Analysis and Generation of Wikidata Descriptions

Sammanfattning

This thesis explores the structure and generation of descriptive texts for Wikidata entities,focusing on cities, universities, and mathematicians. The goal is to develop agrammar-based, language-independent system for automatic description generation.We begin by analyzing multilingual description patterns across six languages, revealinghigh structural consistency within languages and substantial cross-languagevariation, particularly between European and non-European language groups. Adetailed property frequency analysis shows that a small number of attributes accountfor the majority of descriptions. Further label occurrence analysis indicatesthat while human-readable administrative attributes are well represented, identifiersand spatial data are rarely included in natural language descriptions. To mitigatemissing label issues, we design a data augmentation pipeline using GeoNamesand OpenStreetMap, significantly improving label coverage across languages. Wealso compare the grammar-based approach with a Retrieval-Augmented Generation(RAG) system and find that the former performs significantly better in terms ofclarity, structural consistency, and multilingual alignment. Our findings inform thedesign of a multilingual description generation system based on Grammatical Framework(GF), emphasizing clarity, informativeness, and structural consistency. Thisproject is part of a broader collaboration: Bokun Xiao contributed to the developmentof the core grammar, Imtiaz Ayon focused on building the Bengali grammar,and another team was responsible for the Greek grammar.