Edward Sapir (18841939). Language: An Introduction to the Study of Speech. 1921.
without going back on the drift the only possible method is to have the other sounds of the series shift in analogous fashion. If, for some reason or other, p becomes shifted to its voiced correspondent b, the old series p, t, k appears in the unsymmetrical form b, t, k. Such a series is, in phonetic effect, not the equivalent of the old series, however it may answer to it in etymology. The general phonetic pattern is impaired to that extent. But if t and k are also shifted to their voiced correspondents d and g, the old series is reëstablished in a new form: b, d, g. The pattern as such is preserved, or restored. Provided that the new series b, d, g does not become confused with an old series b, d, g of distinct historical antecedents. If there is no such older series, the creation of a b, d, g series causes no difficulties. If there is, the old patterning of sounds can be kept intact only by shifting the old b, d, g sounds in some way. They may become aspirated to bh, dh, gh or spirantized or nasalized or they may develop any other peculiarity that keeps them intact as a series and serves to differentiate them from other series. And this sort of shifting about without loss of pattern, or with a minimum loss of it, is probably the most important tendency in the history of speech sounds. Phonetic leveling and splitting counteract it to some extent but, on the whole, it remains the central unconscious regulator of the course and speed of sound changes.
The desire to hold on to a pattern, the tendency to correct a disturbance by an elaborate chain of supplementary changes, often spread over centuries or even millenniathese psychic undercurrents of language are exceedingly difficult to understand in terms of individual psychology, though there can be no denial of their historical reality. What is the primary cause of the unsettling