Improving Cross-Domain Low-Resource Text Generation through LLM Post-Editing: A Programmer-Interpreter Approach
Zhuang Li, Levon Haroutunian, Raj Tumuluri, Philip Cohen, Gholamreza Haffari
TL;DR
This work tackles the challenge of cross-domain, low-resource text generation when updating LLM parameters is impractical. It introduces a programmer-interpreter post-editing framework where a lightweight Programmer encodes task-specific knowledge via edit-action sequences, and a large Interpreter (an LLM) applies those edits to the Generator's output, preserving the LLM's domain generalization. Empirical results on Kashmiri-English MT and AMR-to-English LF-to-Text show that few-shot action prompts in Prog-Refine outperform zero-shot and existing post-editing baselines, with ORACLE actions yielding substantial BLEU improvements. The approach offers a practical, scalable way to enhance LLM-generated text in low-resource, cross-domain settings, though in-domain performance and privacy considerations warrant further investigation.
Abstract
Post-editing has proven effective in improving the quality of text generated by large language models (LLMs) such as GPT-3.5 or GPT-4, particularly when direct updating of their parameters to enhance text quality is infeasible or expensive. However, relying solely on smaller language models for post-editing can limit the LLMs' ability to generalize across domains. Moreover, the editing strategies in these methods are not optimally designed for text-generation tasks. To address these limitations, we propose a neural programmer-interpreter approach that preserves the domain generalization ability of LLMs when editing their output. The editing actions in this framework are specifically devised for text generation. Extensive experiments demonstrate that the programmer-interpreter significantly enhances GPT-3.5's performance in logical form-to-text conversion and low-resource machine translation, surpassing other state-of-the-art (SOTA) LLM post-editing methods in cross-domain settings.