The stasis of migration discourses 

Professor Charlotte Taylor

One of the key contributions of a conference like AMLI is the way in which it allows us to step outside our own sphere of research and see how migration, language and identity interact and intersect in diverse contexts. And one of the astounding features for a scholar of migration discourses is the extent of overlap we discover as we cross borders of geography, history and perspective. In this keynote, I would like to take the opportunity to pause and reflect on the recurring features of framing migration – from  water metaphors to binary opposition. I will start from a state-of-the-art review and then draw on my own work on historical migration discourses to illustrate and investigate some of these recurring themes and tropes. And I will end by asking, is there nothing new under the proverbial sun? And what are the implications for us as a community of scholars committed to understanding the impact of language on our society?  

Diaspora communities in the Colombian transition to peace 

Associate Professor Catalina Montoya-Londono 

Beyond the polls: Decoding migration sentiment with AI and digital trace data

Professor Francisco Rowe

Abstract

Migration has become one of the most polarising issues in global public discourse, shaping elections, policy agendas, and social cohesion. Yet, our ability to measure migration sentiment and narratives at scale in high geographical and temporal resolutions remains constrained by the limitations of traditional surveys. In this talk, I present a new frontier in migration research, showcasing how artificial intelligence -specifically natural language processing (NLP) and large language models (LLMs)- combined with digital trace data, can transform how we understand migration debates in real time and across languages at high temporal frequency and spatial granularity.

Drawing on over 1.3 million georeferenced tweets spanning more than a decade and 13 languages, I introduce an integrative framework that leverages supervised and unsupervised NLP techniques to detect shifts in migration salience, classify sentiment polarity, and map narrative themes across time and space. The results show strong temporal responses to key events (e.g., COVID-19, Brexit), increasing spatial polarisation, and the emergence of echo chambers dominated by highly active anti-immigration users. These accounts drive rapid, wide dissemination of anti-migration content -often 1.5x faster than pro-migration messages- shaping the public conversation disproportionately.

I also present novel findings on cross-lingual model transfer. LLMs fine-tuned on one or two languages can generalise to classify immigration discourse in unseen languages with high accuracy, especially for identifying topic salience. However, stance classification still benefits from multilingual fine-tuning to overcome pretraining biases. Lightweight, cost-efficient fine-tuning methods such as low-rank adaptation and quantised models enable scalable, inclusive analysis for low-resource settings.

Together, these insights highlight the promise and the perils of using AI and digital data to study migration opinion. They offer new tools to track public attitudes, inform policy, and combat misinformation, while raising ethical questions about algorithmic bias and representativeness in digital platforms. This work calls for an interdisciplinary dialogue on how we can responsibly harness AI to understand and engage with migration narratives in an increasingly digital and multilingual world.