Introducing rich semantic knowledge into large language models improves few-shot Chinese spell checking performance.