The rapid expansion of information on the Web resulted in the development of collections of structured data known as knowledge bases (KBs). Finding necessary pieces of data in these KBs (such as DBpedia and Wikidata) usually requires knowledge of their structure and query languages, which can be challenging for an inexperienced user. MA Elizaveta Vilavuo’s research bridges this gap by developing systems letting users ask questions and receive answers and summaries written in natural language.
The dissertation describes several question-answering systems, such as GQA, a grammar-based system that parses complex questions, and MuG-QA, its multilingual extension supporting English, German, French, and Italian. The most advanced system, TraQuLA, involves flexible pattern matching and allows users to trace the system’s reasoning in transforming natural-language questions into graph queries, thus ensuring transparency. These systems perform competitively with state-of-the-art models while remaining fully interpretable.
The dissertation also tackles the challenge of generating human-readable summaries for multiple KB entities — a task previously unaddressed in research. It might be infeasible for a human to browse numerous data points to get an idea of a collection of entities, so getting a readable overview of their key features can be useful. Alongside with the summarisation system, E. Vilavuo designed an experimental framework with evaluation criteria to assess the quality of machine-generated summaries and their effectiveness in helping humans in writing their own summaries.
Despite the growing popularity of deep learning and neural networks, they are often criticised for being a “black box” for their own developers and users. The systems described in the dissertation are totally transparent and easy to control and debug, at the same time being close in performance to the state-of-the-art methods. This makes them especially suitable for applications in which user trust and developers’ understanding of their algorithms are critical.
Elizaveta Vilavuo currently continues her career in language technology, working at Lingsoft as a software developer.
Public defence on Tuesday 20 May
The doctoral dissertation of MA Elizaveta Vilavuo in the field of computational linguistics titled Transparent RDF Question Answering and Summarisation will be publicly examined at the Faculty of Information Technology and Communication Sciences at Tampere University at 12 o’clock on Tuesday 20 May 2025 in the auditorium Pieni sali 1, Festia building, Hervanta Campus, Korkeakoulunkatu 8, Tampere.
The Opponent will be Associate Professor Haridimos Kondylakis from the University of Crete. The Custos will be Professor Jyrki Nummenmaa from the Faculty of Information Technology and Communication Sciences.
The doctoral dissertation is available online.
The public defence can be followed via remote connection.