WordPerfect Gets Grammatik(al)

First WordPerfect licensed a grammar checker – then it bought the company which developed it.

When WordPerfect acquired Reference Software, best known for its writing tool Grammatik, in early 1993, it was not immediately clear whether the latter would continue to operate autonomously or be integrated within the huge Utah company. Some months later, we now have the answer. In name, Reference is no more; its San Francisco offces have been closed and its employees now work for WordPerfect. Co-founder and president Don Emery has been appointed vice president of marketing solutions; Bruce Wampler, the father of Grammatik, will continue work on Grammatik after a six-month leave. And Reference‘s former vice president of r&d, Rudy Montigny, is now vice president of technical development. Home base for Montigny is again Antwerp.

The first indication that the two companies were beginning to work closely together came last fall with the announcement that WordPerfect had licensed Grammatik for incorporation in its popular wordprocessing software. By subsequently acquiring Reference outright for US$19 million, the Utah company gains exclusive access to all of Reference’s technology, most notably the foreign language versions of Grammatik currently under development. While WordPerfect will continue to honor Reference’s existing oem deals, it will not offer the German Grammatik to its competitors. WordPerfect’s share of the German wordprocessing market lags behind that of Microsoft Word and homegrown package Star. In the ongoing marketing wars, WordPerfect will soon have one more item on the almighty feature checklist when the German Grammatik is released later this year.
    Several years ago, Reference people correctly predicted that grammar checkers would be the next major enhancement to wordprocessing software and, indeed, the more flexible memory architecture of Windows has now made this a reality. This naturally meant a diminishing market for standalone grammar checking packages, once a popular wordprocessor add-on, particulary in America, as both Reference and competitors Que Software and Houghton Mifflin can attest. “We began to plan other retail products,” says Montigny, “as well as direct more attention to the oem market.” This meant a complete range of linguistic resources, from spelling and hyphenation wordlists to morphology analyzers and parsers.

“I didn’t think that there was a market left for spellchecker and hyphenation wordlists. We therefore didn’t give these things a priority,” recalls Montigny. “I was wrong. Software developers began telling us they wanted to be able to have just one single universal API for language-specific functions. And they wanted to get their spelling, hyphenation, and grammar checking resources from one source, ideally for all the languages they want to support.” Reference did not offer spellchecking and hyphenation wordlists for languages other than English. Chief competitor Houghton Mifflin does, but in turn does not yet have foreign language versions of CorrectGrammar, its grammar checking package.
    As a consequence, a major push got underway at Reference to build up “wordbases” for all of the major European languages and important regional variants, such as Brazilian Portuguese, Canadian French, and British English. However, Reference did not find it easy to make the transition from a retail marketing-oriented organization to a development-driven one. Although Reference could boast total Grammatik sales of over two million, with 300,000 registered users, and had lucrative bundling deals generating a steady income, the sales were offset by the high r&d costs (some 30% of its annual revenues) it required to penetrate the oem business. WordPerfect brought relief.
    “Joining WordPerfect provided us with the resources to go immediately full steam ahead with Dutch, Italian, and Spanish versions of Grammatik,” says Montigny. At the former Reference (now WordPerfect) offices in Antwerp, a team of linguists has been working since 1989 on foreign language versions of Grammatik. A French version was launched in 1990; a German version will be ready later this year. Dutch, Italian, and Spanish versions should be ready in the course of next year. Under the current arrangements, the Antwerp team has only to concern itself with language specific issues; the underlying program architecture and user interface is the responsibility of the twenty strong development team in New Mexico, which also develops the English version of Grammatik. Montigny, who for the past few years spent on average six out of eight weeks in Reference headquarters in San Francisco, will again be spending most of his time in Antwerp. He reports directly to WordPerfect’s senior vice president of R&D, Dave Moon, in Orem, not to WordPerfect Europe in Rotterdam, WordPerfect’s European marketing, sales, and support organization.
    For Montigny, being part of the WordPerfect organization means above all access to tremendous human resources, particulary with regard to multilingual products. “There are seventy localization specialists in Orem,” says a clearly impressed Montigny. “Moreover, I also have access to all the local affiliates here in Europe.” He says that when WordPerfect Germany first evaluated the new German Grammatik under development, they “tore it apart.” Henceforth, Montigny wants the local WordPerfect affiliates involved “from the very beginning,” building rule bases, compiling help files, and writing or translating manuals. Montigny is equally aware of the Mormon Church’s significant legacy in computational linguistics and machine translation in terms of human resources in and around Provo.

Rudy Montigny has been involved in the field of linguistic software for more than a decade, although he came to it by a circuitous route. Having a background in economics, he was involved for many years in office automation projects in the insurance sector, working for cap Gemini among other companies. In 1983, he headed to the us with fellow Belgian Paul Janssen with the hope of marketing a wordprocessor for the Hewlett Packard HP 3000 mini-computer. People, however, appeared a great deal more interested in the program’s hyphenation routines.
    In 1984, Montigny and Janssen therefore established Soft-Art in Marco Island, Florida, with the aim of exploiting this apparently wide-open market. Demonstrating their French and English hyphenation routines on a lowly Radio Shack trs-80, they received numerous orders and requests for additional languages. Before long, customers were ordering spellcheckers from Soft-Art for ten languages – sight unseen. This required some hasty cobbling together of lexical data.
    Soft-Art went on to become one of the major suppliers of linguistic resources for spellchecking and hyphenation, counting Microsoft, WordPerfect, Samna, Wang, Data General, Prime, Data Point, SPC, and Lifetree among its many customers. Montigny sold his share of Soft-Art to Janssen in 1988 and moved his family back to Antwerp. The following year, he was approached by Reference’s president Don Emery – also a Soft-Art customer – who asked him to set up a European distribution network for Reference; this led to the establishment of Reference Software Europe in Antwerp.

Version six of Grammatik is currently under development in Albuquerque and Antwerp; Montigny demonstrates for us some of its new features. “Through version five, we flagged potential errors but we didn’t rewrite sentences,” he explains. “But people kept asking ‘If the program can recognize a mistake, why can’t it fix it too?’ You can tell users that they’ve split an infinitive, but they may not know how to correct the error. Of course, you can look up split infinitive in our extensive online help file, but that’s an extra step. You have to understand, people turn to Grammatik because they are insecure about their grammatical knowledge and writing skills.” Montigny says that sentence rewriting has posed an exceedingly difficult challenge for Reference developers. This is partly due to the kind of analysis Grammatik does and partly due to the general nature of some of its rules. Montigny concedes that not everything will be feasible within the given production schedule. Moreover, specific languages pose specific problems, such as the extensive declensions in German which wreak havoc with efforts to generate text strings having proper agreement. However, the company characteristically does what it can under set circumstances, not allowing itself to become paralyzed by overly high ambitions. In any case, Grammatik’s users seem satisfied with the progress the company is making.
    Those who dismissed Grammatik on the basis of its simple pattern-matching techniques in early versions might be intrigued to know that, since version 4, the package does perform shallow syntactic parsing. Reference Software grammarians have started extending the dictionaries with some basic syntactic information, such as codes for uncountable noun and clause conjoiner, (there are ninety-six such specific attributes). The package includes a booklet, Creating a Custom Style Guide, which details how to build your own dictionaries and write your own rules using a utility program included. Large companies such as Hewlett Packard have tailored Grammatik for proofreading documentation along the lines of in- house style guides. Its configurability may indeed be one of the program’s most attractive features. Montigny knows of a Norwegian company which has compiled some 3,300 rules in Grammatik for Scandavanians writing in English. He recognizes English-as-a-second-language (ESL) packages have an enormous potential market. How much of this kind of development work WordPerfect will be willing to do itself and how much it will contract out to local third parties, remains to be seen.

While Grammatik was Reference Software’s best known package, it has never been its only one, online reference works being another important product line. One of the key technical issues – with economic implications – which Reference grappled with is compression. “When we set out to produce the Random House dictionary, we surveyed the field of data compression specialists. AND Software in Rotterdam appeared to have the best. Even so, we found AND couldn’t get the compression ratios we could. We know, because we tested both our AND their compression on exactly the same data.” The Reference compression routines analyze the specific data involved extremely carefully. Compression for the Random House dictionary took two days on a fast pc but enabled the company to squeeze the 45 mb of data to 9.5 mb, 20% of its original size. The better the compression, the fewer the distribution diskettes – and material costs are vital here at the low-end of the retail software market. Going from two to three diskettes, for example, can cut deeply into a product’s profit margin. However, there are other considerations as well.
    “At Reference, I had pushed for us to publish the entire Random House dictionary, not an abridged version, despite the high material costs involved,” says Montigny. “The decision paid off because the package has been a great success. People appreciate having as much information at their fingertips as possible.” He is sceptical about some of the multilingual translation aids coming onto the market because they aim for breadth, not depth. “People want to look up unusual words, not basic vocabulary. I don’t think that old argument about taking up too much diskspace is valid any more. People would rather devote ten megabytes to really good reference that they can use all the time than two megabytes to something which isn’t very useful. They’ll just end up taking it off their harddisk in the end, anyway.” Reference was also behind the new Collins Electronic Dictionary and Thesaurus, currently distributed on diskette and shortly to be shipping on CD-ROM. While the Random House is aimed at the domestic us market, Collins will naturally appeal to British users. WordPerfect will be marketing them both as add-ons.

WordPerfect is clearly changing. It is evolving from a rather insular company known for one enormously popular but not terribly innovative wordprocessor, to a far-flung organization made up of a diversity of cultures and moving in a variety of directions. With the linguistic resources of Reference, WordPerfect’s software will hopefully be getting more language savvy – the cursor control keys in version 5.1 still don’t recognize word boundaries properly – and more content-oriented. The wordprocessing package’s character set extensions, although not exactly standard, are second to none, but there is room for improvement in other places. Better hyphenation, more sophisticated search and replace, and smarter suggestion algorithms in the spellchecker – these are just some of the eminently feasible improvements which come to mind.
    Montigny, of course, realizes this. “Our morphology analyzers, parsers, wordbases – these can be implemented in a lot of WordPerfect’s software,” he says. “I see a huge oem market for us simply within the company. We’ll be merging technologies where needed to get the best of both worlds. wp has great dictionary chaining technology, for example. We can add morphology.” Montigny says you can expect to see a smart spellchecker for WordPerfect “before the end of the year.”
    “WordPerfect wants to listen more to Europe,” sums up Montigny, “particularly on the development side of things. Previously, development plans reflected primarily American needs. That has to change – Europe is a crucial market.” As WordPerfect’s European spokesman for development, Montigny will be a significant channel of communication, explaining the company’s corporate strategy on this side of the Atlantic as well as maintaining an open ear to European needs. He will no doubt have a substantial ally in Dutchman Ad Rietveld, the founder of WordPerfect Europe and recently appointed wp’s director of international marketing and sales. One thing is certain: WordPerfect is no longer just a Mormon company…