Do Multilingual Neural Machine Translation Models Contain Language Pair Specific Attention Heads?
This analysis provides insights into model efficiency for researchers in multilingual NLP, though it is incremental as it builds on existing attention analysis work.
The paper investigates whether multilingual neural machine translation models contain attention heads specific to particular language pairs, finding that the most important heads are similar across languages and up to one-third of less important heads can be removed with minimal impact on translation quality.
Recent studies on the analysis of the multilingual representations focus on identifying whether there is an emergence of language-independent representations, or whether a multilingual model partitions its weights among different languages. While most of such work has been conducted in a "black-box" manner, this paper aims to analyze individual components of a multilingual neural translation (NMT) model. In particular, we look at the encoder self-attention and encoder-decoder attention heads (in a many-to-one NMT model) that are more specific to the translation of a certain language pair than others by (1) employing metrics that quantify some aspects of the attention weights such as "variance" or "confidence", and (2) systematically ranking the importance of attention heads with respect to translation quality. Experimental results show that surprisingly, the set of most important attention heads are very similar across the language pairs and that it is possible to remove nearly one-third of the less important heads without hurting the translation quality greatly.