A Machine Translation-Powered Chatbot for Public Administration

A Machine Translation-Powered Chatbot for Public Administration Dimitra Anastasiou+, Anders Ruge& Radu Ion*, Svetlana Segărceanu#, George Suciu#, Olivier Pedretti+, Patrick Gratz+, Hoorieh Afkari+ +Luxembourg Institute of Science and Technology, &SupWiz, Denmark, *Romanian Academy Institute for AI, #R&D Department, BEIA, Romania a.ruge@supwiz.com, radu@racai.ro, {svetlana.segarceanu, george}@beia.ro {dimitra.anastasiou, olivier.pedretti, patrick.gratz, hoorieh.afkari}@list.lu Abstract This paper describes a multilingual chatbot developed for public administration within the ENRICH4ALL project. We argue for multilingual chatbots powered through machine translation (MT) and discuss the integration of the eTranslation service in a chatbot solution. 1 Introduction In this paper, we introduce the Action ENRICH4ALL (E-goverNment [RI] CHatbot for ALL) which is about the development of a multilingual chatbot service to be deployed in public administration in Luxembourg, Denmark, and Romania. ENRICH4ALL is funded by the Connecting Europe Facility and its duration is from June 2021 to May 2023. The partners are Luxembourg Institute of Science and Technology, BEIA Consulting Romania, Romanian Academy Institute for AI and SupWiz, Denmark. In this paper, we refer to the benefits and challenges of e- government chatbots and to the integration of eTranslation with the chatbot platform. 2 Related Work The benefits of having e-government chatbots are several: they can process service requests in huge © 2022 The authors. This article is licensed under a Creative Commons 3.0 licence, no derivative works, attribution, CC- BY-ND. numbers, work 24/7, provide up-to-date information and consequently reduce operational costs. In some European countries, such as Denmark, Estonia, and Latvia, there are chatbots used in many public authorities, whereas in other countries, such as Romania or Luxembourg, there are not. Some of the challenges of using chatbots in public administration are the large number of relevant services, the complexity of administrative services, the context-dependent relevance of user questions, the differences in expert-language and user-language as well as the necessity of providing highly reliable answers for all questions (Lommatzsch, 2018). To these challenges, we should add the language diversity in Europe. The consequence of language diversity is that each EU country and each administration uses its own initiative to deploy a chatbot (often monolingual) resulting in a scenario where the interaction with e-government through virtual assistants is scarce and fragmented. 3 A multilingual chatbot in public administration Particularly for administrative procedures, there are many requests from expatriates, who enter a new country. Application for residence, importing a car, starting-up a new business, and building a house are some of such requests. Public administration was also burdened with many questions related to the pandemic, which gave rise to COVID-19 chatbots in Europe. We created (and actively develop) three datasets: • COVID-19 (RO2) • Construction permits (RO) • Administrative questions (LTZ–FR–DE–EN) The datasets are available at the project’s website and will soon be available at the European Language Grid. As for BERT language models, we use already existing ones for RO, FR, DE, EN, and we have developed and trained one for Luxembourgish3 to use for detecting question similarity and classification with user intent labels, but this is outside the scope …

Content truncated to 500 words