Do sound changes ever have exceptions, or does every word have its own history?

Do sounds change without exceptions? The classic answer is yes, and many linguists still hold it, though it has been challenged. In the late 19th century the Neogrammarians summarized their view as 'sound laws suffer no exceptions': a sound change affects all words in which its environment is met. That principle is the foundation of the comparative method of reconstruction and of the tree model of how languages descend from one another. When Grimm's law looked broken, Verner's law resolved the apparent exception.

But Hugo Schuchardt, a dialectologist who studied Romance languages, had doubts. Rarely used words lag behind, he wrote, and very frequent ones rush ahead. His fellow dialectologists found so many apparent exceptions that they adopted a motto: chaque mot a son histoire, 'every word has its own history'.

Then came a computer. In 1962 Peking University published transcriptions of 2,444 morphemes in 17 modern varieties of Chinese, and a Berkeley project led by William Wang tried to apply the comparative method to them. The data revealed pervasive irregularities. Wang argued that words change their pronunciation by perceptible increments, but a few at a time. Others replied that the irregularities came from dialect mixture, and one famous case was indeed borrowing from the local literary reading tradition.

The debate remains controversial, and many linguists still follow the Neogrammarian rule. Regularity is useful even when the world is messier than the law: expecting it lets historians of language define regular correspondences between related languages.