Discovering Language-neutral Sub-networks in Multilingual Language Models
This addresses the challenge of improving cross-lingual transfer for NLP applications, but it is incremental as it builds on existing multilingual models and the lottery ticket hypothesis.
The paper tackled the problem of understanding language-neutral representations in multilingual language models by conceptualizing neutrality as overlap between language-encoding sub-networks, and found that sub-networks for different languages are topologically similar, enabling effective cross-lingual transfer with limited degradation.
Multilingual pre-trained language models transfer remarkably well on cross-lingual downstream tasks. However, the extent to which they learn language-neutral representations (i.e., shared representations that encode similar phenomena across languages), and the effect of such representations on cross-lingual transfer performance, remain open questions. In this work, we conceptualize language neutrality of multilingual models as a function of the overlap between language-encoding sub-networks of these models. We employ the lottery ticket hypothesis to discover sub-networks that are individually optimized for various languages and tasks. Our evaluation across three distinct tasks and eleven typologically-diverse languages demonstrates that sub-networks for different languages are topologically similar (i.e., language-neutral), making them effective initializations for cross-lingual transfer with limited performance degradation.