If the Maltese language were ever to be at risk of dying, the technology being developed for the first Maltese chatbot could come to the rescue.

A University of Malta team is currently ‘training’ a basic language model to converse with clients of banking, finance, insurance and investment service providers.

The €3 million project is a collaboration between researchers from the Faculty of Information and Communication Technology, the government and the private sector.

Artificial intelligence senior lecturer Claudia Borg told Times of Malta her team was not reinventing the wheel but had started off with technologies that have been proven to work for the English language.

The team used such technology to train an existing generic language model called BERTu, which was developed through funding allocated for the AI Strategy by the Malta Digital Innovation Authority.

“BERTu is a model with a basic understanding of the Maltese language, and which can be developed into various computer programs, including, among others, a chatbot.

“If I try to converse with BERTu, and ask it, for example, to guide me to open a bank account, it would not know how to answer me. Through our research, we built data that would ‘train’ a chatbot version of BERTu to carry out a conversation.

“The more interactions this chatbot version of BERTu has with people, the better it gets. Right now, if I ask it about a banking account – even if I use broken Maltese – it is likely to provide an answer. But if I ask it about the weather, it won’t understand me… yet.”

In order to ‘train’ BERTu, the team created what is known as ‘user stories’, which are the type of questions that people may ask when seeking help for banking, finance, insurance and investment services.

“We created multiple versions of each question in correct Maltese, such as Liema dokumenti taċċetta għall-ftuħ ta’ kont?, X’dokumenti għandi bżonn biex niftaħ kont? and X’tip ta’ dokumenti rrid biex niftaħ kont?

“We then included some questions with grammatical errors and spelling mistakes, such as Xandi bzonn bix niftah accout? and Nixiteq niftah account. Xirid namel? and the model still performed well.”

Borg said that so-called broken Maltese, or a mix of Maltese and English, was a reality of a bilingual society, so the researchers could not limit themselves to correct Maltese.

“We are experimenting with various technologies to see how we can improve models to understand and process the Maltese language. While there are several aims for our research, one of the main aims is to preserve a language in the digital age.

“Unless we create technologies and tools that work for the Maltese language, only the spoken form of the language will remain… and eventually Maltese might not even be spoken anymore. That’s what happens when you don’t have the tools to process a language.”

The first prototypes of the chatbot will primarily operate in the field of financial services, however, the chatbot could continue being trained to eventually be deployed in other sectors, including public service, tourism, customer services and e-commerce services.

The project, called The New Era of Chatbot, is supported by Malta Enterprise under the Research and Development Scheme.

It constitutes a collaboration between Cartesio Ltd and the University of Malta, in partnership with AI and cloud technology enablers Noovle International.