Product Review Translation using Phrase Replacement and Attention Guided Noise Augmentation

Product Review Translation using Phrase Replacement and Attention Guided Noise Augmentation Kamal Kumar Gupta, Soumya Chennabasavraj,† Nikesh Garera,† and Asif Ekbal Department of Computer Science and Engineering, Indian Institute of Technology Patna, Patna, India †Flipkart, India kamal.pcs17,asif@iitp.ac.in †soumya.cb,nikesh.garera@flipkart.com Abstract Product reviews provide valuable feedback of the customers, however, they are avail- able today only in English on most of the e-commerce platforms. The nature of reviews provided by customers in any multilingual country poses unique challenges for machine translation such as code-mixing, ungrammatical sentences, presence of colloquial terms, lack of e-commerce parallel corpus etc. Given that 44% of Indian population speaks and operates in Hindi language, we address the above challenges by presenting an English–to–Hindi neural machine translation (NMT) system to trans- late the product reviews available on e-commerce websites by creating an in-domain parallel corpora and handling various types of noise in reviews via two data augmen- tation techniques, viz. (i). a novel phrase augmentation technique (PhrRep) where the syntactic noun phrases in the sentences are replaced by the other noun phrases carrying different meanings but in the similar context; and (ii). a novel attention guided noise augmentation (AttnNoise) technique to make our NMT model robust towards various noise. Evaluation shows that using the proposed augmentation tech- niques we achieve a 6.67 BLEU score improvement over the baseline model. In order to show that our proposed approach is not language-specific, we also perform ex- periments for two other language pairs, viz. En-Fr (MTNT18 corpus) and En-De (IWSLT17) that yield the improvements of 2.55 and 0.91 BLEU points, respectively, over the baselines. 1 Introduction Product reviews written by the users on e-commerce websites are useful to get the feedback about the products and provide valuable insights to the user for making the buying decision. The product reviews available on different e-commerce websites are mainly in English language. India is a multilingual country with great linguistic and cultural diversities. There are 22 officially spoken languages, and many of them such as Hindi, Bengali, etc. come into the top 10 most spoken languages all over in the world. Since English is not a first language in India and most of the population (approximately, 68.9%)1 from the rural areas do not have the proper understanding of English language, 1http://mohua.gov.in/cms/urban-growth.php Proceedings of the 18th Biennial Machine Translation Summit Virtual USA, August 16 - 20, 2021, Volume 1: MT Research Track Page 243 Source (A) osm product.i really love it. osm camera quality...nice one Reference बहुत ब�ढ़या ूॉडक्ट. मुझे यह पसंद है. बहुत ब�ढ़या कै मरा �ा�लट�… अच्छा है (Transliteration) bahut badhiya prodakt. mujhe yah pasand hai. bahut badhiya kaim kvaalitee… achchha hai Gen-NMT ओसम उत्पाद. मैं वास्तव में इसे प्यार करता हूँ. ओसम कै मरा गुणव�ा... अच्छा एक (Transliteration) osam utpaad. main vaastav mein ise pyaar karata hoon. osam kaimara kvaalitee… achchha hai Source (B) NYC product,and cloth quilty is too good Reference अच्छा ूॉडक्ट, और कपड़े क� �ा�लट� बहुत ब�ढ़या है (Transliteration) achchha prodakt, aur kapade kee kvaalitee bahut badhiya hai Gen-NMT NYC …

Content truncated to 500 words