Cross-Cultural Transfer Learning for Text Classification

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 3873–3883, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 3873 Cross-Cultural Transfer Learning for Text Classification Dor Ringel1 Gal Lavee2 Ido Guy2 Kira Radinsky1 1Technion - Israel Institute of Technology 2eBay Research {dorringel,kirar}@cs.technion.ac.il idoguy@acm.org glavee@ebay.com Abstract Large training datasets are required to achieve competitive performance in most natural lan- guage tasks. The acquisition process for these datasets is labor intensive, expensive, and time consuming. This process is also prone to hu- man errors. In this work, we show that cross- cultural differences can be harnessed for nat- ural language text classification. We present a transfer-learning framework that leverages widely-available unaligned bilingual corpora for classification tasks, using no task-specific data. Our empirical evaluation on two tasks – formality classification and sarcasm detec- tion – shows that the cross-cultural difference between German and American English, as manifested in product review text, can be ap- plied to achieve good performance for formal- ity classification, while the difference between Japanese and American English can be ap- plied to achieve good performance for sarcasm detection – both without any task-specific la- beled data. 1 Introduction The collection of large, task-specific labeled datasets poses a major challenge to the applica- tion of supervised text classification in many do- mains. Acquiring these datasets is expensive, time-consuming, and error-prone. In this work, we propose leveraging bilingual datasets for classification tasks. The large-scale availability of such datasets counteracts the low availability of task-specific labeled data. The cross-cultural differences expressed in these bilin- gual datasets can be learned and then transferred for text classification tasks with no labeled data. Our work, thus, extends the idea of distant super- vision (Mintz et al., 2009) to tasks for which no relevant large-scale curated datasets exist. Culture was defined by Hofstede et al. (2010) as “the collective programming of the mind, which distinguishes the members of one group of peo- ple from another.” Shweder et al. (2006) defines membership in a cultural group as “thinking and acting in a certain way, in the light of particular goals, values, pictures of the world.” Based on these definitions we can reasonably expect mem- bers of a specific cultural group to think and act in ways that are distinct from others. For example, Hedderich (2010), who investigated cross-cultural differences at the workplace, found that the over- all interaction of employees in Germany is more formal than their American counterparts. House (1997) found that “German subjects tended to in- teract in ways that were more direct, more explicit, more self-referenced, and more content-oriented”. Similarly, several works have studied cross- cultural differences between Japanese and English (both American and British variants) (Koga and Pearson, 1992; Minami, 1994; Martinsons, 2001; Adachi, 2010). Specifically, Adachi (1996) and Ziv (1988) investigated sarcasm in Japanese and the latter showed that American students are more sarcastic than their Japanese counterparts. The main thesis underlying this work is …

Content truncated to 500 words