subsidiary Nippon Zi Corp., claims that it can complete words with an average of four presses compared to the more than seven common on multi-tap entry phones. But not everyone is convinced of the value of word completion, which Tegic is also planning to introduce as a tool. "More times than not the (predicted) word is wrong," says IDC's Williams.

Apart from the in-house tech development teams at Toshiba and other big appliance firms, the most potent threat in Japan to Zi and Tegic is from a little-known Israeli/US startup called Slangsoft. In fact, it is so little known that Tegic's Japan director Ray Tsuchiyama hadn't even heard of the firm. But despite its stealth appearance, Slangsoft aims to make its mark. "Those who already haven't will hear about us very soon," pledges Sara Hirschhorn, Slangsoft's marketing chief in Jerusalem. Though it will not establish a base here until the second quarter of this year, it has been pushing its product through trade shows and directly to Japanese customers. Slangsoft is in talks with a major Japanese appliance maker, says the company, which would be its first big customer in Japan.

While the two North American entries focus on developing intelligent input for mobile phones, Slangsoft has a wider angle that it reckons will help it trump the competition. Its Intelligent Text Input and Display (iTID) technology, which, the firm says, allows text input in a hefty 53 languages, boasts cross-platform functionality, a much sought-after feature. The key word in iTID is display; while Tegic and Zi offer text input technology, mobile phone makers have to go elsewhere to license technology to display the characters. But Slangsoft, says Hirschhorn, offers the two in one package, and with its proprietary rendering system can do so on less than 600 Kbytes of memory. Since footprint size affects the performance of a terminal, the smaller the better. Hirschhorn claims the firm can provide the display fonts for 3,500 kanji characters in only 28 Kbytes. Founded only in 1998, Slangsoft sees itself as the new blood compared to the competition.

Several search engines (including Alta Vista) use Slangsoft's technology to attract non-English speaking users, since it allows input in users' native languages and can also be embedded in chips in Net-enabled mobile phones or other devices. Slangsoft CEO and inventor of iTID, Arie Mazur, stresses that the technology is generic, meaning that it doesn't have to be changed drastically, if at all, for use in other languages. An iTID-enabled phone, says Mazur, could handle many different languages without adding significantly to the software's footprint. This will allow Slangsoft to steal a march on Zi and Tegic, as their clients will have to spend more resources on localizing their products, he says. Hidenori Shimizu of Hewlett Packard Japan, which has just licensed Slangsoft's iTID software, says HP chose Slangsoft over somewhat similar technologies offered by Omron and Just Systems because of its small-size software modules.

But IDC's Williams is still cautious. All the demonstrations of such technologies he has seen so far "have had embarrassing glitches." The problem lies in trying to do too much with the hardware and processing resources available on small platforms, he says. "To try to compact all that intelligence into a small packet is a large problem. The objective is good, but at this stage of the technology you are turning people off."

Venture Spotlight
Company Slangsoft Inc.
CEO Arie Mazur
Location Jerusalem
Phone +972-264-82424
Ownership Private
Founded 1998
Employees 30 (all offices)
Products Intelligent Text Input and Display (iTID)
Partners Hewlett Packard, others
Funding Private venture capital
In seeking business here, though, the fact that Tegic, Zi, and Slangsoft have already won overseas licenses will help. Success abroad is a "phenomenal validation" to Japanese corporate customers, says venture capitalist Bruno Grandsard of Japan Internet Ventures. But a track record in English -- or even Chinese -- intelligent input does not a track record in Japanese make.

Still, the three, and especially Tegic -- which counts Ericsson, NEC, Siemens, and Motorola as customers -- can brandish star-studded client lists from their overseas sales. Zi has also secured a number of attractive business deals and licensing agreements in China, a market that CEO Michael Lobsinger sees as crucial to Zi's fortunes.

To what extent these firms' client lists translate into Japan revenue is not clear. Of the three, only Zi, a listed firm, releases audited earnings data. The three rely on initial licensing fees plus a royalty stream from handset manufacturers, set-top box makers, or other clients who pay a certain amount each time they make a device enabled with one of the firm's technology. This emphasis on royalties means that most recent earnings statements are not the most important indicator of success -- which is just as well. While Zi's sales grew an impressive 96 percent year-on-year in the third quarter of 2000 and its payroll more than quintupled to 234, its operating loss increased by a whopping 232 percent. The firm, however, in a recent extremely upbeat statement, said it was on track for profit in 2001. For Slangsoft, in calendar 2000, when the firm introduced iTID, sales reached around $1 million, not a huge amount of money but about the same as Zi managed in the third quarter of 2000. Tegic offers no hints on sales.

Zi, which is listed on Nasdaq-US and the Toronto Stock Exchange, experienced the loving and leaving that investors bestowed upon many New Economy issues in 2000. The Calgary firm saw its Nasdaq share price climb to over $40 in 2000 -- from a low of 82 cents in 1999. But it dropped so suddenly from February onwards that Lobsinger had to issue a letter to investors in April to reassure them and to put an end to "misinformation" about his firm. Such misinformation included chat site postings alleging insider trading and predictions of his firm's demise. The letter, which began with the gravity-laden line: "Ordinarily, I would never write this type of letter, but the events of the past week are not ordinary," failed to work. Zi's share price ended up dropping further to finish 2000 around the $7 mark.

The high-tech bubble bursting has not outwardly affected Tegic, which had been considering an IPO, according to Tominaga. But by coming under the wing of America Online in 1999, Tegic will for the foreseeable future remain safe from fickle investors.

Slangsoft, which is in the process of obtaining second-round funding and does not intend to go public for a few years yet, was more affected by tech bubble burst fallout; spending enthusiasm among its Net-based clients slumped. "When you sell to Web-based application makers or search engines or Hotmail, of course you feel that those guys have to really consider their spending. They are more quiet now," Mazur says. Customers also tend to look for immediate financial benefits from using his product, he says, and adds that many have told him that they were looking long term, but now they're focusing short term -- to satisfy investors. This has rubbed off on Mazur, who says that rather than going public, his primary goal is to become profitable "like an old-economy company."

Tegic's online demo shows how easy it can be.
The energetic Israeli, who developed the iTID technology, is in no hurry. A speaker of five languages, including that of his birth country Russia (from which he emigrated to Israel with his family 16 years ago as a 12-year-old), he wants to see an Internet that is widely accessible to non-English speakers. Creating technology to allow people to input text in their own tongue is not simply a language question; there is a larger ideal involved. "It's about culture, how people think, their mentality, the way the live their life," he says.

Another pure, as opposed to purely profit, motive also runs through Tegic, whose T9 came from co-inventor Cliff Kushler's work to create a technology that would help disabled people who could neither speak nor use a keyboard to input text. Kushler, a Japanese-speaking martial arts expert who researched his PhD at Tokyo University, and colleagues used a sensor with eight LEDs arranged in a circle that would react to eye movements. The LEDs represented an effective eight-key keyboard but raised the problem of how so few keys could handle a regular alphabet. Solving this dilemma provided the seeds for T9. Tegic is still involved in developing technologies for the disabled. "We have not forgotten our roots," says Tsuchiyama.

Though intelligent input technology may be based in worthy humanitarianism, the reality of its marketing is more bruising. Tegic and Zi have been locked in a war in the US over alleged patent violations on their Chinese-language technologies since March 1999. Insults as well as lawsuits have been flying; Tegic says Zi's main eZiText technology is "cumbersome." Zi says Tegic is merely using the courts to try to block Zi's progress in China. Tsuchiyama refuses to comment on the legal conflict, but the markets interpreted the outcome of the most recent skirmish as a victory for the Canadian firm. He says simply that T9 stands on its own merits. "Our product is validated by the market; we have been licensed by the majority of the world's leading handset manufacturers. One of the reasons companies go with us is because they believe we have superior patents."

Serious legal problems apart, the firms have yet to prove their Japanese text input systems on the market. Venture capitalist Grandsard says he would "definitely consider investing" in firms in this area, though he adds he's not sure he would jump at the chance. "The problem with these types of technologies," he says, "is that often they don't work that well." IDC's Williams has similar doubts. "People have gotten pretty proficient (with existing input systems), and they are used to tapping at the keyboard as it is now." He maintains that adding intelligent input systems complicates matters, since the user then has to go and learn a new system.

There is also the risk that someone will come up with effective voice-activated text input. "I think there is that chance," says Yoshinori Oikawa of Japan- and Israel-based Triangle Technologies, which offers support and investment for Internet ventures. "But the issue is how good is the voice-recognition technology." At the moment, technology to recognize natural speech is not very good, he says, but merging it with intelligent text input may be the answer. "Combining the two could be the killer app," says Oikawa.

But voice activation has its own problems, including considerations of privacy. Keypad text input allows people to silently and privately send messages without disturbing coworkers or exposing (the contents of) their message to coworkers or others nearby. Not so with voice-activated input. Mazur rather amusingly pinpoints one of the main weaknesses of voice-activated input -- you have to speak. "Not many would want to announce the date and time of their annual colon exam while standing in an elevator; even fewer would want to listen."

Note: The function "email this page" is currently not supported for this page.