Generalization Measures for Zero-Shot Cross-Lingual Transfer
This work addresses the lack of generalization metrics for language models in cross-lingual settings, which is an incremental improvement for researchers and practitioners in multilingual NLP.
The paper tackles the problem of evaluating language model generalization in zero-shot cross-lingual transfer, proposing efficient measures including a novel algorithm for computing sharpness in the loss landscape that correlates with generalization success.
A model's capacity to generalize its knowledge to interpret unseen inputs with different characteristics is crucial to build robust and reliable machine learning systems. Language model evaluation tasks lack information metrics about model generalization and their applicability in a new setting is measured using task and language-specific downstream performance, which is often lacking in many languages and tasks. In this paper, we explore a set of efficient and reliable measures that could aid in computing more information related to the generalization capability of language models in cross-lingual zero-shot settings. In addition to traditional measures such as variance in parameters after training and distance from initialization, we also measure the effectiveness of sharpness in loss landscape in capturing the success in cross-lingual transfer and propose a novel and stable algorithm to reliably compute the sharpness of a model optimum that correlates to generalization.