Project leader: Thomas Gammeltoft-Hansen, University of Copenhagen
Project duration: 2021-2026
Participating countries: Denmark, Norway and Sweden
Funding from NordForsk: 14,999,677 NOK
Project website: Nordic Asylum Law & Data Lab
Despite decades of legal harmonisation, the chance of receiving asylum for people from the same country or groups still varies significantly across Nordic and European countries. This has led to repeated accusations of legal decision-making as a “refugee roulette” or “asylum lottery”. Based on a unique dataset and an interdisciplinary team of scholars from law, computer science and social medicine in Denmark, Norway and Sweden the NordASIL project set out to explore what determines asylum outcomes, and why decision-making patterns are so different from country to country.
Impact story
Asylum adjudication is characterised by a lack of hard evidence and decisions often hinge exclusively on the applicant’s own testimony and assessment of her credibility. This places a great emphasis on asylum applicants to convince authorities about the veracity of their claim but also on the decision-makers not to succumb to flawed assumptions, bias, stereotyping, and discrimination.
“For decades legal decision-making in asylum has operated as a kind of ‘human black box’, which was very difficult for scholars to probe from the outside. Our dataset has opened up entirely new possibilities for empirically investigating these dynamics and point to specific areas of concern. It’s been an arduous and years-long process to obtain access to this highly sensitive data, but it’s paid off not only in terms of creating important new insights in the asylum field, but also opening up for much broader transfers of legal data and multiple major follow-up grants for members of our team. It’s been one of the hardest projects I have ever been involved in, but potentially also the one with the most significant long-term impact,” says Thomas Gammeltoft-Hansen.
Key findings
- The researchers built the first ever large-scale Nordic dataset of asylum decisions involving more than 850.000 individual case files from Denmark, Norway and Sweden. The dataset is unique in providing not only legal summaries but also information on case handling, transcripts of interviews and supplementary information such as medical reports. As such it provides the most in-depth resource for understanding and exploring legal decision-making in this area anywhere in the world, and an ideal platform for conducting comparative computational analysis of high value not only for the present but also several follow-up research projects.
- There is a rich but fragmented literature on asylum decision-making across disciplines. The dataset allowed the researchers to test many different hypotheses to validate, nuance or debunk previous findings. For example, decision-maker identity, political climate and gender dynamics have been shown to impact outcomes in other countries, but appear to play a less significant role in the Nordic context. Comparisons with data from other countries also showed that Nordic countries tend to have more coherent decision-making internally, suggesting that Nordic models with e.g. specialised panels and multiple decision-makers for the same case help even out variations.
- The project allowed the researchers to develop new, generally applicable methods for analysing large datasets of legal case files. Using deep neural networks and natural language processing we developed the best performing prediction model of asylum outcomes to date. Adapting methods from image classification similarly enabled the researchers to create word-level heatmaps in the legal decisions, thus providing an important stepping stone for implementing Explainable AI (XAI) in legal text domains.
Key outputs
- An interdisciplinary Nordic lab and training ground for early career scholars working on AI and law, several of which have gone on to obtain competitive grants and follow-up positions.
- A testing ground for producing new knowledge and engaging interdisciplinary dialogue across law, social science, medicine, psychology and computer science from an evidence-based perspective.
- New and cutting-edge methods for computational analysis of legal decision-making that has already been applied in other legal domains and countries beyond the Nordics.
- Publications from the project: https://asylumdata.ku.dk/publications