The Intersection of AI and Language: An Interview with Dr. Lea Frermann
In today’s world, artificial intelligence (AI) is revolutionizing many fields, and language
research is no exception. We recently had the pleasure of sitting down with Dr. Lea
Frermann, senior lecturer in Natural Language Processing (NLP) at the University of
Melbourne. Her interdisciplinary research spans cognitive science, computational
linguistics, and social sciences, giving her a unique perspective on how AI is reshaping our
understanding of human language and communication.
Exploring the Complexity of Language through NLP
Dr. Frermann’s academic journey began in linguistics and evolved into a fascinating
intersection with AI. She explained that her research seeks to uncover how humans learn
and represent language. By leveraging NLP methodologies, Dr. Frermann studies a wide range of topics, from cognitive science, such as how narratives shape understanding, to social science issues like media bias and and the linguistic differences in the communication of complex issues across different cultures or ideologies.
“My recent work focuses on media framing and bias in communication around complex
issues like climate change,” she explained. “Combining NLP methods with established social science theories and frameworks, I can scale these studies to much larger data sets, providing a more comprehensive understanding than traditional methods
of manual coding could.”
The Role of Large Language Models in Research
As AI technologies, particularly large language models (LLMs), continue to evolve, they
bring both opportunities and challenges to linguistic research. Dr. Frermann pointed out
that many traditional tasks like translation and summarization are now highly automated in
high-resource languages like English and German. However, the situation is not the same
for low-resource languages.
“LLMs can be an immensely useful tool, helping us label data on a much larger scale,” Dr.
Frermann said. “But this has to be done carefully, as these models come with their own
biases. While they provide insights into both human and machine language processing, wemust continue to critically evaluate where human language understanding differs from
what LLMs can replicate.”
Concerns Over Bias in AI Models
One of the pressing concerns in the field is the potential for AI models to perpetuate
biases, both in research and in practical applications like education. Dr. Frermann
acknowledged these risks, especially in language learning and cultural representation,
where LLMs trained predominantly on English or other high-resource languages might
misrepresent minority languages and cultures. “Even if we could translate all the available English data into a low-resource language like Swahili, the language model might still ‘speak’ like an English
person, embedding cultural biases into its output,” she noted. This, she believes, is a significant area where further research is needed, and careful depolyment and regulation are crucial to balance the need for broad access to LLM technology with the risk of losing cultural diversity.
The Importance of Interdisciplinary Collaboration
Dr. Frermann stressed the value of interdisciplinary collaboration in addressing these
challenges. She regularly works with experts in cognitive science, sociology, and other
fields to ensure her research asks the right questions and produces meaningful results.
“Tackling issues like bias in machine learning requires collaboration beyond just
engineering—domain experts who understand the human side of these problems are
essential,” she said.
Encouraging Diversity in AI and Language Research
As a leader in a male-dominated field, Dr. Frermann is passionate about mentoring the
next generation of female researchers. “I’ve benefited immensely from mentors, and I
encourage young female researchers to seek out guidance , and for established researchers — both male and female — to create these opportunities.” she said.
Dr. Frermann also emphasized the needthe need to tackle structural problems, e.g., fostering interest in STEM disciplines early on and create a work and research environment that values diversity — be it in gender, culture, domain expertise or otherwise.