The space of love: problematizing the interaction of close and distant reading in the study of courtly love

Abstract

This thesis asks whether computational methods can intervene productively in an established humanistic debate while remaining answerable to the interpretive traditions they seek to extend. It approaches this question as a methodological problem, using the longstanding debate over courtly love in later medieval English literature as a test case. Reconstructing the Anglo-American debate from Gaston Paris through the decisive influence of C. S. Lewis, the thesis argues that its persistent impasse can be understood partly as a problem of archival restriction. The major interpretive traditions identified here as philosophico-theological, sociological, and psychological each recover genuine features of medieval representations of love, but do so from overlapping yet differently delimited bodies of evidence. Their internal coherence therefore derives partly from the evidentiary fields through which courtly love has been constituted as an object of inquiry. Drawing on Rüdiger Schnell, the thesis consequently reframes courtly love not as a unified doctrine or historical system but as a discourse: an ongoing and contested negotiation of love across texts, genres, authors, and linguistic traditions. To make that discourse computationally analyzable, the thesis brings together two traditions of discourse analysis. A post-Foucauldian account provides the concept of the archive as a historically conditioned field of statements, while Zellig Harris’s distributional hypothesis provides an empirical account of how patterned linguistic relations can be recovered from textual data. Spitzmüller and Warnke’s DIMEAN supplies the multilevel framework through which these approaches can be synthesized, specifying how linguistic regularities at the level of words and texts may be interpreted as evidence of larger discursive formations. Vector-space modelling provides the computational means of representing those distributional relations. The thesis also addresses the epistemological conditions under which such an intervention can count as humanistic interpretation. It argues that the opposition between close and distant reading rests on reciprocal misreadings in which each approach privileges a different object of authority: textual complexity and minutiae for close reading, and scale, archive, and evidentiary breadth for distant reading. Through a critique of the falsificatory logic exemplified by Matthew Jockers, this study argues that computational findings do not supersede or invalidate established interpretations, because humanistic interpretations need not succeed one another through falsification. Computational and non-computational approaches instead produce different, partial models of their objects. The consequential distinction is therefore not simply one of scale but of modelling: how evidence is bounded, represented, and made to stand for a larger field. This framework is tested on a deliberately bounded corpus of Middle English texts. At the word level, distributional analysis shows that the theological, sociological, and psychological dimensions isolated by scholarship coexist within the semantic field of love. At the document level, the analysis uses C. S. Lewis as a benchmark because subsequent courtly-love scholarship has repeatedly developed from, revised, or defined itself against his synthesis. Testing the fourfold taxonomy of The Four Loves reveals that its categories do not map evenly onto the corpus and motivates a revised three-part description of literary discourse: courtly-romantic, secular-erotic, and theologico-philosophical. The purpose is not to falsify Lewis but to show how computational modelling can sharpen distinctions within an inherited interpretive framework. Producing this analysis also required confronting the practical limits of computational work on pre-modern textual data. Corpus construction, periodization, textual selection, orthographic variation, and cumulative processing noise make medieval evidence a problem of heterogeneous data integration rather than simply one of scale. The thesis develops an AI-assisted normalization procedure that maps dialectal and scribal variants onto shared Present-Day English lexical descendants while recording cases requiring semantic substitution for audit. These methodological solutions, together with the corpus-construction and modelling framework developed here, demonstrate how computational methods designed for large datasets can be adapted to small, historically mediated evidence. The thesis therefore offers a proof of concept for how computational modelling can complicate, refine, and extend established humanistic interpretation without claiming to replace it.