CultureLab Projects

• Automatically Identifying Inter-Cultural Metaphor Variation [ongoing]

  The way that people in different cultures think is encoded in the language they use. In particular, patterns of metaphorical language, known as conceptual metaphors (Lakoff and Johnson, 1980), may offer a lens into the thought patterns of different cultures in different time periods. In this project, we will be exploring how conceptual metaphors vary between languages, both modern and ancient, in order to investigate whether multilingual metaphor differences are predictive of cultural variation. To achieve these analyses, we are developing develop computational methods to identify conceptual metaphors automatically across a diverse collection of languages.

           Principal investigator: Rowan Hall Maudslay

• Computational Approach to the Cultural Evolution of Food [ongoing]

  This project investigates how food culture evolves over time, combining perspectives from computational, historical, and cultural research. It explores how cuisines, including recipes, ingredients, techniques, and tools, emerge, spread, and change across societies and historical periods. In addition, this study seeks to understand how cultural, social, and technological factors influence cuisine. It integrates methods from computational linguistics and agent-based modelling to uncover the mechanisms driving culinary change, adaptation, and innovation. In this respect, this project focuses on three main research axes: classifying and comparing cuisines across time and space, tracing the integration of new food elements into existing traditions, and studying how digital media contribute to the viral spread of recipes and culinary trends. By bridging computational methods with cultural theory, this research aims to build a scientific foundation for understanding the cultural evolution of food and to provide new tools and datasets for future interdisciplinary studies.

            Principal investigator: Alexandre Bluet

• When Fear Goes Viral : Computational Analysis of the Evolution of Horrific Content Online [ongoing]

  Emerging from early 21^st century forums and social networks, creepypastas represent a new literary phenomenon at the crossroads of folklore, digital culture, and literature. These short and frightening stories, often accompanied by images or videos, have given rise to a new imaginary and poetics based on virality, interactivity, and media hybridity.
  This dissertation analyzes the genesis and evolution of the genre through a computational approach combining the study of texts and images. By using large-scale textual analysis and computer vision tools, it seeks to identify the forms, motifs, and codes specific to creepypastas, while examining the processes of canonization and literary legitimation in the digital age.

           PhD student: Alexandre Lionnet
           Supervisors : Daniel Stockholm (EPHE-PSL) & Florian Cafiero (ENC-PSL)

• The Medieval Transmission of Merovingian Historiography. A Computational-philological Study of Texts and their Manuscripts [ongoing]

• The Lost Manuscripts of Medieval Europe: Modelling the Transmission of Texts [ongoing]

  In the currents of cultural evolution, the fate of written artefacts hangs on the delicate balance between cultural preferences and chance. How do texts, like living organisms, experience a process of preservation, transformation or extinction? To answer this question, the ERC-funded LostMA project will blend AI, complexity science and philological expertise to unravel the mysteries behind the deviation of textual transmission from pure chance. Focusing on chivalric literature in a European context, the team utilises deep learning for large-scale data collection on 4 000 documents. This groundbreaking approach not only scrutinises the transmission of texts but also challenges the role of chance in shaping cultural canons.

• The role of cognition applied to unpredictable content in the evolution of writing (UNPREDICTABLE) [ongoing]

  The appearance of writing was a turning point in the evolution of human communication and cognition; yet writing did not evolve independently in most human societies. For millennia, many societies only used specialised graphic codes representing specific types of information—personal emblems, numbers, calendric units, among others. In the societies that invented writing independently, the notation of speech sounds is preceded by such specialised notations, and is first used for telegraphic writing (encoding only some parts of speech), giving rise to full sentences notation only centuries later. This project attempts to describe this phenomenon systematically, to provide a cognitively plausible explanation for it, and to explore the broader implications of the underlying psychological mechanisms. We hypothesise that contextually unpredictable content is more likely to be encoded: we are more likely to explicitly say or write what cannot be guessed. We will test three hypotheses: (H1) some types of content are consistently more predictable than others, in a way that is to a certain extent robust to historical and cross-linguistic variations; (H2) people can detect unpredictable content and they ascribe cognitive value to it; (H3) the evolution of writing and other graphic codes should follow a sequence, with the invention and use of codes for highly unpredictable content (e.g. numerical quantities) occurring first, followed by the encoding of predictable content (e.g. articles or prepositions). The project is based on three different methodologies: Computational linguistic methods on big text corpora will test H1; H2 will be tested by conducting experiments to test whether participants detect unpredictable content types as such, and whether mentioning unpredictable content enhances the perceived value of explanations and arguments; we will test H3 through a systematic investigation of the evolution of graphic codes dedicated to various types of content.

• Digital Approaches to the Survival and Loss of Old Norse Romances [ongoing]

  The genre of romance began to flourish in the North from the thirteenth century onwards, when the earliest known translations of European romances appeared under the patronage of King Hákon Hákonarson of Norway. During this period, Norse translations of continental works were produced in Norway, including works such as Tristan, Élie de Saint Gille, and Yvain, while the earliest legendary sagas were written down in Iceland, narrating the legendary past of Nordic and Germanic heroes like Sigurður the Dragon Slayer and Ragnar Loðbrók. These two related genres of Icelandic literature—riddarasögur and fornaldarsögur—are rarely studied together, despite their transmission histories being closely intertwined.
  This project examines the survival and loss of Old Norse romances, broadly understood as medieval chivalric and heroic literature, including works traditionally classified as chivalric sagas (riddarasögur), legendary sagas (fornaldarsögur), and related narrative poems (rímur). Its goal is to leverage recent developments in digital humanities to gather a quantifiable dataset describing key features of medieval texts and their manuscripts. This dataset will be used to apply statistical methods commonly used in ecology, such as unseen species modelling, to estimate the loss rates of works traditionally belonging to different genres of Old Norse literature. We aim to uncover which features of medieval texts and their manuscripts may have had the greatest influence on the diverse transmission paths of medieval chivalric and heroic literature from the North.

* A TEI model to encode medieval text and their metadata — Host lab: Centre Jean Mabillon; Supervisors: Virgile Reignier & Jean-Baptiste Camps
* Identifying conceptual metaphors with LLMs — Host lab: Centre Jean Mabillon; Supervisor: Rowan Hall Maudslay
* LatticIA : Les personnages mères dans les grands corpus littéraire — Host lab: Lattice; Supervisor: Olga Seminck
* Origine, circularité et déclin des motifs naturalistes — Host lab: Lattice; Supervisor: Dominique Legallois
* Letter shapes across the world — Host lab: Institut Jean Nicod; Supervisor: Olivier Morin
* Verifiability of fake news — Host lab: Institut Jean Nicod; Supervisor: Hugo Mercier