↓
Developed within the Machine Visual Culture research group led by Leonardo Impett at the Bibliotheca Hertziana in Rome (Max Planck Institute), the Corpus Explorer is an interface for exploring archives, assisted by several artificial intelligence models. Conceived as a digital desk for researchers, curators and archival collections, it offers a new way of visualising large corpora within a critical apparatus of vision and analysis, at once human and algorithmic.
Distant viewing: opening a field of vision limited in space and time
Human vision is limited in space and time. So is its capacity for memory.
The problem is not new. Scientists, historians and anthropologists once spent decades gathering the material they studied, sometimes unable to observe their whole corpus on a single surface, whether the documents themselves or the systems that arrange them. The distant reading proposed by Franco Moretti, which replaced the close reading of one text with the observation of thousands , was followed by distant viewing, theorised by Taylor Arnold and Lauren Tilton, which applies computational methods to the observation of digital archives at very large scale . Gathering archives within malleable digital interfaces has allowed the eye to reach, in a few clicks, a perceptual scale of an order of magnitude never reached before.
These new methods of observation, brought by the digital humanities to the humanities, have introduced new practices for curating and showcasing archives. The digital inspection table is what we like to call "the montage table", after Georges Didi-Huberman, who conceives it, in the tradition of the iconographic practice of atlases and of Eisensteinian montage, as a mental and material space for arranging images which, through their gaps and connections, generate critical thought . The digital interface thus becomes a hybrid space of montage, curation and showcasing of archives, at once human and computational, in the line of the cultural analytics initiated by Lev Manovich.
Artificial intelligence systems, machines for perceiving
In interfaces for exploring and analysing archives, AI systems not only give access to monumental and infinitesimal scales through the automatic processing of contextual, visual and sound data (metadata, recognition of shapes, of sounds, etc.), but also refine the filters through which we perceive this material.
In this context, the parameters of perception depend, on the one hand, on the documents under examination. Processing film material, for instance, involves parameters specific to the medium, such as frame composition, shot scale, camera movement, the various film forms and sound composition, whereas literary documents call for attention to the materiality of the paper and to the layout of text and images on the page, up to the recognition and transcription of manuscripts. Each type of document has its own grammar, defining its methods of analysis and indexing and its modes of interaction.
Operationalising an analytical gaze
On the other hand, they depend on the research hypothesis and on the analytical gaze of the researcher, the curator or the artist. The Corpus Explorer is designed to operationalise an analytical gaze on the material at hand.
For example, the corpus studied in Marta Pizzagalli’s doctoral thesis (Università della Svizzera italiana, Lugano), entitled Intermedial combinations in Guido Gozzano’s work. A study of poems and prose in their periodical publication, 1905-1916 , consists of fifteen Italian illustrated magazines published between 1900 and 1916. So far, six have been analysed with the Corpus Explorer: La Lettura, Corriere dei piccoli, La Donna, Bianco, rosso e verde, La Grande illustrazione and Lettura sportiva, totalling approximately 24,000 pages. By measuring the layout of these magazines at scale, namely the number of images per page, their position and the proportion of the page they occupy, can we discern an editorial approach specific to each magazine and trace its evolution over time?
In other words, it doubles the human gaze with that of a machine, externalising it so that it processes the material (analyses it, indexes it and puts it on view), but also, and above all, so that the model's own perception is put on view.
A double subjectivation
This is a double subjectivation: human vision is parameterised and sets the perceptual parameters of the machine. Yet the machine itself perceives in ways that can escape researchers; it is a situated algorithmic subjectivity whose assumptions must be questioned, not a neutral gaze. So whether AI is used to examine cultural material, or that material is used to reveal algorithmic biases and sensitivities, what sets this interface apart is its use of several AI models, in order to observe and measure their different effects on the processing of a single corpus. It also brings greater transparency to the computer vision tools that usually process this data in complete opacity.
The Corpus Explorer is thus an interface for visualising archives, a tool for observation and analysis assisted by algorithmic models of perception whose capacities differ from human vision; to some extent, it is also an interface that supports AI interpretability. This interpretability takes several forms: visualising the areas of the image on which a model bases its recognition and its transcription into natural language, projecting the documents into the model's representation space (UMAP and PCA), where their proximities trace its way of classifying the world, or comparing results from one model to another. More recent methods, such as sparse autoencoders, make it possible to isolate the concepts a model has learned to distinguish.
Latent spaces of archives
Today, algorithms do more than generate fake archives from those they were trained on. Models trained on vast sets of images and texts become archives of a new kind themselves. Antonio Somaini proposes to think of their latent spaces as "meta-archives", in that they are not places where archives are kept, still less consulted, but structures that compress and transform cultural memory . To question how these models perceive is to question the form our access to the past now takes. These reflections are also at the heart of the documentary Reinscribing the Dondorf, directed by Ania Szczepańska in 2026.
It is therefore important to question these algorithms and to treat them as situated gazes and perceptual prisms whose characteristics, mechanisms and provenance we need to know better, in order to question their biases, absences and erasures.
The Corpus Explorer seeks less to demonstrate the possibilities of artificial intelligence than to put them to the test of archives, and vice versa. By making visible the perceptual filters through which documents are now indexed, processed and consulted, it offers greater methodological transparency, opens the digital humanities to the wider humanities, and presents archives within a hybrid medium of perception, more transparent and more critically examined.
Get in touch with any question or for a possible collaboration!