Two Methods for Domain Adaptation of Bilingual Tasks: Delightfully Simple and Broadly Applicable

Abstract

Bilingual tasks, such as bilingual lexicon induction and cross-lingual classification, are crucial for overcoming data sparsity in the target language. Resources required for such tasks are often out-of-domain, thus domain adaptation is an important problem here. We make two contributions. First, we test a delightfully simple method for domain adaptation of bilingual word embeddings. We evaluate these embeddings on two bilingual tasks involving different domains: cross-lingual twitter sentiment classification and medical bilingual lexicon induction. Second, we tailor a broadly applicable semi-supervised classification method from computer vision to these tasks. We show that this method also helps in low-resource setups. Using both methods together we achieve large improvements over our baselines, by using only additional unlabeled data.

Target-Level Sentiment Analysis on Various Genres

“…2013 (Hangya and Farkas, 2013a) • CogInfoCom 2013 (Hangya and Farkas, 2013b) • ICCIA 2017 • SEMEVAL 2013 (Hangya et al, 2013) • AIRE 2017 • • TSD 2015 (Hangya, 2015) • ACL 2018 (Hangya et al, 2018) • Table 1.1: The relation between the thesis topics and the corresponding publications. In the results presented in Chapters 3, 5 and 6 the author's contribution was prominent while Chapters 4 and 7 are the results of joint works with other researchers.…”

“…We would like to thank Dr. Fabienne Braune, Prof. Dr. Alexander M. Fraser and Prof. Dr. Hinrich Schütze for their invaluable cooperation in the development of this work. The two proposed methods for domain adaptation were developed by the author of this dissertation with the help of invaluable ideas of all authors of the published work of (Hangya et al, 2018). The experiments related to bilingual lexicon induction was performed by Dr. Fabienne Braune.…”

“…The performed experiments showed that the adaptation of embeddings is crucial for the second step and that the proposed methods are task and language independent. Related to his publication (Hangya et al, 2018), the author regards the following results as his main contributions to the field:…”

“…• Dokumentumszintű szentimentelemzés kvantitatív és kvalitatív összehasonlítása különböző doméneken, műfajokon és nyelveken keresztül. (Hangya et al, 2018) alapján a szerző a tézis legfontosabb saját eredményeinek a következőket tekinti:…”

“…We show that high-quality domain-adapted bilingual word embeddings are required in order to use unlabeled data well by the semi-supervised approach. The results of this chapter were published in (Hangya et al, 2018).…”