Research on sound change is usually conducted under the working assumption that it is regular, which means that it is expected to apply mechanically whenever its structural conditions are met, irrespective of any non-phonological factors like the meaning of the words that are affected.
Real-world sound laws often admit exceptions, but the expectation of their regularity or absence of exceptions is of great heuristic value[further explanation needed] by allowing historical linguists to define the notion of regular correspondence by the comparative method.
The only exception is that a sound change may recognise word boundaries, even when they are unindicated by prosodic clues.
Apparent exceptions are possible because of analogy and other regularization processes, another sound change, or an unrecognized conditioning factor.
For example, the Spanish fronting of the Vulgar Latin [g] (voiced velar stop) before [i e ɛ] seems to have reached every possible word.
By contrast, the voicing of word-initial Latin [k] to [g] occurred in colaphus > golpe and cattus > gato but not in canna > caña.
The symbol ">" can be reversed, B < A, which also means that the (more recent) B derives from the (older) A": The two sides of such a statement indicate only the start and the end of the change, but additional intermediate stages may have occurred.
In historical linguistics, a number of traditional terms designate types of phonetic change, either by nature or result.