Consortium investigates the development of chat technology

How can chat technology be used in different languages and for different purposes, with relatively little training and compute power? Arianna Bisazza, assistant professor specialized in computational linguistics and language processing, is part of a large research consortium that will investigate this.

Like a conversation
Many organisations use chatbots to help visitors to their websites and to answer their questions. This is increasingly taking place like a conversation: people put their questions and needs to a system, the system responds naturally, possibly by asking questions for clarification, and meets the needs of its users. But this chat functionality is not developed in every language yet.

Complex research
Transferring chat technology to other languages and for other purposes has not yet been the subject of much research. Bisazza: "A chatbot must understand what someone is chatting about in every interaction and be able to give an appropriate answer, or ask a good question for clarification. Ultimately, a customer needs to be helped via the interaction with the chatbot. To do this well in different languages, not only does the translation have to be good, but you also have to take into account the differences in culture that can affect how two people have a conversation. That makes this research particularly complex."

Applications during the project
In the project, the researchers work together with supermarket chain Albert Heijn, web shop Bol, telecom company KPN and insurer Achmea, among others. The companies provide use cases - specific problems or scenarios that need to be addressed - on the basis of which the researchers can improve the technology. The companies can already implement these technical improvements during the research to improve their customer service. Bisazza: "Working with these different companies - with very different customer contacts - allows us to work with different and realistic use cases, which ultimately makes the technology applicable to more domains and in different languages."

LESSEN
The Dutch Research Council (NWO) recently awarded the consortium a €4.6 million grant from the Dutch Research Agenda (NWA). The consortium, called LESSEN: Chat-based conversational intelligence with few resources, is led by Prof. Maarten de Rijke of the University of Amsterdam and is a collaboration of the University of Amsterdam, Leiden University, University of Groningen, Radboud University, Achmea, Albert Heijn, Bol.com, KPN, Rasa Technologies, Ahold Delhaize and the Dutch national police force.

Commissioned by the Ministry of Education, Culture and Science (OCW), NWO has been funding research in the framework of the Dutch Research Agenda (NWA) since 2018. The aim of the NWA is to make a positive and structural contribution to the global knowledge society of tomorrow, where new knowledge flows easily from researcher to user and where new questions from practice and society are quickly and naturally incorporated into new research.