Language-talent AI benefits many industries

"I am investigating how AI models can be trained in different languages and dialects simultaneously, and how they can be continuously updated with new information to remain relevant to the needs of society," says Evangelia Gogoulou.

A language model is an AI model of text in natural languages. ChatGPT is an example that learnt not only language itself but also the content of the texts it was trained on. The challenge is that large language models require a lot of training data, computational power, and large investments. The models are trained on data collected over a specific period, which can lead to missing information. A clear example of the need for continuous updating is the lack of information in AI models trained before the pandemic.  The challenge is to capitalize on the fact that humans continue to generate information, plus society's growing need to manage multilingualism.

"Just like humans, it is easier for the computer to learn languages in the same language family. Nevertheless, we could see in an experiment with Arabic and English, which are far from similar, that the models learn a pattern that it can reuse," Evangelia reveals.

What do you want to achieve in your thesis?

"I hope to produce some methods and results on how to train models in different languages in an efficient way and with less computational power."

The results of the research could be of great importance to many sectors, such as government, industry, law enforcement and health care. Language models can be used to summarize, translate, and classify texts, helping to make sense of many texts in different languages. This can be useful for specific professions, such as lawyers and doctors who need a quick overview of history and cases. In addition, it can help authorities to fulfil the requirement for information in different languages, help companies to sort and translate documents into different languages. Classification of text can be used to improve services and products or to efficiently send specific customer queries to the right person.

"Language technology is a very exciting technology with many opportunities. Through our research, we are building expertise that is relevant to RISE in the future," concludes Evangelia.