
UPHG blog
Here we post occasational stories and ad-hoc blog posts about our work.
New research: The Principle of Public Access as a Barrier to Patentability
By Katarina Halldin, September 2026.
The Principle of Public Access as a Barrier to Patentability
The Scope of Confidentiality Protection for University Research Funding Applications in Swedish Law
Master thesis summary.
Katarina Halldin, Uppsala University, Spring 2026
Supervisor: Professor Bengt Domeij
This thesis examines a structural tension in Swedish law: the constitutional principle of public access to official documents can unintentionally undermine the patentability of university research. When a researcher at a Swedish university applies for research funding from a state funding agency such as the Swedish Research Council, Forte, Formas or Vinnova, that application becomes an official document under the Freedom of the Press Act, and in principle, anyone can request to see it. This matters because, under patent law, technical information can become prior art once it is publicly accessible which can either destroy an invention’s novelty outright or, more subtly, combine with other known technology to undermine its inventive step. This second, less obvious risk was illustrated by the Patent and Market Court of Appeal in case PMT 2840-24, where a Swedish ethics-review decision, which had become public, contributed to invalidating a European patent for lack of inventive step, even though the decision itself did not reveal the invention outright.
Examining Sweden’s main confidentiality provision in this area, Chapter 30, Section 23 of the Public Access to Information and Secrecy Act, the thesis identifies three structural weaknesses: uncertainty over whether the provision applies to basic research funding, which lacks a clear trade and industry connection; a damage requirement that places the burden of proof on the researcher despite the harm being irreversible; and a right to disclose confidential information to the media that, for most public funders, remains unaffected by the secrecy provisions. A disclosure request submitted to Vinnova as part of the thesis further illustrates the unpredictability of the current framework in practice.
A particularly important part of the analysis was the comparison with Finland. Although both countries share the same constitutional tradition of public access to official documents, they have adopted different approaches to research funding applications. Swedish law rests on a presumption of openness, whereas Finnish law rests on a presumption of confidentiality. The Finnish model also requires no commercial connection and expressly identifies patent protection as a purpose of the rule. Case law from Finland’s Supreme Administrative Court shows that administrative information, such as what is funded and how much, can remain public while the technical core of the research stays protected.
The thesis concludes that Swedish law would benefit from a reform inspired by, but not directly adopted from, the Finnish model. A time-limited presumption of confidentiality, combined with a reversed burden of proof and an explicit patent-law purpose, would address shortcomings in the current openness-first approach. At the same time, an exception for information that can be separated from the technical core of the research could help avoid the risk of over-protection inherent in the Finnish model. Such a reform would provide researchers, universities, and innovation offices with a more predictable starting point for protecting potentially patentable research, without sacrificing public insight into how government research funding is spent.
About the author:
Katarina Halldin holds a law degree from Uppsala University, specializing in intellectual property. This thesis was written as her final degree project, under the supervision of Professor Bengt Domeij.
Contact: Katarinahalldin@live.com
Download the thesis here (inSwedish): https://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-597990
Internship reflection Spring 2026: Yiwei Wang "LLMs as your best friend, and your worst enemy"
By Yiwei Wang, June 2026.
On June 5, I completed a five-week internship at the Uppsala Patent History Group (UPHG), a project at the intersection of business studies, history, and digital humanities. It gives me a comprehensive idea of how a research project runs, provides me with hands-on experience in multi-degree problem-solving, and prompts me to reflect on the role of large language models (LLMs) in research.
I am mainly involved in two projects. The first concerns the real cost of patent application in 19th Century Sweden. In addition to application fees and investment in experiments, innovators at the time also had to hire agencies for filing patent documents, or have the prototypes of patented objects drafted by designers. These hidden costs are revealed, at least partially, in the 1900s account books of Swedish companies. My task, then, is to automate a process for transcribing and inspecting these handwritten documents, to help the researchers identify patents of interest and their related costs.
The second project is to look into 18th Century Swedish biography books, which documented, in blackletter typeface prints, the most famous people at the time, and trying to find among them the innovators at a time where patent was yet a concept.
Both tasks present great HTR (handwritten text recognition) or OCR(optical character recognition) challenges, partly due to low recognition rate of old Swedish texts, and partly due to the huge volume of documents. By running test trials with a variety of tools, carrying out CER & WER tests of results, and carefully inspecting the original historical documents for smarter transcription strategy, I have succeeded in building automated processes that could ensure high-quality and consistent output for transcription and interpretation.
During my exploration process, LLMs are used extensively, both in generating codes and improving results. The use of the latest LLM models has facilitated my work, but I have also encountered several issues with these tools.
For example, while other documents in the same batch returns more or less accurate results with the same prompt, Gemini-3-flash generated the following transcription, where the DEBIT side is empty, and the CREDIT side quotes totally irrelevant texts (hello, a railway station revitalization in Czech Budejovice?).


While the above is a rare case, more common is the one shown below, where all information at the DEBIT side is displayed at the CREDIT side.
As previous researches pointed out, this could be due to the model’s over-reliance on linguistic priors (Lyu et al., 2026) in cases of uncertainty during the image decoding-and-encoding process (He et al., 2025). Despite continuous efforts from computer science scholars in evaluating (Dayoda and Kunimoto, 2024) and addressing the issue, the problem persists.
In a small-sample test (processing 10 images at once), the newly released Gemini 3.5 performs more stable than previous models and doesn’t have this issue. However, large scale testing is still required to check the stability of results.
Despite the above issue, my method explorations echo with a former research on financial invoice OCR (Loukil et al., 2024), finding that traditional OCR methods do not only lack the capability of recognizing layout and cultural contexts, but also are too often overly sensitive to noises. While I eventually mainly used LLM for this task, a perhaps better sol
ution is to incorporate traditional OCR with LLM to create a more effective pipeline for information extraction.
Another issue with the LLMs is laziness. I encountered this both in using the Gemini API and the Gemini Pro interface, with the latest model (3.5). In transcribing a 300+ page PDF document, Gemini API stopped transcription in the middle of the file and reposted the last line again and again in the output txt files (as shown below) to “fake” the length of the transcription. In dealing with a 101-page PDF, again, the Gemini web interface jumped contents 10 times and skipped nearly half of the pages. Based on how the codes and prompts are structured, my hypothesis is that the former issue may be due to limited capacity of a free API, while the later could be emissions caused by the pause and restart of Gemini when it reaches the limit (In the prompt, I instructed Gemini to be aware of the capacity issue, pause and restart if necessary).


My final solution to the later issue involves preprocessing the PDF into images and manually process them in batches through Gemini. The solution to the former issue could be, again, a combination of the traditional OCR tool as a pre-processor and LLMs as performance enhancer.
To conclude, I could not have completed the tasks that I was assigned in such a short time span without the help of LLMs, but with it, new problems emerge as the time spent in inspecting and correcting the AI-generated results is considerably long. For me, this indicates the importance of curating a more thoughtful research process to minimize the effect of errors and emissions, which requires both in-depth knowledge of the research subject and how the LLMs work.
During the internship, I had a lot of support from my supervisors and my peers. I would like to thank Sofia for the help in deciphering 18th century Swedish words and paragraphs, Risa for inspirations in code works. I would like to thank Yunting for the many good suggestions at work and the thoughtful information sharing as an effort to constantly engage us in the research community. I would also like to thank David, Fredrik and Matti for their professional insights in research, as well as their warm attitudes in creating a free and cozy working environment for all of us, in which we were able to participate in numerous workshops, meetings, and seminars within the department. Even lunches at the dinging hall table with all the PhD students have been fun and inspiring for me, as I get to see what real, daily academic life is like. I am utterly grateful to this precious chance of working alongside this brilliant group of people, and will forever treasure this experience.
References
Dayoda, R. and Kunimoto, M., 2024. Automated Evaluation of Visual Hallucinations in Commercial Large Language Models: A Case Study of ChatGPT-4V and Gemini 1.5 Pro Vision. [online] Authorea. Available at: https://doi.org/10.22541/au.172288308.83969872/v1 [Accessed 4 June 2026].
He, Z., Zhang, C., Wu, Z., Chen, Z., Zhan, Y., Li, Y., Zhang, Z., Wang, X. and Qiu, M., 2025. Seeing is Believing? Mitigating OCR Hallucinations in Multimodal Large Language Models. In: Advances in Neural Information Processing Systems (NeurIPS 2025). [online] Available at: https://proceedings.neurips.cc/paper_files/paper/2025/hash/6b628a3d7cb8055eb6fd2dd48c586347-Abstract-Conference.html [Accessed 4 June 2026].
Loukil, F., Cadereau, S., Verjus, H., Galfré, M., Salamatian, K., Telisson, D., Kembellec, Q. and Le Van, O., 2024. LLM-centric pipeline for information extraction from invoices. In: International Conference on Foundation and Large Language Models (FLLM2024). Dubai, United Arab Emirates, November 2024.
Lyu, J., Wu, D., Cao, T., Li, G., Shen, H., Ma, C. and Zhou, Y., 2026. HalluText: Towards Benchmarking and Mitigating OCR Hallucination for LVLMs. [online] OpenReview. Available at: https://openreview.net/forum?id=LRnt6foJ3q [Accessed 4 June 2026].
Internship reflection Spring 2026: Sofia Jönsson Tellefors
By Sofia Jönsson Tellefors, June 2026.
This year, during five weeks between May and June, I have been doing my internship as a Digital Humanities master student with the Uppsala Patent History Group, together with Risa Yagisawa and Yiwei Wang. I was tasked with digitising patent litigation documents spanning 1887–1935, that were available through the National Archive in Stockholm.
The work I had to do was straight forward, but time intensive: familiarise myself with the material by traveling to the archive, scan the documents at the archive to make them digitally available, organise and the rename all the image files (with the help of some Python coding), and lastly write the documentation for the images, such as what kind of patents or their owners the litigations revolved around.
While working towards my task I also had the opportunity to experience the Department of Business Studies more academic side and the other research that was conducted here at Uppsala University. Through my participation in the group meetings, seminars, workshops and even a couple of doctoral thesis defences I have gained more insight into the kind of activities that happen at the University, aside from teaching courses. Most notable to me is how collaborative the work is, not only within a research group or department but also with scholars internationally or agencies and institutions outside of academia.
Since I was not the only one doing my internship with UPHG from the Digital Humanities programme, I also gained a lot of insight into the flexibility of how the things we have learned can be applied for different types of patent research. I am very thankful to have been given this opportunity!



Internship reflection Spring 2026: Risa Yagisawa
By Risa Yagisawa, June 2026.
From May to June 2026, I experienced a five-week internship at Uppsala Patent History Group (UPHG). This opportunity brought me intersectional perspectives between economic history and digital humanities.
I was mainly involved in the research project collaborating with Kobe University. The project examines the roles of patent agents during the 20th century in Japan, and I focused on a German patent attorney, Karl Vogt, who worked as a patent agent in Japan. I had two tasks: one was doing OCR the patent documents regarding him, and the other was extracting information from the OCR’ed text files. For OCR, I compared three different OCR AI tools, which made me realize the unique features of Japanese historical documents. I also did Python programming to get invention titles, inventor/patentee names and addresses, and patent agent names, etc., from the documents. This experience enabled me to get familiar with digital methods and get to know how these digital tools are implemented into actual historical research projects.
During the internship, I also had opportunities to attend different academic events, such as workshops, higher seminars, and doctoral thesis defenses. I could feel real life as a PhD student in the Department of Business Studies at Uppsala University, other than researching and teaching, and all of them were precious experiences as a master student. Also, I had a lot of time to enjoy conversations with my supervisors, colleagues, and PhD students, which made the internship experience even richer.
Thanks to this opportunity, I could apply my knowledge learned from the DH program to the actual research project. Also, I am very grateful to my supervisors and colleagues for supporting me in different ways.



UPHG is hiring a postdoc!
By UPHG, April 2026.
THE DEADLINE HAS PASSED.
Matti La Mela is hiring a postdoc for his project “Making public property: A Digital history of allemansrätten in Sweden and Finland, 1880–1950” (Vetenskapsrådet 2024-02209, PI Matti La Mela).
The project investigates the emergence and meaning of allemansrätten (right of public access to nature) through digital analysis of historical source materials, with a particular focus on how access to nature and natural resources has been formulated and contested over time. The project aims to reassess established interpretations of the right of public access and to understand how open access rights developed in parallel with increasing commercialization of natural resources (such as wild berries and mushrooms).
Deadline: Tuesday, 12 May 2026.
https://www.uu.se/en/about-uu/join-us/jobs-and-vacancies/job-details?query=921843
Project info: https://github.com/mlamela/dh-allemansratten/blob/main/README.md

UPHG is expanding with up to 2 PhD candidates
By UPHG, February 2026.
THE DEADLINE HAS PASSED AND THE POSITIONS ARE FILLED.
We are hiring!
The Department of Business Studies at Uppsala University and UPHG are announcing 1-2 PhD positions commencing Fall semester 2026 (start date in mid-August). For more information about the positions, see the weblink below. Deadline for applications is March 31.
Read more about the positions here: https://www.uu.se/en/about-uu/join-us/jobs-and-vacancies/job-details?query=903344
One position covers strategy and innovation more generally, while the other is part of a project proposal on social capital and innovation (funding pending).
If you have questions regarding these positions, please don’t hesitate to contact Fredrik Tell.

Internship Reflection: My experience at the Uppsala Patent History Group
_edited.png)
By Lu Zhang, February 2026.
From November to December 2025, I completed a five-week internship as a Master's student at the Uppsala Patent History Group (UPHG). This experience offered me a valuable opportunity to step into the actual practice of digital humanities research.
During the internship, I was involved in two main tasks. The first was in collaboration with the Swedish Patent and Registration Office (PRV), where I analyzed layout changes in historical patent documents in the Svensk Patentdatabas. This foundational work will support future efforts to extract bibliographic information using AI models like ChatGPT. The second task was carried out with the Swedish National Archives (Riksarkivet), where I and my classmate Sara reviewed and verified historical patent documents transcribed by Transkribus. I carefully checked each line for accuracy, and ensured the reading order matched the original documents.
The most valuable aspect of this internship was observing how the research team collaborates in practice—from regular project meetings and discussions with external partners to the actual execution and progression of tasks. I gained firsthand insight into how academic research operates within a team setting and how researchers communicate with experts from other organizations. The team also provided me with relevant literature to support my goal of developing this internship experience into my Master's thesis.
This experience showed me that clear communication, curiosity, and a willingness to learn from mistakes are essential in a professional research environment.I am grateful to the UPHG team for their guidance and support. This experience will serve as an important foundation for my continued journey in digital humanities.