From the seminal works of Moretti onward, the evolutionary approach has been central to the field of distant reading, offering a large-scale, data-driven framework for understanding the historical development of literary forms. However, equating literary evolution with natural selection poses major methodological challenges, particularly in modeling the ‘environment’ in which literary forms circulate and compete. This environment comprises multiple overlapping dimensions —social, political, economic, and more— whose effects are difficult to isolate and assess empirically. In my work, I focus on one of these dimensions, drawing from Bourdieu’s theory to model literary reception as staged within a social field. I then turn to Digital Social Reading platforms —specifically Goodreads— which, by combining the functionalities of a social network (making friends, following users, joining groups) and a book cataloging system (reviewing, rating, creating lists), provides an empirical basis for studying the social landscape of literary reception.

Through network analysis, I identify distinct reader communities characterized by coherent taste patterns that emerge and stabilize over time. By contrasting trends in readers’ choices within a specific community against a null model of cultural drift, I test whether genre preferences evolve due to value-driven selection or stochastic imitation dynamics. The findings suggest that while cultural drift plays a role, selective pressures within reader communities actively shape literary preferences, particularly in the rise and decline of specific genres. Ultimately, this study models one of the key ways in which the social environment drives literary change —by forming niches that exert a detectable selective pressure over the literary landscape, favoring certain forms, or genres, over others.

Stylometric authorship attribution is the task of assigning texts of unknown, pseudonymous or disputed authorship to their most likely author, often based on a comparison of the frequency of a selected set of features that represent texts. The way stylometric methods typically approach authorship attribution is to use the frequencies of a large number of simple features, such as words or character sequences, to determine of the degree of similarity between texts. These similarities, in turn, are interpreted as an indicator of the likelihood for two texts to have been written by the same author: the more similar the feature vectors, the more likely is identical authorship.

The parameters of the analysis, such as feature selection and the choice of similarity measure or classification algorithm, have received significant attention in the past. Two additional key factors for the performance and reliability of stylometric methods, however, are corpus composition and corpus language. They are relevant not only for the results in a specific case, but also for the overall performance and reliability of stylometric methods of authorship attribution. Therefore, the aim of ongoing research by our group is to disentangle the influence of corpus composition and language on the performance of stylometric authorship attribution: To what extent do the attribution accuracy and robustness of such approaches depend on the language of the materials, on the one hand, and on corpus composition, on the other? How do these two factors interact with each other, and how do they interact with feature selection?

This paper reports on results relevant for one part of this issue, that of language, by investigating four distinct but broadly comparable corpora in a classification scenario. In order to investigate the role of language independently of corpus composition, all four corpora were automatically translated into the other three languages using the DeepL machine translation system.

We can show that corpora of different language and composition lead to different attribution accuracy levels and different best-performing features, an expected result. We can also show that translated corpora (at least when all texts have been translated by the same machine translation system) usually lead to a lower attribution accuracy, overall, compared to their counterpart in the original language.

We will also report on preliminary results concerning the second part of the question, namely the influence of corpus composition on attribution accuracy, also in a multilingual setting. Here, we aim to show that using corpora composed of texts with high within-author similarity of texts and low between-author similarity of texts, in terms of basic metadata (such as author gender, subgenre, narrative perspective, or time of composition) generally leads to higher attribution accuracy than when the inverse is true (low within-author similarity and high between-author similarity).

Un nouvel outil numérique de visualisation est proposé pour l’analyse des textes littéraires dans une perspective comparative, à partir d’une approche novatrice. Il offre une lecture à distance particulière dans la mesure où il ne s’applique ni à une grande masse de données ni à un large corpus de textes à la fois, mais à un seul texte, dont il ne retient que la dimension visuelle, indépendamment de sa mise en page. Cette forme visuelle du texte est façonnée par les paragraphes et les chapitres, qui rythment le texte en fonction de leur longueur respective.

Un logiciel, Narra 2.0, a été développé afin de mesurer ces longueurs textuelles successives et générer un tableau de mesures, donc une suite numérique à partir de laquelle sont produites des données statistiques et, grâce à des algorithmes, des visualisations. Ces dernières montrent ainsi le rythme du texte en fonction de la longueur de ses paragraphes ou de ses chapitres, soit la fréquence des changements – et de locuteurs et de thèmes – dans le texte, une dynamique propre à l’écrit.

Cette méthodologie offre la possibilité de comparer les textes dans le temps (au fil des éditions), dans l’espace (de diverses régions géographiques) et pour un même auteur ou courant littéraire. Elle permet également d’appliquer la méthode éprouvée des atlas – stellaires du XIXe et XXe siècles –, aux recherches comparatives. À titre d’exemple, Un Atlas des spectres de textes littéraires, a confirmé l’existence d’une corrélation entre la longueur des paragraphes et le genre littéraire ou la période d’écriture.