The Making of Data Liquidity: Embeddings in a navigable milieu and the hermeneutics of latent space

English translation of: Dumas Primbault, S. & P. Guhennec (2026) "La fabrique de la liquidité des données : Plongements dans un milieu navigable et herméneutique(s) de l’espace latent", Intelligibilité du Numérique 7 [online]. Abstract According to the discourses accompanying the fourth paradigm, one of the conditions for its realisation would be the fluidity of the data — “data liquidity” — on which it is based. The metaphor of the “data lake”, for example — used from 2011 onwards to describe a storage format for data of very diverse natures and formats, structured or unstructured, often without knowing what use they will be put to — has subsequently emancipated itself from its technical context to embody the hydraulic ideal of perpetual, frictionless recombination. In short, it is the fluidity of supposedly unstructured masses of data that would enable practices to be “un-siloed” and new knowledge to be produced. This would open up the new analytical possibilities of the fourth paradigm: the changing nature of results when processing large quantities of data, the equivalence of empirical materials of heterogeneous natures in a common space, the automation of these processes in their embodiment by self-learning machines. Yet there is nothing spontaneous or natural about the liquidity of data, which paradoxically needs to be structured to guarantee seamless, frictionless navigation. This contribution aims to shed light on the making of data liquidity by examining the operational chain of the word2vec embedding algorithm in digital humanities.