Homepage for UWB Portal News

What does it actually mean to translate sign language using a computer? And what would it be useful for?
One possibility is converting text into sign language, meaning generating a signed video. I focus on the opposite direction: taking an input video and turning it into understandable text. The system first has to recognize what is happening in the video — hand movements, facial expressions, and spatial use. It then has to understand the meaning and translate it into a natural, grammatically correct sentence. The goal is an application that makes communication with the Deaf community easier. At the moment, though, we are still mostly in the research phase and producing partial outputs, for example, used in presentations or academic papers.

Did you have to learn sign language yourself? Where do you get the data for this kind of research?
The data comes from two sources. Either we create it intentionally by asking someone fluent in sign language to sign specific sentences, or we use existing videos, typically from YouTube, where Deaf creators run their own channels and often provide subtitles. I personally haven’t learned to sign properly yet. My own signing would not look natural and would be full of mistakes. But I can sign “it will rain tomorrow” and “llama” (laughs).

Do you work with the Deaf community as well? What do they think about your research?
Collaboration is essential because sign language is far more complex than it may seem at first glance. It is not just about hands; it also involves facial movements, expressions, and context. Without consulting experts and native signers, we would not understand these things. We have worked with people from the United States and Saudi Arabia, as well as Czech linguists at Charles University, and we are now establishing cooperation with the Deaf community in Pilsen. Their feedback is valuable, but often critical as well. For example, they point out that “artificially” created data — when someone signs a pre-written text — feels unnatural. For us, however, the problem is that there is very little natural data available, so we have to work with this compromise. Without sufficient data, the system cannot be properly trained.

There are several projects at FAV focused on different sign languages. Why? How much do they differ?
Sign languages are not universal. They have their own language families, much like spoken languages, and they evolved completely independently of spoken languages. For example, American Sign Language and British Sign Language are two entirely different languages, even though they share the same spoken language. Interestingly, some signs are similar across languages — for instance, “nose” is usually signed near the nose. Abstract concepts, on the other hand, tend to differ much more.

How long will it take for this research to move from the lab into everyday life? Why is progress so slow?
The main obstacle is translation quality. Even the best current systems are still inaccurate — it resembles the early versions of online machine translators that were only half reliable. The reason is simple: we do not have enough data. A high-quality system would require roughly ten thousand hours of data. For American Sign Language, we have around 2,000 hours, and not all of it is high-quality. For Czech Sign Language, we currently have only about thirty hours of publicly available data. Right now, for example, I am trying to add recordings from Czech Television broadcasts or parliamentary press conferences.

Why did you decide to pursue a Ph.D. instead of going straight into industry after your master’s degree?
It was not an easy decision, but the team environment and open discussions about the pros and cons helped me make it. If I had gone straight into industry, I would never have returned to a Ph.D. Program. In the end, curiosity and the desire to try it won out.

How much do financial considerations influence students’ decisions about whether to pursue or continue a Ph.D.?
A lot. A Ph.D. offers many opportunities, but the financial side is weak. For many people, that is the main reason they choose industry instead. You can think of it as an investment — in the short term, I earn less, but in the long term, it will pay off. The problem is that not everyone can afford that.

What should someone know before applying?
The key is choosing the right topic and supervisor. If you pick something you genuinely enjoy, it opens doors to top researchers, conferences, and experiences that would otherwise be much harder to access.

Where do you see yourself after completing your Ph.D.?
I want to do a postdoctoral fellowship abroad. At the same time, I will definitely return to the Czech Republic afterward and put the experience I gain here to good use.

Three words that describe a Ph.D.?
Flexibility, research, opportunity.

Do you have any advice from a supervisor?
“From now on, you’ll only have more money.” It helped me stop hesitating whenever I wanted to buy something more expensive, wondering whether I could afford it.

Biggest myth about Ph.D. students?
That they are boring nerds.

What would you change tomorrow if you could?
Stronger connections with foreign universities and greater involvement of international experts in teaching and research projects.

Are you currently studying for a follow-up master’s degree, interested in science, and unsure what comes next? Stay and discover more. Applications for doctoral studies at the Faculty of Applied Sciences are open until May 31, 2026.