Announcing the Winners of the CDCS Digital Research Prizes 2024

7 Jun 2024

We are thrilled to announce the winners of this year’s CDCD Digital Research Prizes. Congratulations to all our winners and to all those whose fabulous work was nominated!

Winners of the CDCS Digital Research Prizes 2024 — Event photography by Gintare Kulyte.

Best Small Data-Driven Project: Systemic structure of kinship is shaped by evolutionary processes

Maisy Hallam

This project uses a digital database of kinship terminology to look at cross-linguistic constraints on kinship systems. Kinship terms are used to talk about kin - for example, in English, kinship terminology includes words like "father", "mother", aunt", "cousin", etc. The number of terms and which relatives are categorised together varies across languages. For instance, the English kin term "uncle" groups both parents’ brothers together, but the same relatives are split into three categories in Hindi: cācā (father’s younger brother), tāū (father’s older brother), and māmā (mother’s brother). Despite the fact that different languages do this differently, there seem to be commonalities across languages in how kinship systems work - not all logically possible kinship systems are attested, and some common patterns recur across unrelated languages. Anthropologists studying kinship terminology have known about this since at least the 1970s, but surprisingly their methods for identifying these recurring cross-linguistic regularities are largely qualitative, and in particular they don't typically use rigorous baselines to show that some putative cross-linguistic regularity is actually not just a product of unconstrained random variation. This project brings computational rigour to this question, using a large digital database of kinship terminologies from 500+ languages, quantifies a putative cross-linguistic regularity from the anthropology literature using information theory, and then uses Monte Carlo sampling techniques to verify that the degree of internal co-selection seen in those languages is actually distinct from what you'd expect to see if there was no internal co-selection constraint. This is a creative and innovative approach to a long-standing problem in anthropology; as well as being very powerful the method is surprisingly simple and has great potential to be applied to other claims and by other researchers.

Maisy presented on this work at the Evolang conference in May 2024.

Best Dataset: Beyond Humanitarian Emergencies

Kate Wright, Dani Madrid-Morales

This project challenged the most significant theory within humanitarian communication using a creative blend of computational and manual methods, deployed by interdisciplinary researchers. Craig Calhoun (2004, 2008, 2010) famously argued that Western journalists and aid workers collectively reproduced an “emergency imaginary” via their media coverage of humanitarian crises and related fundraising appeals. Indeed, he argued that this interpretative frame had spread internationally, to become the main way of understanding and organising humanitarian responses in the world. To interrogate Calhoun’s claim, this project constructed a global corpus of Anglophone media, spanning 10 years (2010-2020) of texts including the word “humanitarian”, before training a word-embedding layer to generate and compare country-level estimates of closeness to each dimension of the emergency imaginary. An article on this work is forthcoming.

Find the data here: https://datashare.ed.ac.uk/handle/10283/8739

Best Impact from a Data-Led Project: Joint Winners

Gender-ing ELT: International perspectives, practices, policies

Vander Viana

Collaborators: Aisling O’Boyle (Northern Ireland), Danielle de Almeida Menezes (Brazil), Sibonile Edith Ellece (Botswana), Udi Samanhudi (Indonesia), Anna Chesnokova (Ukraine), Fatima Sadiqi (Morocco), Gina Lontoc (Philippines), Gulnahar Begum (Northern Ireland), Hang Nguyen (Vietnam), Teresita Sevilla (Colombia), Yan Wen (China), Anna Robinson-Pant (England), Gulsah Kutuk (England), Elmakki Amiri (Morocco), Florence Nwaefuna (Botswana), Haiqin Wu (China), Hannah-Gazelle Gabrielle Ponce (Philippines), Hesam Sadeghian (Colombia), Mariana Nunes Monteiro (Brazil), Nika Marushchak (Ukraine), Pamela Joy Mercado (Philippines), Rahmi Amalia (Indonesia), Ranjit Podder (Bangladesh), Tung Vu (Vietnam), and Via Allison Del Rosario (Philippines)

Informed by Goal 5 – gender equality and the empowerment of women/girls – in the United Nations’ Sustainable Development Agenda, this project explored the contribution that English language stakeholders (can) make to promote effective social change in education worldwide, through research into the socially relevant but often overlooked topic of gender equality in ELT across several countries. With national teams in ten countries, six stakeholder groups and four different instruments of data collection, the breadth of influence of the research project is unparalleled.

One of the project’s key feature is its early embedding of community engagement, resulting in the use of data for research and impact purposes. Essays and lesson plans collected throughout have been used to support teacher educators, student teachers and teachers to reconceptualize their current/future practice, and two open-access impact-driven books have arisen from this project. “Gender equality in English language teaching practice” contains 30 reflective essays written by research participants, accompanied by reflection questions, discussion points and practical tasks for teacher education. “Gender-ing English language classes across the globe” encompasses 40 lesson plans, modelling how gender can be part of the work undertaken by teachers. These two books have changed the traditional way in which ELT and English language teacher education are conceptualized and practiced and provide a means through which participants in low-resourced countries can have their voices heard and, as they have pointed out, “feel proud to be a part of this project”.

The Influence Policing Project

Ben Collier, James Stewart, Lydia Wilson, Daniel R Thomas, Shane Horgan

This research constituted the first empirical and theoretical exploration of an entirely novel mode of ‘algorithmic’ policing, which we term ‘influence policing’. This involves the law enforcement use of domestic digital influence campaigns using behavioural advertising for crime prevention. In a 150-page landscape report based on extensive fieldwork with Police Scotland and innovative computational analysis of a new dataset – 12,000 campaign segments and targeting details from the Meta Ad Library – we outlined a distinctive ‘Scottish’ model of influence policing and identified serious emerging harms and ethical issues in its wider use in UK policing.

Widely reported in the press, the findings of the Influence Policing project have actively impacted policy in Police Scotland, the FBI, UK government and National Crime Agency. The team were also invited to, and took part in a Europol sprint developing methods for these campaigns, and separately to the international InterCOP prevention network – via ACOP and InterCOP meetings., and their work has also been referenced as a critical source in a recent OECD Communications Scan of public sector strategic communications in the UK.

Best Novel use of a Digital Method: News Media as a Lens: Exploring Neighbourhood Health through the Analysis of Geoparsed and Clustered Local News

Eleojo Abubakar, Andreas Grivas, Claire Grover, Richard Tobin, Clare Llewellyn, Chunyu Zheng, Chris Dibben, Alan Marshall, Jamie Pearce, Beatrice Alex

The understanding of the multifaceted impact of communities on health remains challenging to define, with limited insight into place-based processes and their implications for health and inequalities. This knowledge gap undermines the development of effective policy interventions aimed at enhancing local health and well-being. This project investigated the innovative use of computational news analysis to complement statistical information collected to analyse neighbourhood health. This project explored whether it is possible to capture aspects of social dimensions and community of place using themes in local newspapers which had previously not been leveraged in place-based health research relying only on statistical information in databases like the Scottish Index of Multiple Deprivation (SIMD).

The project applied established NLP techniques, such as geoparsing via the Edinburgh Geoparser combined with topic modelling using hierarchical clustering, to over 70K news articles reporting on areas in the City of Edinburgh in this novel research setting. The NLP output (news topic distribution of areas) feeds into a regression analysis related to health resilience and vulnerability to explore if computational analysis of local newspaper text can capture social aspects of place that are neglected in SIMD and other area-based measures. To ensure that the news clustering is accurate, researchers performed cluster evaluation and showed that it performs well. As a check, they also showed that news topics, which are already well captured in statistical data, such as ones related to crime or property, correlate as expected with deprivation statistics. More crime is reported for areas with higher deprivation and reporting on the property market increases for areas of lower deprivation. Most notably, a regression-based analysis showed that resilient areas are associated with increases in local press on topics like heritage, community sports (boxing) and managed crime, whereas vulnerable areas are associated with increased local press on topics of littering and anti-social problems, unsolved crime and local politics. These results show that combining NLP of news with statistical analysis can provide insights on local neighbourhood characteristics which can assist in explaining their resilience.

This work has been presented at the GeoExT Workshop at ECIR 2024 in Glasgow.

See the full list of winning and commended projects in the CDCS Digital Research Prizes 2024.

Event photography by Gintare Kulyte.