Shapley Head Pruning: Identifying and Removing Interference in Multilingual Transformers
This addresses interference in multilingual models for NLP applications, offering a parameter-efficient solution rather than adding capacity.
The paper tackled interference in multilingual transformers, where adding languages degrades performance, and showed that pruning language-specific attention heads identified via Shapley Values improves target language performance by up to 24.7%.
Multilingual transformer-based models demonstrate remarkable zero and few-shot transfer across languages by learning and reusing language-agnostic features. However, as a fixed-size model acquires more languages, its performance across all languages degrades, a phenomenon termed interference. Often attributed to limited model capacity, interference is commonly addressed by adding additional parameters despite evidence that transformer-based models are overparameterized. In this work, we show that it is possible to reduce interference by instead identifying and pruning language-specific parameters. First, we use Shapley Values, a credit allocation metric from coalitional game theory, to identify attention heads that introduce interference. Then, we show that removing identified attention heads from a fixed model improves performance for a target language on both sentence classification and structural prediction, seeing gains as large as 24.7\%. Finally, we provide insights on language-agnostic and language-specific attention heads using attention visualization.