How did a 21-year-old predict a lost sound that no one had ever heard?

Can a sound be predicted before anyone has heard it? In 1879 the Swiss linguist Ferdinand de Saussure, then 21, proposed that Proto-Indo-European, the reconstructed ancestor of English, Latin, Greek and Sanskrit, contained a series of consonants that left no direct consonantal descendants in languages outside the Anatolian branch. This is the laryngeal theory, and it was meant to explain irregularities in the patterns of the daughter languages.

The theory allowed a better reconstruction of PIE ablaut and roots. But there was a problem: at the time there was no direct evidence of the proposed sounds in any language. The views were accepted by a few scholars, but they did not achieve any general currency, as they were too abstract and had little direct evidence behind them.

Then the evidence turned up in the ground. Hittite, an extinct Indo-European language attested in cuneiform tablets from the archives of Hattusa, was deciphered. Its texts, recovered from 1906, contained a consonant written ḫ that did not match any consonant then reconstructed for PIE. In 1927 the Polish linguist Jerzy Kuryłowicz discovered that this sound appears in many of the places where the laryngeal theory predicted it should.

The theory is now widely accepted, because of the number of explanations it gives and the simplicity of the system it posits. Hittite texts confirmed Saussure's theory by featuring direct reflexes of PIE *h₂ and *h₃.