We have reached the point in our discussion of neutralization – or, more to the point, neutralization; neutralization as derived homophony – to test our predictions against a body of data. Recall that neutralization involves limitations on the distribution of phonetic information, limitations that possess genuinely function-negative consequences. Assuming that language structure and language change are influenced by functional considerations, there may exist passive diachronic pressures against the proliferation of homophone-deriving alternations. Thus, recall from Chapter 1 my overarching proposal: alternations are more likely to be present in a language if they do not significantly increase the level of homophony; derived homophony is not excessive.

A particularly rigorous testing ground for these predictions would involve a pattern or patterns of sound change that do significantly increase the amount of (traditionally defined) neutralization. The proposal would be that, in such cases, despite an overall decrease in word-internal phonetic distinctions, evidence for semantic distinctions emerges largely intact, as derived homophony remains negligible. In this chapter then, we report on a single case study: neutralization in Korean (Silverman 2010, Kaplan 2011a). Indeed, it emerges that despite this language's huge amount of traditionally defined neutralizing alternation, the amount of derived homophony resulting from such alternations – that is, the amount of neutralization – is surprisingly meager.