Barack’s Wife Hillary: Using Knowledge Graphs for Fact-Aware Language Modeling

TLDR

Modeling human language requires the ability to generate fluent text while encoding factual knowledge, yet traditional language models struggle to recall facts seen only at training time. To address this, we introduce the knowledge graph language model (KGLM), a neural language model that selects and copies facts from a knowledge graph relevant to the context. Its mechanisms enable rendering unseen information and generating out‑of‑vocabulary tokens, and we also present the Linked WikiText‑2 dataset aligning text to Wikidata. Experiments show KGLM significantly outperforms a strong baseline and even very large language models at completing sentences requiring factual knowledge.

Abstract

Modeling human language requires the ability to not only generate fluent text but also encode factual knowledge. However, traditional language models are only capable of remembering facts seen at training time, and often have difficulty recalling them. To address this, we introduce the knowledge graph language model (KGLM), a neural language model with mechanisms for selecting and copying facts from a knowledge graph that are relevant to the context. These mechanisms enable the model to render information it has never seen before, as well as generate out-of-vocabulary tokens. We also introduce the Linked WikiText-2 dataset, a corpus of annotated text aligned to the Wikidata knowledge graph whose contents (roughly) match the popular WikiText-2 benchmark. In experiments, we demonstrate that the KGLM achieves significantly better performance than a strong baseline language model. We additionally compare different language model’s ability to complete sentences requiring factual knowledge, showing that the KGLM outperforms even very large language models in generating facts.