The Devil Is in the Word Alignment Details: On Translation-Based Cross-Lingual Transfer for Token Classification Tasks

Translation-based strategies for cross-lingual transfer (XLT) such as translate-traintraining on noisy target language data translated from the source language-and translatetest-evaluating on noisy source language data translated from the target language-are competitive XLT baselines. In XLT for token classification tasks, however, these strategies include label projection, the challenging step of mapping the labels from each token in the original sentence to its counterpart(s) in the translation. Although word aligners (WAs) are commonly used for label projection, the low-level design decisions for applying them to translation-based XLT have not been systematically investigated. Moreover, recent markerbased methods, which project labeled spans by inserting tags around them before (or after) translation, claim to outperform WAs in label projection for XLT. In this work, we revisit WAs for label projection, systematically investigating the effects of low-level design decisions on token-level XLT: (i) the algorithm for projecting labels between (multi-)token spans, (ii) filtering strategies to reduce the number of noisily mapped labels, and (iii) the pre-tokenization of the translated sentences. We find that all of these substantially impact translation-based XLT performance and show that, with optimized choices, XLT with WA offers performance at least comparable to that of markerbased methods. We then introduce a new projection strategy that ensembles translate-train and translate-test predictions and demonstrate that it substantially outperforms the markerbased projection. Crucially, we show that our proposed ensembling also reduces sensitivity to low-level WA design choices, resulting in more robust XLT for token classification tasks. We first detail our pipeline for label projection with WA focusing on the low-level design choices we investigate: mapping labeled spans, filtering strategies, and pre-tokenization (García-Ferrero et al., 2022) . We then describe our new translation-based XLT approach for token-level tasks that ensembles T-Train and T-Test. Label Projection with WA: Design Choices Translation-based XLT entails two paradigms: (1) creating noisy target language training data by