Decoding Prompt Syntax: Analysing its Impact on Knowledge Retrieval in Large Language Models
Stephan Linzbach, Tim Tressel, Laura Kallmeyer, Stefan Dietze, Hajira Jabeen
Abstract
Large Language Models (LLMs), with their advanced architectures and training on massive language datasets, contain unexplored knowledge. One method to infer this knowledge is through the use of cloze-style prompts. Typically, these prompts are manually designed because the phrasing of these prompts impacts the knowledge retrieval performance, even if the LLM encodes the desired information. In this paper, we study the impact of prompt syntax on the knowledge retrieval capacity of LLMs. We use a template-based approach to paraphrase simple prompts into prompts with a more complex grammatical structure. We then analyse the LLM performance for these structurally different but semantically equivalent prompts. Our study reveals that simple prompts work better than complex forms of sentences. The performance across the syntactical variations for simple relations (1:1) remains best, with a marginal decrease across different typologies. These results reinforce that simple prompt structures are more effective for knowledge retrieval in LLMs and motivate future research into the impact of prompt syntax on various tasks.