Control Large Language Models via Divide and Conquer

TL;DR

This paper explores controlling large language models through prompt-based lexical constraints, identifies key limitations, and proposes a Divide and Conquer strategy that significantly improves success rates in constrained text generation.

Contribution

It introduces a novel Divide and Conquer Generation method that enhances LLMs' ability to satisfy lexical constraints, addressing key limitations in prompt-based control.

Findings

Over 90% improvement in success rate for challenging LCG tasks

Identified position bias, low responsiveness to decoding parameters, and handling complex constraints as key limitations

Proposed strategy works for both white-box and black-box LLMs

Abstract

This paper investigates controllable generation for large language models (LLMs) with prompt-based control, focusing on Lexically Constrained Generation (LCG). We systematically evaluate the performance of LLMs on satisfying lexical constraints with prompt-based control, as well as their efficacy in downstream applications. We conclude that LLMs face significant challenges in consistently satisfying lexical constraints with prompt-based control. We identified three key limitations of LLMs for LCG, including (1) position bias, where LLMs tend to satisfy constraints that appear in specific positions within the input; (2) low responsiveness to decoding parameters, which render minimal impact on control of LLMs; and (3) struggle with handling the inherent complexity of certain constraints (e.g., compound words). To address these issues, we introduce a Divide and Conquer Generation strategy, effective for both white-box and black-box LLMs, to enhance LLMs performance in LCG tasks, which demonstrates over 90% improvement on success rate in the most challenging LCG task. Our analysis provides valuable insights into the performance of LLMs in LCG with prompt-based control, and our proposed strategy offers a pathway to more sophisticated and customized text generation applications.