The language of prompting: What linguistic properties make a prompt successful?

TL;DR

This paper systematically examines how various linguistic properties of prompts affect the performance of large language models, revealing that common assumptions about prompt optimization are often invalid and proposing a new evaluation standard.

Contribution

It provides a comprehensive analysis of linguistic factors influencing prompt effectiveness and challenges existing beliefs, offering insights for more robust prompt design and evaluation.

Findings

Performance varies with linguistic structure, not just perplexity or frequency.

Prompts transfer poorly across datasets and models.

Optimal prompts are not necessarily those with lower perplexity.

Abstract

The latest generation of LLMs can be prompted to achieve impressive zero-shot or few-shot performance in many NLP tasks. However, since performance is highly sensitive to the choice of prompts, considerable effort has been devoted to crowd-sourcing prompts or designing methods for prompt optimisation. Yet, we still lack a systematic understanding of how linguistic properties of prompts correlate with task performance. In this work, we investigate how LLMs of different sizes, pre-trained and instruction-tuned, perform on prompts that are semantically equivalent, but vary in linguistic structure. We investigate both grammatical properties such as mood, tense, aspect and modality, as well as lexico-semantic variation through the use of synonyms. Our findings contradict the common assumption that LLMs achieve optimal performance on lower perplexity prompts that reflect language use in pretraining or instruction-tuning data. Prompts transfer poorly between datasets or models, and performance cannot generally be explained by perplexity, word frequency, ambiguity or prompt length. Based on our results, we put forward a proposal for a more robust and comprehensive evaluation standard for prompting research.