Radical Prompting: Enhancing Chinese Language Models with Character Visual Analysis
Abstract: As a glyphic language, Chinese incorporates information-rich visual features, with distinct characters combining to form compounds that inherit the meaning or pronunciation of their components. However, we argue that Large Language Models (LLMs) fail to effectively harness this valuable feature. This study designs 'radical prompting' to improve LLMs' effectiveness across general NLP tasks such as Part-Of-Speech (POS) tagging and investigates the limitations of contemporary LLMs in accurately identifying the visual information of characters. Results demonstrate that the introduction of 'radical prompting' markedly improved LLM performance across various NLP tasks, particularly when correct radicals were provided, highlighting its potential as a crucial tool for optimizing Chinese language processing. However, most LLMs struggle to correctly identify the visual fundamentals of Chinese characters, which limits their effectiveness. Despite some progress achieved through prompting and fine-tuning, the current accuracy levels still fall short of the desired excellence.
Paper Type: Long
Research Area: Phonology, Morphology and Word Segmentation
Research Area Keywords: Morphology,
Contribution Types: Model analysis & interpretability, Data resources, Data analysis
Languages Studied: Chinese
Submission Number: 468
Loading