Using Large Language Models for Zero-Shot Natural Language Generation from Knowledge Graphs
This work addresses the problem of generating human-readable text from structured data for users of knowledge-based systems, but it is incremental as it builds on existing pretraining concepts.
The paper tackled zero-shot natural language generation from knowledge graphs using large language models, showing that ChatGPT achieves near state-of-the-art performance on some WebNLG 2020 metrics but lags on others, with output quality strongly linked to the model's prior knowledge of the data.
In any system that uses structured knowledge graph (KG) data as its underlying knowledge representation, KG-to-text generation is a useful tool for turning parts of the graph data into text that can be understood by humans. Recent work has shown that models that make use of pretraining on large amounts of text data can perform well on the KG-to-text task even with relatively small sets of training data on the specific graph-to-text task. In this paper, we build on this concept by using large language models to perform zero-shot generation based on nothing but the model's understanding of the triple structure from what it can read. We show that ChatGPT achieves near state-of-the-art performance on some measures of the WebNLG 2020 challenge, but falls behind on others. Additionally, we compare factual, counter-factual and fictional statements, and show that there is a significant connection between what the LLM already knows about the data it is parsing and the quality of the output text.