Beyond Computational Formalism or, Architecture Matters

Despite frequently avowed commitments to formalist methodologies, computational literary studies (CLS) has insufficiently accounted for the importance of the formal architectures of the computational models it employs—particularly deep learning neural networks. Arguing against the tendency to treat neural networks with an abstract gloss of their operation or to focus attention on the outputs, this article posits that architecture is not merely a technical detail but a crucial site where meaning is made and historicity registered. By examining the genealogy of neural network architectures—from Frank Rosenblatt’s Perceptron to contemporary transformer-based models—this article demonstrates how these architectures materially shape the capacities, outputs, and interpretive possibilities of machine learning models.