Prefix Text as a Yarn: Eliciting Non-English Alignment in Foundation Language Model

TL;DR

This paper introduces PreTTY, a training-free method that uses minimal prior tokens to improve cross-lingual performance of foundation LLMs, matching supervised fine-tuning without additional training.

Contribution

The paper proposes PreTTY, a novel training-free approach that enhances multilingual LLM alignment using minimal prior tokens, reducing costs and data requirements.

Findings

PreTTY achieves comparable performance to SFT in cross-lingual tasks.

Decoding with one or two prior tokens suffices for effective alignment.

PreTTY offers a cost-effective alternative to supervised fine-tuning.

Abstract

While supervised fine-tuning (SFT) has been a straightforward approach for tailoring the output of foundation large language model (LLM) to specific preferences, concerns have been raised about the depth of this alignment, with some critiques suggesting it is merely "superficial". We critically examine this hypothesis within the scope of cross-lingual generation tasks, proposing that the effectiveness of SFT may be constrained by its reliance on prior tokens to guide cross-lingual generation. Based on this crucial insight, and in response to the challenges posed by the costly and limited availability of non-English data for SFT, we introduce a novel training-free alignment method named PreTTY, which employs minimal task-related prior tokens to bridge the foundation LLM and the SFT LLM, achieving comparable performance without training. Experiments on machine translation and part-of-speech tagging across eight languages demonstrate the efficacy of PreTTY in cross-lingual settings. Remarkably, by initiating the decoding process with only one or two prior tokens, foundation LLMs can achieve performance comparable to their SFT counterparts. This method presents a cost-effective alternative to SFT and advances the democratization of multilingual LLMs.