LISA: The language technology market is changing fast. What will be the critical success factors for tools providers in the next five to ten years?

Jochen Hummel: That’s difficult. One critical success factor is definitely maintaining the growth of the company, because only then can you attract the first-class staff you need to develop the technologies.

LISA: What do you have to do internally to be able to service that sort of growth?

Jochen Hummel: You have to keep up with the technology, and this is another critical issue. Once you are established, you start to devote substantial resources to maintaining your current product line. This means that there may come a time when maintaining the installed base of your current product harms the development of new ones. I think this problem is specific to our industry, since our clients are worldwide and demand worldwide solutions. You need to have a global network of sales and support offices, you need to show worldwide presence, and your technology needs to be able to run on all the different platforms that your clients are using. This is difficult to accomplish for companies that are still fairly small.

LISA: What about your clients? Have they also matured in the past three to five years?

Jochen Hummel: This is definitely the case in the localization industry in Europe, and in some other industries. We have seen our products—terminology databases and translation memory—become standard applications here. However, in some countries, such as the Far East, a lot of pioneering work still needs to be done in education and in selling the concept of computer-aided translation tools. We have to make top management aware of the opportunities and risks involved in documentation, translation and localization, and of the need for tools support. The area still doesn’t get the attention it deserves: very few top managers and consultants care at the moment. This is especially true for applications like Multiterm, our terminology database. Translation memory is simple and saves you money so it’s an easy sell, but it will probably take another few years until the idea of maintaining terminology is a known and accepted concept in larger corporations.

LISA: What are the barriers to this?

Jochen Hummel: The problem with terminology work is that there are so many hidden costs. It is very, very difficult to really figure out how much proper terminology work will save a company, or how much you will waste if you don’t do it properly. The cost is very high, but it’s hidden inside the general complex of enterprise communications. The costs of buying the software aren’t that high, but creating and maintaining the terminology database is very expensive, because companies need to hire people, collect the information, update it, and so on.

LISA: Isn’t it more than a cost issue? Companies are prepared to invest heavily in other large-scale areas, such as just-in-time manufacturing or knowledge databases, to obtain long-term payouts.

Jochen Hummel: Right, but these are concepts for top management. It’s nothing you can sell to the head of a translation department.

LISA: So how should the language industry get the message across to top executives and consulting companies? Isn’t the industry itself part of the problem?

Jochen Hummel: Definitely. Terminologists are kind of dusty—they have to be more innovative and aggressive in making their point. They all complain that they don’t get enough attention, but they have to go out and get it. The Internet will be very helpful here because, for the first time, companies are having to talk to a worldwide clientele, and solving this problem will translate directly into higher revenues.

LISA: So how is TRADOS positioning itself here?

Jochen Hummel: We haven’t put much effort into Multiterm recently, and one of the reasons is that it is much easier to sell translation memory than a terminology database. However, we shall be releasing a new software product and we shall be trying to address top management. One of the vehicles will be to work together with top consulting companies, such as Andersen Consulting, KPMG or PriceWaterhouseCoopers, because they are already talking to this target clientele. Our strategy will be to start in countries where there is a higher appreciation of terminology work, e.g., Switzerland or other countries which are multilingual, get some good cases and then take it from there. In Switzerland we have already done several projects with large companies such as banks, which have invested and will continue to invest quite a lot of money. The main new feature of the product is that it will have a client-server architecture, and will therefore allow intranetbased terminology distribution, querying and updating. This will make it possible for enterprises to obtain and distribute terminology globally. The ultimate goal should be to allow companies to use the terminology when communicating with clients as well. There is no way that even the biggest companies in the world will be able to translate their Web sites into all languages, but at least they could offer terminology in most languages. Then, people with a passive knowledge of English plus the term base could read the Web sites.

LISA: What about other forms of technological integration?

Jochen Hummel: I think that translation memory and machine translation will eventually grow together. The current method is for the translation memory system to serve as a front end. When it cannot find something in its memory, it passes the sentence to an MT system, which provides a draft translation. This is then corrected, and from that point on it is available in the translation memory. This approach already makes a lot of sense, but as retrieval technology develops, translation memory systems may work on smaller units such as phrases or subphrases rather than sentences alone. You could then use some of the technology currently available in MT systems to recycle these phrases. Eventually we might see example-based machine translation systems, which are a sort of mixture between the translation memory and the MT concept.

LISA: When will such systems become available commercially?

Jochen Hummel: That’s very difficult to foresee. The idea of example-based machine translation is not really new, but recently we have seen new developments in statistical algorithms. Also, of course, machines are becoming much more powerful. However, example-based machine translation systems are based on existing materials, so they will only work for specific subject areas, whereas current MT systems try to do everything.

LISA: Will this type of collaboration increase business?

Jochen Hummel: I would say so, since it gives both providers a bigger market. Also, with linguistic programming you are faced with the need to support almost all languages on the market. Since it’s impossible to do all the groundwork yourself, you will always need partners from whom you can license certain components. It then depends on where individual companies see themselves—whether they are component suppliers or applications suppliers—and in some cases on their mood of the day.

LISA: So users will end up with the workbench they’ve been asking for?

Jochen Hummel: Yes, of course. Clients are currently faced with all sorts of tools, and even vendors have different tools that are not always easy to work with, or that don’t interface that well. One of the challenges in the next few years will definitely be to provide a broader workbench offering convenient access to the different tools, and it will probably have to be better interfaced than the current desktop packages. However, this is not a trivial task.

LISA: What effect will this have on industry structures? Will tools vendors buy up whole companies as well as components?

Jochen Hummel: It will have a certain effect. If you decide to become a component supplier, you can stay much smaller and focus on the work you are doing. On the other hand, you lose real estate on the screen, and this is something that is difficult to defend and maintain. My guess is that there will be some major players who will provide the applications and that they will work with many dedicated components, especially in the different linguistic areas.

LISA: Will memory tools and interactive translation become just another button on the office interface?

Jochen Hummel: Certain parts of the technology we are currently offering—even some kind of translation memory technology—could find their way into standard word processors. However, in the case of the more sophisticated tools, my guess is that the translator market is simply too small to warrant integration into a standard office application. What is more, the tools are too complex—indeed, the whole subject is too complex—to be reduced to the level of a simple feature.

LISA: Do you see functionality moving upstream to the authoring level as well?

Jochen Hummel: The next big boost in productivity could definitely be achieved by using terminology databases and translation memory when writing texts for the first time. Again, these ideas are not exactly new, but organizational structures are a major obstacle. Very often, documentation and translation are two completely different departments, and there are simply no communication channels. I think this will change, though, to keep up with shorter cycle times. Workflow components and software support for the entire documentation and translation workflow will soon become a must. Once you have this in place, organizational structures will also need to adapt, and this will clear the way for using translation memory and terminology database technology for authoring as well. Of course, this raises the question of whether these kinds of tools will be part of bigger workflow or authoring systems, or whether translation memory and localization technology will be able to keep the real estate on the screen and simply interface with document management systems or authoring tools.

LISA: What role does controlled language play in this?

Jochen Hummel: To be honest, I really don’t know. So far, I’ve always seen a very strong resistance by authors to using controlled language. It might become more important than it is today, but I think it will always play a minor role in comparison to the other developments we’ve been talking about.

LISA: Why are so many service providers developing their own, proprietary tools?

Jochen Hummel: Probably, the tools they really wanted were not available, and they wanted to customize the technology. Sometimes vendors believe that tools give them a competitive advantage, or use them as a door-opener for their service business. However, all these strategies fail in the end, because if you want to develop software, it’s better to be a software development organization and none of them are. Having a separate software development department isn’t enough nowadays to produce state-of-the-art software. Given the growing complexity of the issues involved, you need to be totally focused.

LISA: The counter-argument would be that, as vendors, they really understand user requirements.

Jochen Hummel: They may understand their own workflow, but there are so many different workflows and so many different needs. The danger is that you will end up with specific tools rather than ones that can be used in a broader context. However, as soon as vendors start to market their tools, as opposed to using them for promoting their services, they will have to respect the needs of a broader clientele. This is very difficult if you’re just a software development department. It’s then that they realize that they won’t really get the revenue they had forecast, or that other tools provide better services to organizations at lower costs. With a small team, you can come up with a nice product—you can even come up with a fairly powerful product, because you can use the newest software technology and you’re not bound by existing needs, but that’s only the first step. You still need to package it, market it, support it, and ensure its ongoing development. If you don’t have an organization that is dedicated to this, you’ll fail. We’ve seen this phenomenon many, many times.

LISA: Many respondents to the 1998 Industry Survey said they were only partially satisfied with their tools. What do you think the reasons are and what do you want to do about it?

Jochen Hummel: These tools are still fairly young and, since natural language processing is a very, very difficult task I guess that language tools users will never be completely satisfied. Most people aren’t even totally satisfied with human translators. Also at TRADOS, we get a really large number of new ideas about what could be done and how we could improve with every new release. In a way it’s good that people are only partially happy, because this keeps them buying new versions of products and finding new ways of doing things.

LISA: So what developments can we expect to TRADOS’ products? When can we expect morphology or conceptrelated memory, for example?

Jochen Hummel: I think you’re talking here about a big ‘R’ and a smaller ‘D’ here. As regards morphology, we are happy with the approach we are currently using, but for MT suppliers, it’s a different issue. And once you start talking about example-based machine translation, more of this technology will be required and in this case we would probably go the same way as others and license existing morphological engines. Some work has been done on concepts, but most of it is in English, and many clients outside the IT industry have other source languages and need to treat all of them in more or less the same way. I wouldn’t really want to make a statement as to when we shall see this materializing.

LISA: What role do standards such as OSCAR play for you in technology development?

Jochen Hummel: Well, of course, being one of the initial members of OSCAR, I think it’s a very good thing. We are using TMX as a format for our new HTML-SGML editor, and as an export and import format. It’s not the full TMX yet, but it is good enough to provide basic data exchange. We talked earlier about cooperation between different tool vendors, and one of the problems you hit there is defining interfaces. TMX makes this very easy, because all the work is already done, and I think we will see this format being used more and more internally in tools alliances. Thus, over time, most tools providers will base their work on the work already done in this group, but it needs people who are enthusiastic about it and try it. Luckily we had some pragmatic and dedicated software developers in the OSCAR group, and they produced a pretty good end result. Of course, developing such a standard requires a considerable investment in resources and makes valuable know-how available to others. Any new entrant to the market will take a look at TMX, and so you could ask why we bothered in the first place, but we are confident that we can maintain our cutting edge and keep ahead of new developments. TMX will be an advantage for us, since it will allow translation memory systems to talk to each other easily, and to communicate with MT systems, for example, without clients even being aware of it.

LISA: What about OSCAR’s proposed TBX (term base exchange) standard?

Jochen Hummel: Term base exchange is a bit more tricky, because it is a much more complex subject, both conceptually and in terms of the technical complexity involved. Also, a lot of work has already been done, so you find that a lot of people are religious about the need to push through their own standards or the stuff they are used to working with. TMX was more something that happened under the hood. If TBX happens, it will help TRADOS define standard layouts for terminology databases. The next Multiterm will be a kind of XML-based database, but users may not have the capability of designing their own DTD, or their own database, so they will want to work with standard models. If such standard models exist, you’ve already done a lot of work for clients. They can then build on this experience and, if they keep to the standard, they will also be able to exchange terminology.

LISA: How is TRADOS affected by the shortage of skilled people in this industry?

Jochen Hummel: The skillbase is certainly limited at the moment and this provides a challenge for our company. It is difficult to find people who have good technical skills and a real interest in languages. On the other hand, though, it’s not very easy for such people to find a company that focuses on this area, and this has helped us to find good staff at the language level. However, when it comes to talking to top management, you need consultants with business skills and a feeling for language, localization and translation. These people are really difficult to find, and the whole industry is suffering as a result. This situation is unlikely to improve, because the demand for IT specialists is outrageous. Every industry wants to get onto the Net and needs to build up big internal IT infrastructures to communicate. You only have to look at the millions and millions of lines of software that you use unconsciously when you drive your car, or take a plane, or make a cup of coffee.

LISA: So why aren’t we collectively or singly investing in universities to influence their curriculums, and get the people we need?

Jochen Hummel: I thought that LISA is actually doing this through the LEIT initiative. Isn’t that, among other things, why companies like TRADOS pay membership fees, to leverage the influence of the collective of organizations in our industry?

LISA: How does TRADOS try to attract and keep people?

Jochen Hummel: I think our biggest asset is our corporate culture. Most people who work for TRADOS—I hope everybody—think it’s a big advantage. That’s a way both of keeping and of attracting people, provided they know about that culture. It’s something we are definitely proud of and we are willing to devote more time and effort to develop it further and maintain it. A second thing we do is to make use of our network of offices. Although this is often difficult to manage, it gives us access to many different labor markets with the specific language skills we need. We also try to work together with some universities—this is on a small scale, but there are several projects going on. Thirdly, we are trying to grow a very successful business that makes it financially interesting for people to join our industry. Top-notch developers want to work in an environment in which you can offer them career opportunities, competitive compensation and new, interesting projects. I don’t think anyone can afford to be static in this business. Incidentally, this is another problem for service providers that want to develop tools: they need to have the same commitment to expand their departments very quickly into a big business. This is very difficult to do if your main business is services, and your management understands a lot about services, but perhaps less about software development.

LISA: Are the tools providers going to able to help the industry cope with the vast increase in demand?

Jochen Hummel: Most companies are aware that internal development is merely the lesser of two evils, and are often willing to change to an external product as soon as something is available on the market, because it’s much, much cheaper in the long run. We shall try and make our business grow as fast as we can, increase our development resources, and use the newest methods and software development technologies in order to deliver products at the speed at which the market requires them. This will definitely be a very difficult task, because language technology can be used in many different contexts, and there are many directions we could go in—we mentioned authoring, workflow, and providing NLP technology on the Internet earlier, to name just three examples. In addition, we have to find the right balance between serving the needs of existing clients and providing new technologies. There’s a commercial aspect involved here, but also development strategy and product strategy issues. What is more, it’s currently quite a fragmented market because we support everyone from individual translators to multinationals. Our next step will be to introduce a client-server architecture: this solves many problems for big corporate users, but isn’t necessarily very interesting for freelance translators or smaller translation groups. We therefore need to find a way to develop this technology so that it doesn’t become too complex, because we definitely intend to continue to service the entire breadth of the market.

LISA: So where do you see TRADOS in five years?

Jochen Hummel: Well, first of all we want to continue being what we think we are—the market leader for computerassisted translation tools. We want to defend and expand this position by having a piece of our software on every translator’s desk. We also want to exploit some of the opportunities that our technology offers to broaden our product range. There are several directions we could go in: what we definitely don’t want to do is end up as a provider of components. We want to keep our real estate on the screen and supply applications to the market. This means that, at a certain point, we will have to broaden the functionality or the scope of our applications. I also hope that TRADOS can maintain its corporate culture, despite the fact that it’s growing very quickly. That will be a challenge, but it’s essential to maintain the fun we are currently having working at this company. We feel very strongly that this is one of the most important tasks we need to achieve our goal. I also want us to be smart enough in the way we manage our growth and finances to remain independent and to grow under the TRADOS flag. Finally, I would like us to find sufficient time to realize some of the great ideas we have. These are difficult to implement right now because we are simply too busy with current developments. In other words, the ‘R’ dimension of our business should grow relative to the ‘D’ dimension.

LISA: Thank you very much.