Reading a scientific paper is one thing, but understanding the data it is based on is quite another. Even when datasets are open, answering an additional question often requires downloading files, figuring out their structure and writing code. Researchers from TIB (Leibniz Information Centre for Science and Technology) have proposed a tool that allows users to ask questions of research data in plain language. The work was presented at the JCDL (Joint Conference on Digital Libraries) 2026 conference and published in its proceedings.
The tool was created by Hadi Ghaemi, Zakieh Alizadehsani and Markus Stocker. It relies on TIB Knowledge Loom, an open digital library that links scientific claims to supporting data, code and methods. The new tool allows users to ask questions of this data through a natural-language question-and-answer interface.
A user may want to reproduce a published result, compare groups within a dataset or study a trend that was not mentioned in the paper. The artificial intelligence (AI - here: a program that translates a question into a database query) system converts the question into a query, executes it and returns a table. If needed, the result can be viewed as a chart. In addition, the user can view the generated query itself and understand how the answer was produced.
The main difficulty for AI is accessing the right data. During testing, the researchers found that linking each question to the dataset associated with a specific scientific finding eliminates errors caused by accessing the wrong dataset and improves answer accuracy. However, even correct linking does not guarantee reliability. The main operating scenario produced correct answers to only about 62-67% of questions. This result shows that simplifying access to data is promising, but users still need to check whether the analysis answers their scientific question.
For researchers, the practical benefit is obvious: when scientific conclusions are linked to supporting data, differences can be studied, observations compared and trends sought using plain-language questions. Because the tool shows the automatically generated database query, the user can check how the answer was obtained - this promotes transparency and reproducibility.