13\/02\/2025<\/small><\/p>\n<\/div> It’s Quality Time again. Welcome to a new episode in our series of interviews with experts about language services. This time, our Head of Terminology Management Jasmin Nesbigall met with Tom Winter, First Chair of Deutscher Terminologie-Tag e. V. (German Terminology Association, DTT). It was an interesting discussion about the importance of terminology in companies and the role of DTT.<\/p>\n<\/div> Tom Winter works as a terminologist, computer linguist and data scientist in language management at Deutsche Bahn AG. He specialises in process automation and integrating, influencing and performing quality evaluations for MT systems. He is chair of the German Terminology Association<\/a> and a member of the Rat f\u00fcr Deutschsprachige Terminologie (Council for German Terminology, RaDT)<\/a> (only available in German).<\/p>\n oneword has been a member of the German Terminology Association since 2014 and is a regular participant in the association’s symposium, which takes place every two years. Our head of department Jasmin Nesbigall is contributing to the conference programme again this year with a presentation and is looking forward to 2.5 days of exchanging ideas, input and networking.<\/p>\n<\/div> Jasmin Nesbigall (JN):<\/em> Hello Tom! How nice that we can share our ideas about terminology today. This topic hasn’t been covered by Quality Time so far and I’m delighted to welcome you here today to talk to me as the Chair of the German Terminology Association.<\/p>\n<\/div> Tom Winter (TW):<\/em> Thank you for inviting me! I’m delighted that you’ve chosen me for this topic.<\/p>\n<\/div> JN:<\/em> You’re a terminologist at Deutsche Bahn AG and First Chair of the DTT. How did you come to the terminology field in the first place?<\/p>\n<\/div> TW: <\/em>I studied translation, but quickly realised that it would be difficult to earn a living from it because I dwelt endlessly on individual phrasings. That’s why I was glad that TH K\u00f6ln offered a Master’s degree in Terminology and Language Technology with Klaus-Dirk Schmitz, which I enrolled in after my first degree. This path was a good one, as developments in recent years have shown that terminology is a subject area of growing interest and great impact in which you can thrive.<\/p>\n I’ve always been fascinated by how ideas are formulated in different languages and what that says about the thinking process. Why do we ask for “the<\/span> meaning of life” in German, but for “a<\/span> meaning of life” in English? In German, we assume that there is one meaning \u2013 and only one! \u2013 in English it remains open. JN:<\/em> Speaking of talking to each other: we’re in a symposium year, as the biennial meeting of the DTT is taking place in Worms from 27 to 29 March. How did you come to DTT and your position as First Chair?<\/p>\n<\/div> TW: <\/em>Traditionally and logically, the head of the DTT must come from industry, as it must not be someone who influences teaching or is responsible for their own company. When the former first chair Mark Childress stepped down, I was asked to fill this position.<\/p>\n<\/div> JN:<\/em> How well do you think the different groups in DTT fit together? It brings together industrial companies, service providers and freelancers, companies that produce tools and university representatives. Every group has different requirements. Do these diverse groups enrich each other or does each interest group stand for itself?<\/p>\n<\/div> TW: <\/em>This is definitely a challenge that we’re working on. Unfortunately, the different interests and needs of the different areas can’t always be combined well in many places. Freelancers, for example, want to network and are looking for new assignments. However, the latter is not the aim of DTT events. We founded the Industry Forum for industry members in 2019. More than 30 companies come together there to exchange ideas about their processes and developments in the field of terminology. The focus is clearly on sharing experience and not on whose help or which tools are ultimately used to implement something. As DTT, we endeavour to serve the interests of all groups. Comprehensive networking is a real challenge, but that’s what the biennial symposium is for, where everyone comes together.<\/p>\n<\/div> JN:<\/em> However, not every presentation at the symposium is relevant for every participant. When putting together the programme, do you look for a balanced mix of topics from all areas?<\/p>\n<\/div> TW:<\/em> Each symposium has a topical theme. The initiators of the DIT (German Institute for Terminology) target speakers who can contribute something to this topic. We make sure that there’s something for every group. Independent expertise is absolutely paramount. In recent years, we’ve also incorporated additional elements to liven up the event, such as the musical accompaniment by Dad’s Phonkey and terminology speed dating.<\/p>\n<\/div> JN:<\/em> In your view, when does a symposium feel like it has been a success?<\/p>\n<\/div> TW:<\/em> If we have managed to come across as up to date and groundbreaking despite a one-year lead time. Three aspects are important for the symposium: expertise, independence and networking. If the symposium is recognised and appreciated for its professionalism and if the participants go home satisfied, with new input and well networked, and if even the contributions that didn’t interest someone so much were well received, then it was a successful event. In addition to these non-material goals, we must of course also cover our costs. I’ve definitely been very satisfied with the event in recent years and the number of participants speaks for itself.<\/p>\n<\/div> JN:<\/em> However, whereas in the past people may have been satisfied with getting two or three good inputs from a conference day, today they try to maximise the benefits of events. This is certainly also due to the large number of competing events on offer, such as webinars, where I can continue my training from my desk. Do you feel like expectations have changed about the personal benefit of an event like the symposium?<\/p>\n<\/div> TW:<\/em> Yes. In many companies, if you participate, you have to create a report afterwards and they assess how much the employee has learnt from the presentations. Benefits are therefore scrutinised and demanded in a completely different way. Perhaps this pressure of expectation also blocks us from letting go and getting involved in this type of event. After all, an event like this should be seen as a whole. The networking element, for example over dinner together, should be just as important as the professional input. If companies demand only expertise, this alone can hardly be effective.<\/p>\n<\/div> JN:<\/em> This is made even more acute when there are parallel events on terminology topics. Because then you’ll certainly also be weighing up which programme is more interesting and which one you can get more from to take with you.<\/p>\n<\/div> TW:<\/em> Absolutely. Parallel events are great and enriching in and of themselves and are enjoyable from a professional point of view. But the market doesn’t suddenly have twice as much money. We simply have to keep an eye on what participants weighing up the greatest possible benefit means for participant numbers in the long term.<\/p>\n<\/div> JN:<\/em> But it also shows that interest in terminology is growing. For a long time, the topic tended to live in the shadows at companies. However, this has changed significantly in recent years. Why do you think that is?<\/p>\n<\/div> TW: <\/em>Clearly technological developments, starting with digitalisation and now the use of AI. Natural language processing (NLP) emerged with digitalisation and has now been given a further boost by AI. Much more text is automatically generated and processed in chains. Terminology comes into play here because of the increasing realisation that this text chain and text processing must be supported. Terminology work is a supporting process in this value chain.<\/p>\n However, the whole thing is made more difficult by the perception that language is completely normal and a given. Very few companies realistically calculate the costs of multilingualism and this is where terminology fails.<\/p>\n<\/div> JN:<\/em> Yes, for many it’s still a nice-to-have. We’re familiar with the entire range of terminology issues in companies: some haven’t even started yet and others are building something on the side. Some work with good, well-maintained databases and others have 15,000 entries that have been lying dormant for 20 years in an almost forgotten database. In your experience, what’s the most pressing terminology issue in industrial companies?<\/p>\n<\/div> TW: <\/em>It’s actually a bit of everything. If you want to implement terminology work properly, it has to be a company-wide process. To achieve this, it’s important to communicate the value and necessity and to present the whole thing as a support function, similar to the HR department or Controlling, whose existence is also taken for granted. Yet, when the HR department takes over recruitment interviews for me, I’m happy about the direct support I receive. Terminology is different: it requires input from the specialist departments, some of which can’t see the benefits directly. Rather, the benefits are for the company as a whole and are more difficult for terminologists to communicate.<\/p>\n Because terminology work concerns the whole company, it must be organised centrally, with a central budget, the goodwill of colleagues and support “from above”. The management team must recognise the benefits and purpose and needs a sound knowledge of language, technologies and, above all, knowledge management. And that’s where many people get stuck: too often they only see the costs, which then causes it to fail.<\/p>\n The situation is similar when it comes to the distribution of the budget: new topics such as AI are at the top of the list here. This then leads to the budget for other areas being cancelled. But AI without terminology is inconceivable. Of course, no department gives up its budget voluntarily. This only happens when you reach a point where you can’t go any further without terminology.<\/p>\n AI is currently being rolled out everywhere and there are always increased benefits without you having to do anything, simply because the systems are always getting a little better. That’s why you can put off improving things yourself for a long time. For the last few yards, however, company data, i.e. terminology, must be integrated to get the full benefit.<\/p>\n<\/div> JN:<\/em> It’s not only the case that terminology is needed for AI, but also that AI can provide support for terminology tasks. What are your experiences with it? In which processes in the terminology lifecycle does it already make sense to use AI?<\/p>\n<\/div> TW:\u00a0<\/em>I still take a critical view of AI being used to create definitions. Where is a generic system supposed to get specialised definitions from? It may be able to carry out preparatory work, which must then be checked by experts. Unfortunately, I don’t see any great benefit here, but rather a danger of inaccurate information.<\/p>\n The situation is different with metadata. Let’s take the example of plain language. Plain language requires consistency and transparent terms. In the railway sector, for example, the term “St Andrew’s cross” is meaningless. I could call this sign a “train warning cross”, which is much more meaningful: it’s a cross and it warns of approaching trains. Following this formula, you could now take a terminology database and use AI to reformulate all existing definitions into plain language. The system could then be asked to suggest a descriptive term for each concept that’s also understandable in plain language.<\/p>\n If I’m then not allowed to use one of these terms for legal reasons, I can correct it easily because the terminology is consistent. Because then I could, for example, replace my standardised “train warning cross” in the documents with “St Andrew’s cross” again.<\/p>\n<\/div> JN:<\/em> What about support in MT processes, for example glossary creation?<\/p>\n<\/div> TW: <\/em>When terminology is integrated into MT systems, it sometimes works excellently and sometimes not at all. For example, if you enter train = Bimmelbahn into a glossary, this will be correctly translated in a sentence such as “The train is delayed”. However, if you then enter “The training was hard to follow”, “training” is translated into “Bimmelbahn” in German. The problem here is the same word stem, because the integration procedure for glossary specifications is often based on stemming.<\/p>\n However, if you know that terms with identical word stems can lead to mistranslations, then AI can provide wonderful support. To do this, you create an attribute field “word stem” in the database and let the AI fill it in for all terms. I estimate that at least 90 per cent of the results will be correct. You can continue working with this attribute and, for example, exclude terms with identical word stems from the glossary or check them separately. This type of process would be inconceivable \u2013 at least for large databases \u2013 without AI. AI can achieve a lot if you know how to integrate it.<\/p>\n<\/div> JN:<\/em> We’ve carried out interesting tests with definitions. We had a terminologist define a concept and also defined it using four different prompts. Looking at all the results together, none of the five definitions contained all of the applicable elements. The best result would therefore be the sum total of all the definitions.<\/p>\n<\/div> TW:\u00a0<\/em>Interesting! I think you have to let go of the principle of perfection a bit and be satisfied with 80 per cent sometimes. In many areas, 100 per cent simply does not exist. There can also be disagreements about definitions created by humans, because there are always different perspectives. Can there even be a gold-standard reference definition? Five people would also define a given concept differently.<\/p>\n I find it increasingly exhausting to deal with the two opposite opinions: some people are happy with everything that comes out of AI and forget to scrutinise it critically, while others reject it in principle and only cite negative examples.<\/p>\n<\/div> JN:<\/em> We know this well enough from machine translation (MT). Right from the start, there was the camp that accepted MT without asking questions and said that human translators were no longer needed. And the other camp picks up on every little mistake made by the machines and says “A human would never have made that type of mistake, MT is no good at all!”<\/p>\n<\/div> TW: <\/em>That’s right. But somehow nothing seems to have been learnt from the eight-year head start in machine translation.<\/p>\n<\/div> JN:<\/em> I think the industry has become more relaxed. With the breakthrough of neural machine translation, there was more excitement because everyone wanted to test and use it quickly. With LLMs, the reaction felt a little more detached and the attitude was more “Ah, the next hype. But it won’t deliver on its promises.”<\/p>\n<\/div> TW: <\/em>Absolutely. But with MT, the discussion only took place within the industry, whereas with AI it’s all-encompassing. Now everyone’s talking about it.<\/p>\n<\/div> JN: <\/em>We’ve noticed that there’s more demand from our customers to use AI. In many discussions, they ask whether a process can be supported by AI, reducing costs. What’s your perception: is AI already being used in companies for terminology work or is this still a test phase?<\/p>\n<\/div> TW:\u00a0<\/em>When I look at my employer and at our Industry Forum, it varies greatly depending on the company. In extreme cases, there’s the idea that processes can simply be replaced by AI, saving the budget entirely. This is particularly the case with translations and is currently on the rise. However, similar demands are also being made in terminology, with the question of whether ChatGPT could simply do it all. This phenomenon needs to be addressed because GPT is perceived as an all-powerful tool and, with the current economic situation, this will become even more critical.<\/p>\n On the other side are the terminologists who have been working in their processes for a long time. Here, too, there are two camps: some realise that a lot would be possible with AI support, but wonder what would be left for them. So there’s a fear of becoming superfluous and losing their job, which is why they prefer to keep things as they are. Others are very open and already have concrete ideas and visions for how they might use the time that AI will free up for them.<\/p>\n The question is: in a few years’ time, will we still reach a point where terminology for knowledge management can still fulfil its full potential, if we’re already seeing companies abandoning their long-term commitment to established processes in favour of placing false hope in AI?<\/p>\n<\/div> JN:<\/em> You just mentioned the contrasting views: terminologists who don’t want to deal with AI at all and others who see it as an opportunity. What skills do you currently need for terminology work?<\/p>\n<\/div> TW:\u00a0<\/em>Skills also encompass the mindset. Terminology must be removed from the field of translation and translations must generally be regarded as information processing, i.e. as information and data. Language is perceived as complicated, but the complexities are only the exceptions. And in digitalisation and AI, the most important thing is to distinguish between the exception and the rule. Rules can be automated, and then humans take care of the exceptions.<\/p>\n When I talk about data and information, I’m only one step away from programming languages, digitalisation and AI. This is where I see the requirements for the terminology sector in the coming years. Terminology can only exist as an element of knowledge management. It’s a support process, for example for consistency. Everything in the company is based on the corporate language, which also serves the internal search, the ERP system, chatbots and machine translation. All text-processing processes are based on the corporate language, which must be maintained and provided accordingly.<\/p>\n Terminologists must have the entire company in mind. They need to seek out contacts, know how other departments work and know how and where they can provide support. To do this, you have to look at the company architecture and see how the terminology process can be anchored there. Then terminology has a future.<\/p>\n The skills required are therefore to change mindset and then to engage with the technology. You need knowledge of Python, SPARQL and XML, an understanding of interface APIs and how systems interact. Ideally, you can code yourself.<\/p>\n<\/div> JN:<\/em> Will the job title change at some point? “Terminologist” still sounds like a dictionary and a magnifying glass. What you describe would perhaps be more of a “Language Data Manager” or “Corporate Language Manager”. This would perhaps also demonstrate the overarching role required to the outside world.<\/p>\n<\/div> TW:\u00a0<\/em>Kara Warburton has published the book “The Corporate Terminologist”. There, she emphasises the methodology of terminology. She doesn’t limit it to terminology, but instead speaks of “microcontent”. In other words, any and all information is stored in a terminology database so that it can be transferred to other systems that are in desperate need of it. But “content” can be anything. When someone tells me that this is a “technical report”, I have a completely different expectation of it than of “content”. But what do we want to call it? “Data” is far too unspecific.<\/p>\n “Terminology” may be outdated, but it has value and standing. My hope is that more value will be placed on this term again. And that you have more confidence when you call it a “terminology database” than when you call it a “microcontent database”. The word may be outdated and is perhaps a hindrance to getting a budget, but in five or ten years, it may be of value again because it builds trust.<\/p>\n<\/div> JN:<\/em> Perhaps other synonyms will establish themselves too.<\/p>\n<\/div> TW: (laughs) Do you have any ideas?<\/em><\/p>\n<\/div>Quality Time with Tom Winter: Discussion with the DTT Chair about terminology, AI and quality<\/h1><\/div>
\nHowever, many linguistic inaccuracies or errors can be corrected by looking at the definition of a concept. Because you have to distinguish between what’s a defining characteristic and what’s an attribute.
\nLet’s take racism as an example. Skin colour is made into a defining characteristic; a distinction is made between black and white people. But skin colour is a human attribute, not a defining characteristic. There’s only one definition of “human”. So conflicts can be resolved using terminological groundwork, i.e. logical data structuring. Terminology can cover the entire framework \u2013 in terms of content, politics and society \u2013 by defining and constantly comparing: what does that mean for you and what does it mean for me? Do we understand a given concept as the same thing? This leads to transparency and starting points for approaching and talking to each other.<\/p>\n<\/div>