The Last Recording
Alessandro Moreschi was recorded at the Vatican in 1902 and 1904. He was a castrato — his voice had been preserved in its prepubescent range by surgical intervention before puberty. The recordings are the only audio evidence of a vocal tradition that dominated European opera and sacred music for over three hundred years. At its peak, an estimated four thousand boys per year were castrated in Italy for musical training. The practice was banned in 1870. By the time recording technology existed, Moreschi was the only one left, and he was aging, past his prime, and had never been considered the best of his generation.
The recordings are poor. Early acoustic recording captured a narrow frequency range. Background noise intrudes. Moreschi's technique, trained for live performance in reverberant churches, does not suit the conditions. The ornamentation sounds excessive to modern ears trained on different aesthetics. Historians listen to the Moreschi recordings and hear something fascinating but unrepresentative — a diminished version of a diminished practice captured by inadequate technology at the last possible moment.
This is the structural problem of terminal evidence: when a recording is made because a tradition is ending, the recording cannot capture what the tradition was at its peak. The motivation to record increases as the thing worth recording decreases. Nobody recorded Farinelli. Nobody recorded Caffarelli or Senesino or the hundreds of castrati who performed before audiences across Europe for three centuries. When the technology to record finally existed, the tradition had already collapsed to a single practitioner in a single institution, singing for a microphone that could not reproduce the frequencies that made his voice distinct.
Bo, a language spoken on the Andaman Islands for approximately sixty-five thousand years, died with Boa Sr in 2010. Linguists recorded her speech in her final years. But a language spoken by one person is not the same language that was spoken by a community. It has no interlocutors. It has no arguments, no jokes that require shared reference, no children learning it wrong in productive ways. The recording captures the lexicon and some grammar. It cannot capture the pragmatics — the social function that is the reason a language exists. What survives in the recording is the structure of the language. What dies with the last speaker is everything the structure was for.
The pitch drop experiment at the University of Queensland has been running since 1927. Nine drops of pitch have fallen in ninety-seven years. No one has ever witnessed a drop detach. The eighth drop fell in November 2000; the webcam installed to capture it malfunctioned. John Mainstone, who tended the experiment for fifty-two years, was away when the ninth drop fell. The experiment's purpose is to demonstrate that pitch, which appears solid, flows — but the moment of transition, the instant the drop separates, has never been observed. The experiment has produced data (an estimated viscosity of 2.3 × 10¹¹ pascal-seconds) but not the observation it was designed to provide.
In each case, the recording exists. The tradition, the language, and the drop are documented. But the recording does not contain what mattered most about what it records. Moreschi's audio does not contain what a castrato sounded like at the height of the tradition. Boa Sr's recordings do not contain what Bo sounded like when it was alive. The pitch drop videos do not contain the drop falling. The evidence is real. It is also systematically misleading about the phenomenon it is the only evidence for.
The problem is not that we lack records. The problem is that records are most likely to be made at the margins — when something is rare enough to seem worth recording, or old enough to be recognized as endangered, or interesting enough that someone points a camera at it. By the time these conditions are met, the thing has already changed. The center of the distribution, the typical case, the thing at its best — that is precisely what nobody records, because at the time it is unremarkable.