Language Model Sentence Completion with a Parser-Driven Rhetorical Control Method

TL;DR

This paper introduces a parser-driven decoding method for controlled text generation that guides large language models to produce sentences with specific rhetorical relations without needing to fine-tune the model.

Contribution

It presents a novel CTG algorithm that enforces rhetorical relation adherence during sentence completion without model fine-tuning.

Findings

Effective in guiding LLMs to produce rhetorically coherent sentences

Validated through automatic and human evaluations

Code available on GitHub

Abstract

Controlled text generation (CTG) seeks to guide large language model (LLM) output to produce text that conforms to desired criteria. The current study presents a novel CTG algorithm that enforces adherence toward specific rhetorical relations in an LLM sentence-completion context by a parser-driven decoding scheme that requires no model fine-tuning. The method is validated both with automatic and human evaluation. The code is accessible on GitHub.