Assessing the Role of Lexical Semantics in Cross-lingual Transfer through Controlled Manipulations

While cross-linguistic model transfer is effective in many settings, there is still limited understanding of the conditions under which it works. In this paper, we focus on assessing the role of lexical semantics in cross-lingual transfer, as we compare its impact to that of other language properties. Examining each language property individually, we systematically analyze how differences between English and a target language influence the capacity to align the language with an English pretrained representation space. We do so by artificially manipulating the English sentences in ways that mimic specific characteristics of the target language, and reporting the effect of each manipulation Figure 1: A. Sentences from the UM parallel corpus. In on the quality of alignment with the representation each sentence, the word mind is colored along with its space. We show that while properties translation in Simplified Chinese. B. A weighted graph such as the script or word order only have a which results from the UM corpus. The edge weights limited impact on alignment quality, the degree indicate how many times mind is translated into each of lexical matching between the two languages, instance in Simplified Chinese. C. Calculation of the which we define using a measure of translation translation entropy of the word mind in the UM corpus.