PARADISE: Polyglot Courseware

Start simple, listen to your market, and sell something that works — Belgian developers De WildeBuyck offer a modest lesson in how to market NLP.

Flemish software house De Wilde,Buyck has developed a Computer Aided Language Learning (CALL) package, PARADISE, which exploits some rudimentary NLP techniques to make it easy to develop language learning materials for English, Dutch, or French. PARADISE, stress its makers, is an open learning system, meaning that using the system instructors can compile their own language teaching materials. “What a language instructor reads in the newspaper in the morning,” says Product Development Manager Johan Vandewalle, “he or she can use for teaching purposes in the afternoon.” He points out that it is exceptionally inspiring for students to work with texts from their sphere of interest.
    How does PARADISE work? First, an instructor types or otherwise imports a text into this Dos,based package, such as an article from yesterday’s Financial Times (or for that matter Le Monde or de V olkskrant). PARADISE then scans the text. Using morphological analysis, it tags the gram, matical forms of words, reduces them to their roots, and identifies the part,of,speech. Having inventoried the vocabulary of a given text, the instructor can then develop exercises targeting specific problems, such as specialized vocabulary, verb forms, or pronouns.
    Onscreen, the PARADISE exercises resemble those of their traditional paper counterparts, where you fill in the blanks. No radical new paradigm here. But from the instructor’s point of view, PARADISE makes it possible to partly automate the process of designing materials and to provide supplementary information by simple hyperlinks to entry fields. Moreover, scores can be calculated automatic, ally. From the students perspective, PARADISE provides a rich context for text comprehension: it can offer contexts for unfamiliar words, definitions, and, where needed, translations.
    Some form of local parsing, what Vandewalle calls “context analysis,” to improve part,of,speech tagging was deemed impractical. “Simple errors irritate users,” he says. ”It’s easy enough for an instructor to disambiguate whether a word is a verb or a noun. We stick to things that we can do quickly and accurately.” Among other things, parsing would slow things down too much and make users im, patient; the program needed to be fast, even when running on minimal equipment. Where there is ambiguity regarding part,of,speechassignment, PARADISE prompts the instructor.
    PARADISE makes minimal hardware demands; the exercises themselves can be run individually on floppy, based PCs or printed out. However, if the PC is equipped with a sound card, the program has hooks which allow instructors to linked their own digitized speech files to words to demonstrate pronunciation. Has the company looked into to text,to,speech algorithms for this purpose? ”T ext-to-speech is not yet suitable for language learning,” says Vandewalle. “You really need the human voice.” Digitized speech is one obvious enhancement to the basic system. Are there others? To expand the package’s usefulness, the company is now incorporating electronic reference dictionaries licensed from Antwerp publisher Spectrum. On a more sophisticated operating system like Windows or OS/2, video images might be a future enhancement.
    Christian de Wilde believes that his company has found a perfect niche with PARADISE “It’s neither a closed system nor a full-blown authoring system,” he says. “Today’s authoring systems require some understanding of programming principles. They have a fairly steep learning curve. Not PARADISE. Because it is easy to use, instructors can get straight to work.” De Wilde-Buyck was established five years ago to exploit a language learning package called Adam & Eve which originated at the University of Leuven. The forbearer of PARADISE, Adam & Eve was more limited; it had smaller wordlists and lacked morphology. With PARADISE, De Wilde-Buyck started from a clean slate. The first version of the product was tailormade for the Belgian Banking Association. Called PARADISE Banking, it is used for training new tellers and counter personnel, who need to be proficient in banking terminology in the languages Dutch and French. PARADISE Banking includes 7,000 terms ranging across 125 topics, such as basic day-to-day transactions (”opening an account,” “renting a safe,” etc.) Terminology evolves, points out De Wilde, and materials need to evolve with it. PARADISE Banking can easily be kept up to date. De Wilde-Buyck has also produced a custom version of PARADISE for the Belgian air force which uses it to train its personnel aeronautical English. A third custom PARADISE application is being developed at the behest of the Dutch Ministry of Employment which wants to use the system for improving the marketable skills of young, first-generation immigrants (so-called allochtones), many of whom are unemployed. The Ministry has targeted two domains for this application: metalworking and office procedures.
    De Wilde-Buyck has been involved in EC-funded projects, but De Wilde says it has been a frustrating experience. “For one thing, the administrative and travel costs are enormous — impossible for a small company like ours,” he says. The company’s participation in the DELTA (long-distance training and education) program was less than fruitful. “It was the best DELTA project ever,” says De Wilde. “Our product was very well received because it was simple and effective. But in the long run, we got elbowed out by the big industrial partners.” He adds with dismay, “do you know who gets funded in this program? Berlitz — an American company!”
    Mainstream commercial software publishers have traditionally eschewed educational software for a variety of reasons, leaving a largely untapped market potentially enormous proportions. Says De Wilde, “there are ten thousand schools in Belgium. If we can penetrate ten percent, that’s a thousand sites right there.” The company is by no means focusing solely on the Benelux area; according to De Wilde, it is close to finalizing an agree- ment with IBM for distribution of the package in a number of European companies. Question: which shrewd American publisher with a foot in the educational market will get to Gent first?
    With PARADISE, De Wilde-Buyck offers liberal site licensing arrangements, starting at BFR 30,000 (±US$900) for educational organizations and BFR 100,000 (±US$2,700) for companies, for the first thousand users.

De Wilde-Buyck, Ottergemsesteenweg 439, B-9000 Gent, Belgium; Tel: +32 92210724, Fax: +32 92211865