Exploring Representational Disparities Between Multilingual and Bilingual Translation Models
This addresses a performance disparity problem for machine translation researchers, but it is incremental as it focuses on analyzing existing differences rather than proposing a new solution.
The study investigated why multilingual translation models sometimes underperform bilingual models, finding that multilingual decoder representations are less isotropic and occupy fewer dimensions, which is partly due to modeling language-specific information.
Multilingual machine translation has proven immensely useful for both parameter efficiency and overall performance across many language pairs via complete multilingual parameter sharing. However, some language pairs in multilingual models can see worse performance than in bilingual models, especially in the one-to-many translation setting. Motivated by their empirical differences, we examine the geometric differences in representations from bilingual models versus those from one-to-many multilingual models. Specifically, we compute the isotropy of these representations using intrinsic dimensionality and IsoScore, in order to measure how the representations utilize the dimensions in their underlying vector space. Using the same evaluation data in both models, we find that for a given language pair, its multilingual model decoder representations are consistently less isotropic and occupy fewer dimensions than comparable bilingual model decoder representations. Additionally, we show that much of the anisotropy in multilingual decoder representations can be attributed to modeling language-specific information, therefore limiting remaining representational capacity.