University of Twente Student Theses
As of Friday, 8 August 2025, the current Student Theses repository is no longer available for thesis uploads. A new Student Theses repository will be available starting Friday, 15 August 2025.
LLM-based Type Alignment Performance for Under-resourced Languages
Karkalasev, Filip (2025) LLM-based Type Alignment Performance for Under-resourced Languages.
|
PDF
478kB |
| Abstract: | This research investigates whether LLM-based, knowledge graph type alignment performance is affected across a set of languages (German, Spanish, Dutch, Croatian and Macedonian). LLM-based type alignment refers to an LLM selecting a matching pair of entities between two knowledge graphs, where the entities are types instead of specific instances, to determine whether under-resourced languages form a factor in the performance of type alignment. For this, perturbations (simulating sparse attribute information, incomplete coverage, reduced granularity and other structural inconsistencies) are inflicted into a knowledge graph extracted from Wikidata, and alignment is tested for different languages. The findings show that the overall alignment performance is minimally affected by different languages for the different perturbations. However, for the incorrect pairings made, the amount of hallucinations in an under-resourced language are larger, indicating less stability for difficult to align pairs and a lesser robustness to failure handling in an under-resourced language. These findings point to an area of potential in alignment failure investigation for LLM-based approaches, focusing on typographical and character-based perturbations, as these manipulations were found to have a more detrimental effect on alignment performance than semantic alterations. |
| Item Type: | Essay (Bachelor) |
| Faculty: | EEMCS: Electrical Engineering, Mathematics and Computer Science |
| Subject: | 54 computer science |
| Programme: | Computer Science BSc (56964) |
| Link to this item: | https://purl.utwente.nl/essays/107513 |
| Export this item as: | BibTeX EndNote HTML Citation Reference Manager |
Show download statistics for this publication
Show download statistics for this publicationRepository Staff Only: item control page