The Obscure Limitation of Modular Multilingual Language Models
This work addresses a specific problem for researchers and practitioners in multilingual NLP, but it is incremental as it focuses on evaluating and improving an existing approach.
The paper tackles the limitation of modular multilingual language models in real-case multilingual inference by exposing how excluding language identification modules obscures performance, and it discusses ways to close the performance gap caused by pipelining these components.
We expose the limitation of modular multilingual language models (MLMs) in multilingual inference scenarios with unknown languages. Existing evaluations of modular MLMs exclude the involvement of language identification (LID) modules, which obscures the performance of real-case multilingual scenarios of modular MLMs. In this work, we showcase the effect of adding LID on the multilingual evaluation of modular MLMs and provide discussions for closing the performance gap of caused by the pipelined approach of LID and modular MLMs.