Prompt Segmentation and Annotation Optimisation: Controlling LLM Behaviour via Optimised Segment-Level Annotations

TL;DR

This paper introduces PSAO, a structured prompt optimisation framework that decomposes prompts into segments with annotations to improve LLM response quality and controllability.

Contribution

It proposes a novel method for segmenting prompts and adding annotations, enhancing prompt optimisation efficiency and effectiveness compared to unstructured approaches.

Findings

Optimised segment-level annotations improve reasoning accuracy.

Annotations enhance self-consistency in LLM responses.

Retaining the original prompt as a candidate prevents performance loss.

Abstract

Prompt engineering is crucial for effective interaction with generative artificial intelligence systems, yet existing optimisation methods often operate over an unstructured and vast prompt space, leading to high computational costs and potential distortions of the original intent. We introduce Prompt Segmentation and Annotation Optimisation (PSAO), a structured prompt optimisation framework designed to improve prompt optimisation controllability and efficiency. PSAO decomposes a prompt into interpretable segments (e.g., sentences) and augments each with human-readable annotations (e.g., {not important}, {important}, {very important}). These annotations guide large language models (LLMs) in allocating focus and clarifying confusion during response generation. We formally define the segmentations and annotations and demonstrate that optimised segment-level annotations can lead to improved LLM responses, with the original prompt retained as a candidate in the optimisation space to prevent performance degradation. Empirical evaluations indicate that PSAO benefits from annotations in terms of improved reasoning accuracy and self-consistency. However, developing efficient methods for identifying optimal segmentations and annotations remains challenging and is reserved for future investigation. This work is intended as a proof of concept, demonstrating the feasibility and potential of segment-level annotation optimisation.