What else belongs on this shelf

I · The quarrel about reading

Augustine, De doctrina christiana (ca. 397–426). The source of the oldest rule Eco still uses: an interpretation of one part of a text stands or falls by the rest of the same text. The criterion of internal coherence, sixteen centuries before anyone called it that.

W. K. Wimsatt & Monroe Beardsley, "The Intentional Fallacy" (1946). The classic argument that the author's intention is neither available nor desirable as the standard of judgment. Eco's intentio operis presupposes this battle was won — and then asks what limits readers, if not authors.

Hans-Georg Gadamer, Truth and Method (1960). The hermeneutic tradition's monument: understanding as the fusion of horizons, prejudice as the condition (not the enemy) of reading. Eco's "social treasury" and his unashamed hermeneutic circle are in constant, mostly silent, conversation with it.

Susan Sontag, "Against Interpretation" (1964). The elegant counter-position this site needs: that interpretation itself — the digging for deep content — can be "the revenge of the intellect upon art." Where Eco wants better interpretation, Sontag wants less of it.

E. D. Hirsch, Validity in Interpretation (1967). The strongest modern defense of authorial meaning as the only stable norm. The position Eco declines on his other flank — useful for seeing exactly how narrow the path he walks is.

Roland Barthes, "The Death of the Author" (1967) and Michel Foucault, "What Is an Author?" (1969). The two short texts that made the author's removal a slogan and then a question. Barthes's birth-of-the-reader is the position Eco caricatures as textual Gnosticism; Foucault's "author function" is closer to Eco's model author than either party advertised.

Stefan Collini (ed.), Interpretation and Overinterpretation (Cambridge UP, 1992). The book of these very lectures, with the responses: Richard Rorty's pragmatist rejection of the use/interpretation distinction ("The Pragmatist's Progress"), Jonathan Culler's "In Defence of Overinterpretation," and Christine Brooke-Rose. The obvious next text for this site to read — Eco's criteria, under fire from people in the room.

Umberto Eco, The Open Work (1962), The Role of the Reader (1979), The Limits of Interpretation (1990), Six Walks in the Fictional Woods (1994). The before and after: the book whose "open" side readers overstressed, the book that built the model reader, and the two books where the limits got their full treatment.

II · The machine readers

Alan Turing, "Computing Machinery and Intelligence" (1950). Where the question "can a machine mean anything by what it says?" got its modern form — and where deflecting it in favor of observable behavior was first proposed as a method.

Franco Moretti, Distant Reading (2013). Machine reading before language models: literature studied in corpora and counts rather than passages. A useful reminder that "the machine as reader" did not begin in 2020 — and that it began as the deliberate opposite of close reading.

Emily M. Bender & Alexander Koller, "Climbing towards NLU: On Meaning, Form, and Understanding in the Age of Data" (2020). The octopus thought experiment: a system trained only on the form of messages, intercepting an underwater cable, never grounded in what the messages are about. It is Eco's letter-in-the-bottle rebuilt as a learning-theory argument — and it reaches the opposite conclusion about what the foundling reader can know.

Emily M. Bender, Timnit Gebru, Angelina McMillan-Major, Margaret Mitchell, "On the Dangers of Stochastic Parrots" (2021). The most-cited statement of the skeptical position: language models as "stochastic parrots," producers of form whose meaning is supplied entirely by the human reader. In Eco's terms: the claim that with a model there is no intentio operis at all, only intentio lectoris projected onto noise.

Murray Shanahan, "Talking About Large Language Models" (2022). The careful middle position: a philosopher's audit of which intentional words (knows, believes, means) survive translation to LLMs and at what discount. The closest thing the machine-reading debate has to Eco's tone.

III · Interpreting the machines

Somewhere around 2020 the word "interpretation" doubled. The machine is no longer only a candidate reader of our texts; it is the largest unread text we have produced — billions of parameters about which the question "what does this mean?" is asked in earnest. The field that asks it even took the name interpretability. The methodological echoes are uncanny: this literature, too, has its signatures, its suspicious readings, and its own running argument about which discovered patterns are real and which are the reader's.

Chris Olah et al., "Zoom In: An Introduction to Circuits" (Distill, 2020). The manifesto of the close-reading school of neural networks: the claim that individual features and circuits inside a network are legible, the way an anatomist's tissues are.

Anthropic, "Towards Monosemanticity" (2023) and "Scaling Monosemanticity" (2024). Dictionary learning applied to a real language model: millions of human-interpretable features extracted from activations — including, famously, one for the Golden Gate Bridge whose amplification changed the model's self-description. The strongest evidence to date that there is something in there to interpret.

Anthropic, "On the Biology of a Large Language Model" (2025). Circuit-tracing case studies on a production model — including cases where the model's stated reasoning and its traced computation come apart. Directly load-bearing for this site's claim that a model's self-exegesis is testimony in Eco's weak sense, not authority.