Abstract
In recent years, the application of artificial intelligence (AI) to the study of non-human communication has gained significant momentum. This article critically explores the intersection between AI and zoosemiotics by analysing a corpus of three emblematic case studies – Project CETI, ISPA, and current bioacoustic classification systems – which apply generative models and machine learning techniques to animal communication. Through close textual and conceptual analysis, the article shows how these approaches, despite their technical sophistication, tend to reduce animal semiosis to its external form, overlooking its embodied, ecological, and relational dimensions. The study exposes the epistemological, methodological, and semiotic limits of such models, demonstrating how the situated nature of animal signification is often flattened into abstract algorithmic encoding, devoid of pragmatic context. In response to these issues, the article introduces the concept of Generative Zoosemiotics (GZ), a new theoretical framework that combines tools from zoosemiotics and cultural semiotics with a critical inquiry into AI technologies. GZ proposes to analyse how machines interpret, simulate, or reconfigure animal signs, questioning the epistemic and ethical implications of these practices and asserting the irreducibility of interspecific communication to the logic of symbolic translation.