[Paper Review] Should We Respect LLMs? A Cross-Lingual Study on the Influence of Prompt Politeness on LLM Performance

Ziqi Yin, Hao Wang+3 more|arXiv (Cornell University)|Feb 22, 2024

Artificial Intelligence in Law9 citations

TL;DR

The paper studies how prompt politeness affects LLM performance across English, Chinese, and Japanese tasks, finding that overly polite prompts do not guarantee better results and that the best politeness level varies by language.

ABSTRACT

We investigate the impact of politeness levels in prompts on the performance of large language models (LLMs). Polite language in human communications often garners more compliance and effectiveness, while rudeness can cause aversion, impacting response quality. We consider that LLMs mirror human communication traits, suggesting they align with human cultural norms. We assess the impact of politeness in prompts on LLMs across English, Chinese, and Japanese tasks. We observed that impolite prompts often result in poor performance, but overly polite language does not guarantee better outcomes. The best politeness level is different according to the language. This phenomenon suggests that LLMs not only reflect human behavior but are also influenced by language, particularly in different cultural contexts. Our findings highlight the need to factor in politeness for cross-cultural natural language processing and LLM usage.

Motivation & Objective

• Motivate investigation into whether LLMs mirror human politeness norms in prompts.
• Examine how politeness levels in prompts influence LLM performance across multiple languages.
• Identify whether overly polite prompts improve or worsen performance and if optimal politeness is language-dependent.

Proposed method

• Conduct cross-linguistic experiments evaluating LLM responses to prompts with varying politeness levels in English, Chinese, and Japanese.
• Analyze performance differences to determine if impolite prompts degrade results and whether excessive politeness helps.
• Synthesize findings to discuss cultural/cross-lingual implications for prompting in NLP.

Experimental results

Research questions

• RQ1 Does prompt politeness influence LLM performance across languages (English, Chinese, Japanese)?
• RQ2 Is there an optimal level of politeness, and does it vary by language or culture?
• RQ3 Do impolite prompts consistently degrade performance compared to polite prompts?
• RQ4 What are the cross-cultural implications for prompting in multilingual NLP tasks?

Key findings

• Impolite prompts often lead to poorer LLM performance.
• Overly polite prompts do not guarantee improved outcomes.
• The best politeness level varies by language, suggesting language-specific cultural effects on LLM behavior.