What question did this study set out to answer?

The aim is to explore the effects of prompt design strategies on hallucination rates in large language models across different languages.

July 15, 2026Open Access

Optimizing Prompt Design for Efficient Hallucination Reduction: An Experimental Study on Language, Task Difficulty, and Prompt Length

View Full Paper

Ask AI

Bookmark

Authors

Experimental study analyzes how prompt design affects hallucination rates across multiple languages, highlighting key findings.

The aim is to explore the effects of prompt design strategies on hallucination rates in large language models across different languages.
Developed MultiLang-HalluBench, a multilingual evaluation benchmark with 90 questions in English, Indonesian, and Mandarin Chinese.
Evaluated an LLM using zero-shot, instructed anti-hallucination, and chain-of-thought prompting methods across three difficulty levels.
Conducted a cross-lingual analysis to measure performance gaps in language model outputs.
The transition from zero-shot to instructed prompting led to 86.3% of the total improvement in reducing hallucinations.
Two categories of hallucination were identified: reasoning errors, which can be reduced through prompt engineering, and knowledge errors, which cannot.
Indonesian language performance showed a significant gap that persisted under various prompting conditions.

Jhon Thimoty Junior Sihite (2026) studied this question.

View Full Paper

Ask AI

Bookmark