Parameter-Efficient Quality Estimation via Frozen Recursive Models

TL;DR

This paper explores the transferability of Tiny Recursive Models to Quality Estimation for low-resource languages, finding that frozen pretrained embeddings with weight sharing offer a parameter-efficient alternative to fine-tuning, with competitive performance.

Contribution

It demonstrates that recursive mechanisms do not transfer well to QE, and shows that frozen pretrained embeddings with shared weights achieve similar results with significantly fewer trainable parameters.

Findings

Frozen embeddings match fine-tuned performance.

Recursive mechanisms do not improve QE.

Parameter reduction by 37× with frozen embeddings.

Abstract

Tiny Recursive Models (TRM) achieve strong results on reasoning tasks through iterative refinement of a shared network. We investigate whether these recursive mechanisms transfer to Quality Estimation (QE) for low-resource languages using a three-phase methodology. Experiments on $8$ language pairs on a low-resource QE dataset reveal three findings. First, TRM's recursive mechanisms do not transfer to QE. External iteration hurts performance, and internal recursion offers only narrow benefits. Next, representation quality dominates architectural choices, and lastly, frozen pretrained embeddings match fine-tuned performance while reducing trainable parameters by 37$\times$ (7M vs 262M). TRM-QE with frozen XLM-R embeddings achieves a Spearman's correlation of 0.370, matching fine-tuned variants (0.369) and outperforming an equivalent-depth standard transformer (0.336). On Hindi and Tamil, frozen TRM-QE outperforms MonoTransQuest (560M parameters) with 80$\times$ fewer trainable parameters, suggesting that weight sharing combined with frozen embeddings enables parameter efficiency for QE. We release the code publicly for further research. Code is available at https://github.com/surrey-nlp/TRMQE.