Navigation – Plan du site

AccueilAppels à contributionAppels en coursSignata 17 / Multimodal Translati...

Signata 17 / Multimodal Translation in Generative AI: Fidelity and Creativity

Maria Giulia Dondero et Massimo Leone

Issue 17 of Signata aims to offer a comprehensive state-of-the-art survey on generative artificial intelligences (GAIs) and to reinvigorate interdisciplinary research on this topic. Building on the work initiated by the 2024–2025 edition of the Paris International Seminar of Semiotics, Generative Artificial Intelligence and New Semiotic Challenges: Translation and Creative Appropriations, which brought together scholars from semiotics, linguistics, anthropology, aesthetics, philosophy of art, and media theory to explore the generative dimensions of language, it serves simultaneously as an up-to-date overview and a critical reflection on the most recent developments, encompassing both the relentless pace of technological innovation and the semiotic approaches devised to analyze these phenomena.

This issue centers on questions that have long animated semiotics and the language sciences. Foremost among these is translation, which now occurs between human-generated prompts and machine-refined responses, between natural language and images, between visual forms and textual descriptions, between prompts and audiovisual outputs. Intersemiotic translation has traditionally provided a space for comparing and contrasting media-specific features (such as spatial versus linear arrangements, material substrates, or the organization of written traces). Today, however, intersemiotic translation unfolds within a technical framework of digital encoding, which tends to homogenize diverse materials. Yet this process still produces texts that we recognize as intelligent, functional, or creative—texts whose compositional logic demands renewed investigation.

At its core, translation—especially in its intersemiotic form—is a matter of enunciation. In the case of generative AI, it mobilizes multiple enunciative instances and agencies, each operating within distinct planes of mediation. GAIs already constrain us to consider a layered multiplicity of such instances: the training databases used to train the models, the algorithms governing training and inference, the numerical representations known as embeddings, which convert visual or verbal texts into sequences of values, the code that structures these processes within specific programming languages; and, finally, the prompts, which act as user-facing gateways into the system. Within the outputs generated by GAIs, these enunciative instances register with varying degrees of presence and intensity. They are combined in diverse ways, involving both human and non-human actions, enabling us to better define the nature and degree of creativity expressed in each generated output.

But how do we measure, evaluate, and assign value to creativity? According to which criteria, and within which frames of reference? Should we refer to the visual arts, literature, logical argumentation, or common sense? This issue seeks to differentiate the many forms of creativity according to the domains in which they emerge—visual arts, graphic design, advertising and marketing, as well as politics, strategy games, legal reasoning, and decision-making. In each of these fields, creativity is judged by its own specific parameters, which are also shaped by the discursive genres within which these language models operate. Creativity manifests and is negotiated quite differently in the hard sciences versus the humanities. Ultimately, generative artificial intelligence permeates all spheres of individual and collective life, operating in varied ways across different contexts. In each case, enunciative instances differ in weight and intensity, both in how they recombine prior texts and in how they affect users. More broadly, these instances influence the ongoing production and reconfiguration of this emerging discursive economy.

Our object of study is thus not limited to the semiotic analysis of texts co-produced with ChatGPT, DALL•E, Midjourney, Stable Diffusion, DeepSeek, Perplexity, and others. Rather, it encompasses the analysis of any socio-cultural practice involving generative AI—from education to political debate, from scientific research to artistic production, including the computational practices that implement and continually transform these tools. Within this framework, we investigate what may be called a meta-perspective generated along the operational chain of generative artificial intelligences, a kind of “reduced self-monitoring” attributed to the machine that enters into tension with our own “limited supervision.”

This issue of Signata explores translation between machine and human languages, creativity, impacts on different social domains, and the self-reflexivity of machines through theoretical, methodological, and genealogical perspectives, in order to grasp the evolution, diversification, and specialization of these generative intelligences.

Rechercher dans OpenEdition Search

Vous allez être redirigé vers OpenEdition Search