Learning Language-Specific Layers for Multilingual Machine Translation
This addresses the challenge of maintaining translation quality while scaling multilingual models, which is incremental as it builds on existing transformer architectures.
The paper tackles the problem of reduced model capacity per language in multilingual machine translation by introducing Language-Specific Transformer Layers (LSLs), achieving improvements of 1.3 chrF (1.5 spBLEU) points over a baseline without LSLs in one architecture and 1.9 chrF (2.2 spBLEU) in another.
Multilingual Machine Translation promises to improve translation quality between non-English languages. This is advantageous for several reasons, namely lower latency (no need to translate twice), and reduced error cascades (e.g., avoiding losing gender and formality information when translating through English). On the downside, adding more languages reduces model capacity per language, which is usually countered by increasing the overall model size, making training harder and inference slower. In this work, we introduce Language-Specific Transformer Layers (LSLs), which allow us to increase model capacity, while keeping the amount of computation and the number of parameters used in the forward pass constant. The key idea is to have some layers of the encoder be source or target language-specific, while keeping the remaining layers shared. We study the best way to place these layers using a neural architecture search inspired approach, and achieve an improvement of 1.3 chrF (1.5 spBLEU) points over not using LSLs on a separate decoder architecture, and 1.9 chrF (2.2 spBLEU) on a shared decoder one.