How to Tune a Multilingual Encoder Model for Germanic Languages: A Study of PEFT, Full Fine-Tuning, and Language Adapters

TL;DR

This study compares different fine-tuning methods for multilingual models on Germanic languages, revealing that parameter-efficient fine-tuning works best for high-resource languages like German, but results vary for Swedish and Icelandic.

Contribution

It provides a comprehensive comparison of full fine-tuning and PEFT methods across multiple Germanic languages and tasks, highlighting their relative effectiveness and limitations.

Findings

PEFT outperforms full fine-tuning for German in some cases

Full fine-tuning is better for named entity recognition tasks

Adding PEFT modules trained on unstructured text does not improve performance

Abstract

This paper investigates the optimal use of the multilingual encoder model mDeBERTa for tasks in three Germanic languages -- German, Swedish, and Icelandic -- representing varying levels of presence and likely data quality in mDeBERTas pre-training data. We compare full fine-tuning with the parameter-efficient fine-tuning (PEFT) methods LoRA and Pfeiffer bottleneck adapters, finding that PEFT is more effective for the higher-resource language, German. However, results for Swedish and Icelandic are less consistent. We also observe differences between tasks: While PEFT tends to work better for question answering, full fine-tuning is preferable for named entity recognition. Inspired by previous research on modular approaches that combine task and language adapters, we evaluate the impact of adding PEFT modules trained on unstructured text, finding that this approach is not beneficial.