# Exploring Intrinsic Language-specific Subspaces in Fine-tuning Multilingual Neural Machine Translation

arXiv: https://arxiv.org/abs/2409.05224 · HTML: https://mlsanity.com/paper/2409.05224/

## Key facts

- No citing papers in the map yet (data through 14 Oct 2026).
- Builds on 26 of its 34 references that are in the map.
- First posted on arXiv 8 Sep 2024.

## Builds on (most cited first, 26 of 26)

- [Language Models are Few-Shot Learners](https://mlsanity.com/paper/2005.14165/) (2020)
- [LoRA: Low-Rank Adaptation of Large Language Models](https://mlsanity.com/paper/2106.09685/) (2021)
- [Prefix-Tuning: Optimizing Continuous Prompts for Generation](https://mlsanity.com/paper/2101.00190/) (2021)
- [fairseq: A Fast, Extensible Toolkit for Sequence Modeling](https://mlsanity.com/paper/1904.01038/) (2019)
- [Transformer Feed-Forward Layers Are Key-Value Memories](https://mlsanity.com/paper/2012.14913/) (2020)
- [No Language Left Behind: Scaling Human-Centered Machine Translation](https://mlsanity.com/paper/2207.04672/) (2022)
- [BERT Rediscovers the Classical NLP Pipeline](https://mlsanity.com/paper/1905.05950/) (2019)
- [Intrinsic Dimensionality Explains the Effectiveness of Language Model Fine-Tuning](https://mlsanity.com/paper/2012.13255/) (2020)
- [To prune, or not to prune: exploring the efficacy of pruning for model compression](https://mlsanity.com/paper/1710.01878/) (2017)
- [Beyond English-Centric Multilingual Machine Translation](https://mlsanity.com/paper/2010.11125/) (2020)
- [The FLORES-101 Evaluation Benchmark for Low-Resource and Multilingual Machine Translation](https://mlsanity.com/paper/2106.03193/) (2021)
- [Google's Multilingual Neural Machine Translation System: Enabling Zero-Shot Translation](https://mlsanity.com/paper/1611.04558/) (2016)
- [Measuring the Intrinsic Dimension of Objective Landscapes](https://mlsanity.com/paper/1804.08838/) (2018)
- [AdaLoRA: Adaptive Budget Allocation for Parameter-Efficient Fine-Tuning](https://mlsanity.com/paper/2303.10512/) (2023)
- [VeRA: Vector-based Random Matrix Adaptation](https://mlsanity.com/paper/2310.11454/) (2023)
- [Lifting the Curse of Multilinguality by Pre-training Modular Transformers](https://mlsanity.com/paper/2205.06266/) (2022)
- [Toward Multilingual Neural Machine Translation with Universal Encoder and Decoder](https://mlsanity.com/paper/1611.04798/) (2016)
- [Investigating Multilingual NMT Representations at Scale](https://mlsanity.com/paper/1909.02197/) (2019)
- [Learning Language Specific Sub-network for Multilingual Machine Translation](https://mlsanity.com/paper/2105.09259/) (2021)
- [Parameter Sharing Methods for Multilingual Self-Attentional Translation Models](https://mlsanity.com/paper/1809.00252/) (2018)
- [Exploring Universal Intrinsic Task Subspace via Prompt Tuning](https://mlsanity.com/paper/2110.07867/) (2021)
- [Multilingual Machine Translation: Closing the Gap between Shared and Language-specific Encoder-Decoders](https://mlsanity.com/paper/2004.06575/) (2020)
- [Neuron Specialization: Leveraging intrinsic task modularity for multilingual machine translation](https://mlsanity.com/paper/2404.11201/) (2024)
- [Learning Language-Specific Layers for Multilingual Machine Translation](https://mlsanity.com/paper/2305.02665/) (2023)
- [Towards Higher Pareto Frontier in Multilingual Machine Translation](https://mlsanity.com/paper/2305.15718/) (2023)
- [Clustering Pseudo Language Family in Multilingual Translation Models with Fisher Information Matrix](https://mlsanity.com/paper/2312.02820/) (2023)

## Abstract (arXiv)

Multilingual neural machine translation models support fine-tuning hundreds of languages simultaneously. However, fine-tuning on full parameters solely is inefficient potentially leading to negative interactions among languages. In this work, we demonstrate that the fine-tuning for a language occurs in its intrinsic language-specific subspace with a tiny fraction of entire parameters. Thus, we propose language-specific LoRA to isolate intrinsic language-specific subspaces. Furthermore, we propose architecture learning techniques and introduce a gradual pruning schedule during fine-tuning to exhaustively explore the optimal setting and the minimal intrinsic subspaces for each language, resulting in a lightweight yet effective fine-tuning procedure. The experimental results on a 12-language subset and a 30-language subset of FLORES-101 show that our methods not only outperform full-parameter fine-tuning up to 2.25 spBLEU scores but also reduce trainable parameters to 0.4% for high and medium-resource languages and 1.6% for low-resource ones.