Multilingual Chatbot for E-Commerce: Data Generation and Machine Translation

Authors

Agnieszka Wołk, Polish-Japanese Academy of Information TechnologyFollow
Helena Skowrońska, NextsellFollow
Ida Skubis, Jan Dlugosz University, PolandFollow

Paper Type

FP

Paper Number

82

Abstract

In a world still rushing forward, waiting is a waste of time, opportunity and money. In response to this market need, we are building a chatbot allowing e-commerce shops to offer 24/7 customer support at a fraction of the cost. Cost efficiency is a major issue when implementing dialogue agents for inflectionally rich languages, such as Polish. That is why end-to-end systems use machine translation (MT) as a pivot in communication, whereas the chatbot itself operates in English. In our research, we cover bidirectional translation between Polish, English, German, and Italian. We achieve good quality results that exceed 48 BLEU for EN– DE. We show how to generate significant amounts of speech resources in the form of dialogues and use them in MT. We believe our findings are replicable in other languages.

Recommended Citation

Wołk, Agnieszka; Skowrońska, Helena; and Skubis, Ida, "Multilingual Chatbot for E-Commerce: Data Generation and Machine Translation" (2021). PACIS 2021 Proceedings. 232.
https://aisel.aisnet.org/pacis2021/232