The AI Journal

A continuo player deciding how much ornament to add to a figured bass line, and a classicist deciding which of two fifteenth-century manuscripts preserves the true reading of a disputed line in Catullus, are not obviously doing the same kind of work. One restores sound, the other restores words. Both, however, are choosing among reconstructions that their own discipline’s rigour cannot rule out, and both describe the choice as recovery rather than decision. Neither field’s evidence actually supports that description, and it fails to support it for the identical structural reason in both places.

Historically informed performance announced itself, from the 1960s onward, as a correction: Baroque and early Classical music had been played for two centuries on instruments, at pitches, and with expressive habits its composers never intended, and reconstructing period practice would let audiences hear the music as it actually sounded. Richard Taruskin’s Text and Act (1995) remains the most influential rebuttal of that claim. Taruskin does not argue that the movement is fraudulent, but that it is not doing what it says: its aesthetic of lean, literal, text-following performance is a twentieth-century style, continuous with the geometric manner of the interwar neoclassicists, dressed in historical costume. The documentary record it draws on, he stresses, is fragmentary and frequently ambiguous about the specifics that matter most, tempo relationships, the degree of rhythmic flexibility, how closely a given treatise’s advice was actually followed, so that the historical choice among several defensible performance styles is rarely forced by the evidence. It is a judgement, dressed as a finding.

John Butt’s Playing with History (2002) takes the more sympathetic view that the movement’s intellectual seriousness has been underrated, but he does not deny the evidentiary gap; he relocates the debate to what historical performance is actually true to, the work, the composer’s intention, or something else again, since the three need not coincide and the record rarely settles which should win when they conflict. Peter Kivy’s Authenticities (1995) makes the multiplicity explicit by naming distinct and sometimes incompatible targets a performance can aim at: the composer’s intended sound, the literal sonic materials of the period, the performance conventions of the period, and the performer’s own considered interpretation. A reading faithful to one can violate another; an eighteenth-century violinist’s actual performance, on Kivy’s argument, mixed personal inflection into practice that most self-styled authentic recordings suppress. There is consequently no single fact, how it actually sounded, for a performance to be right or wrong about. There are several, each compatible with the same scraps of treatise and iconography, and a performer must choose among them.

Classical textual criticism built its method on the opposite hope: that manuscript relationships, properly mapped, would converge on a single original. Paul Maas’s Textkritik, published in 1927 and translated into English in 1958, systematised the nineteenth-century method associated with Karl Lachmann: trace the shared errors across surviving copies, the errores significativi, to reconstruct a family tree of manuscripts, a stemma, and let that tree identify which readings the lost archetype actually contained. The method works cleanly when copying is linear, each scribe working from a single exemplar. Maas himself identified the case in which it does not: contamination, where a scribe consulted more than one exemplar and mixed their readings, for which he concluded there is no remedy within the method itself. Later critics, Giorgio Pasquali prominent among them, went further, arguing in 1934 that many traditions are not closed but ‘open’, shaped by the horizontal contact between copies that a single stemma cannot represent, which undercuts the assumption that there is one recoverable archetype to find. Where contamination runs heavy, as it does for many widely copied classical texts, the stemmatic method does not fail to find the original; it fails to certify that any procedure can find it, and an editor settles on a reading by judgement the method cannot underwrite.

Jerome McGann’s A Critique of Modern Textual Criticism (1983) pressed the same gap from the other direction, arguing that even an uncontaminated tradition would not yield a single authoritative text, because the editorial ideal of an author’s final intentions misdescribes how texts are actually produced, through negotiation with printers, censors and collaborators, in versions that are each, in their own historical moment, the text. D. C. Greetham’s survey of the field (1992) records the practical consequence: editors choosing between best-text editing, which prints one witness as copy-text and corrects its evident errors sparingly, and eclectic editing, which composes a text from readings scattered across several witnesses, are not applying a decision procedure so much as expressing a judgement about which kind of error is more tolerable. Two editors working from the same manuscripts, under the same stated method, can and regularly do produce different texts, each defensible by the discipline’s own lights.

The parallel is not an analogy; it is the same structure wearing two notations. Both fields posit a determinate original, a sound-world in one case, an authorial text in the other, that their surviving evidence is supposed to recover. In both, what survives, treatises and iconography on one side, manuscript variants and scribal patterns on the other, is compatible with more than one reconstruction that meets every test the discipline can apply. In both, the gap is closed not by further evidence, which would only relocate the same underdetermination to a finer grain, but by a judgement the practitioner then reports in the language of discovery: this is how it was done, this is what the author wrote. Early music did not borrow its authenticity problem from philology, and philology did not solve the problem before early music arrived at it. They are the same unresolved argument, conducted in two departments that have mostly managed not to notice the other was already having it.

The strongest objection to treating these as one problem is that the evidence differs in a way that matters. A manuscript variant is a fact: a scribe, at some actual moment, wrote a specific different word, and that word survives on the page for an editor to weigh against its rivals. A disputed point of eighteenth-century bowing practice is not a fact of that kind; it is an inference from a treatise’s general advice, itself possibly aspirational or locally specific, to a sonic practice that left no direct trace of its own. Textual critics are choosing among witnesses to something that demonstrably happened. Performers are often inferring from indirect testimony toward something for which no comparably direct witness exists, or ever did. Collapsing the two into a single structural problem, on this view, flatters performance practice with a rigour, the discipline of weighing competing direct evidence, that it has not earned.

The objection is right about the evidence and wrong about what follows from it. It shows that textual criticism’s underdetermination is narrower, bounded by a finite set of actual witnesses, while performance practice’s is broader, bounded only by what a historical imagination finds plausible. That is a difference in how much latitude the judgement has, not in whether a judgement is being made and then redescribed as a finding. A best-text editor who prints a plainly corrupt reading because the alternative requires conjecture she cannot defend, and a fortepianist who declines a written-out ornament because the treatise recommending it addressed a different, more virtuosic context, are making the identical move: declining to let evidence settle more than it can bear, and owning the remainder as a choice rather than a discovery. Narrowed to its structural core, both fields reach a point past which the record cannot adjudicate, and both have, for the better part of a century, preferred to call what comes next recovery rather than name it.