Knowledge graph boosting language models in Latin American cultural questions
Knowledge-Graph Based Augmentation versus Retrieval Augmented Generation for Cultural-Related Question Answering
Computation and LanguageArtificial Intelligence
Summary
Language models often miss facts about less-represented regions like Latin America because these facts appear less in training data. The authors compare two techniques: one uses external text documents (retrieval-augmented generation), the other uses structured knowledge graphs built automatically from Wikipedia. Their knowledge-graph method performs competitively, reducing errors significantly and even works well in Portuguese without extra training. This suggests knowledge graphs can help language models answer culturally specific questions more accurately.
What this means in practice
- •For cultural content platforms: Improve question answering about Latin American culture by integrating language models with automatically built knowledge graphs.
- •For multilingual chatbot developers: Create chatbots that answer culturally specific questions in Portuguese without needing dedicated fine-tuning on target language data.
Authors
Pablo Poulenard, Yannis Karmim, Valentin Barrière
Abstract
Large language models (LLMs) suffer from a long-tail deficit: culturally specific facts, particularly those concerning underrepresented regions such as Latin America, appear too rarely in pretraining corpora to be reliably memorized. Retrieval-Augmented Generation (RAG) addresses this by grounding generation in external text, but structured alternatives such as Knowledge Graphs (KGs) offer tighter control over what enters the context, along with potential gains in explainability and updatability. We benchmark Graph-RAG against standard RAG on LatamQA, a culturally grounded multiple-choice dataset spanning eight thematic categories. The graphs are built end-to-end from Wikipedia articles with KGGen, a recent open-domain extractor, without manual curation in our main setting. G-Retriever is competitive with RAG and reduces the error of the base LLM by 72\% with a standard KG and 78\% with a benchmark-aware variant, the gap to RAG narrowing further as the graph is oriented toward task-relevant content. The trained projection transfers zero-shot to Portuguese without target-language fine-tuning, indicating multilingual reach.