Parameter-Efficient Quality Estimation via Frozen Recursive Models

2026-03-15

AI Summary

This work addresses the challenges of low parameter efficiency and suboptimal performance in quality estimation for low-resource languages by proposing a highly parameter-efficient approach. Specifically, it freezes the XLM-R embeddings and replaces less effective recurrent architectures with a weight-sharing mechanism. This strategy drastically reduces the number of trainable parameters while maintaining or even surpassing the performance of larger, fully fine-tuned models. Experimental results demonstrate a Spearman correlation coefficient of 0.370 across eight language pairs—comparable to that of fine-tuned baselines—with 37 times fewer trainable parameters. Notably, on Hindi and Tamil, the proposed method achieves superior performance over MonoTransQuest using only 1/80 of its parameters, significantly enhancing both efficiency and effectiveness in low-resource quality estimation scenarios.

Technology Category

• Natural Language Processing: Language Grounding & Multi-modal NLP
• Machine Learning: Large Multimodal Models (LMMs)
• Search and Optimization: Learning to Search

Application Category

• Search and Retrieval-Augmented AI: Web evaluation methodologies and metrics
• Economics, Online Markets and Human Computation: LLM based quality controls for crowd work
• Web Mining and Content Analysis: Large pretrained models with web data

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.

Problem

• Quality Estimation
• low-resource languages
• parameter efficiency
• frozen embeddings
• recursive models

Innovation

• Parameter-Efficient Learning
• Quality Estimation
• Frozen Embeddings
• Recursive Models
• Low-Resource Languages