Another class of alleged unmergers are those which Labov ascribes to near-merger, in which two sounds get so phonetically close to each other that observers claim they are merged and even their speakers have trouble perceiving the distinction. Nonetheless the two classes are consistently distinguished acoustically (and therefore also in articulation), and children, whose phonetic perceptual abilities are perhaps better than adults, are able to acquire the distinction. Later phonetic shifts may modify these phonetic targets so that they no longer are so close to each other, and observers then may claim that the sounds have mysteriously unmerged. However, they never had actually merged in the first place.
It may help dispel the common disbelief among linguists in the existence of near-mergers to describe an example. The best studied near-merger is that of ferry vs. furry in Philadelphian White English. For a class project some years ago, for example, I conducted a commutation test with a friend who was born and raised in Philadelphia. We tape-recorded him reading a randomized list of 13 tokens of the words ferry and furry. I backed the tape up to the 4th token, replayed the just-read list at a comfortable volume, and had my friend categorize his own pronunciations as ``a boat''(ferry) or ``an animal''(furry). His performance was 5 correct and 5 incorrect, which is exactly the expected score on a forced-choice, two-answer test if the answers are generated randomly, by coin-toss. That is, he couldn't tell which was which, in listening to his own speech. Later acoustic and impressionistic analysis of the tape showed that the vowels were in fact perfectly categorizable. A single line could be drawn on an F1-F2 chart to divide all tokens of one category from all tokens of the other. And I could teach myself to hear which was which, so that I could pass the test with a 100% correct score.
Thus the near-merger situation is this: speakers properly associate with two particular lexical sets a measurable phonetic difference which is audible to a trained phonetician. Despite this apparent phonemic difference, the same speakers are unable to correctly categorize the minimal pairs that they themselves produce.
The cognitive explanation for near-mergers is unknown at this time. However, I here propose an empirically testable explanation, based on the following two facts.
First, children are excellent impressionistic phoneticians; they acquire the most detailed features of phonetic production, which they must be able to hear, since they can imitate them (cf. Read, 1975, referred to in Kiparsky 1988). As maturation progresses, however, this perceptual ability apparently degrades. In fact, what it means to acquire a phonology is to a great extent simply learning not to be sensitive to non-distinctive differences; an overall loss of phonetic sensitivity is consistent with (though not logically implied by) this general effect.
Second, adults largely retain the phonetic patterns of their youth. In their youth, humans seem to learn an ingrained phonetic production system, precisely modeling the finest details of pronunciation and rendering them thoroughly automatic and unconscious. This pattern remains ingrained, crystallized, largely unmodified throughout the remainder of life. Studies of sound change in progress depend upon this fact in order to infer that the way teenagers were pronouncing things decades ago is much the same as the way the very same, now much older individuals pronounce things today. Thus a snapshot of the speech community at one moment in history that shows a pattern of age-grading in phonetic forms, could only reflect historical changes in progress if the people had this crystallized pattern of phonetic production.
These two facts may explain the near-merger situation. Children hear speech around them in which some phonetic distinction occurs and is associated with particular sets of words; they learn to match the pronunciations and to use the right sound with the right words. But what happens when the children mature? First, the pattern of production has become ingrained and remains unchanged. And second, their perceptual abilities degrade. Some phonetic distinctions may quite well be so small that a degradation in perceptual capabilities could make it impossible for mature speakers to categorize the sound itself. Thus speakers could acquire a distinction, retain it in production, yet be unable to use it in perception.
Notice that children can learn such a distinction from adults that produce it, whether or not their adult models are able to perceive the distinction. Children learn language based on the productions of their models, not based on what their models can perceive! In this way a historically stable situation could occur in which two sounds are produced with a small phonetic difference, by an entire community, but where none of the adults are able to actually make linguistic, perceptual use of the distinction.
Experiments to test predictions of this hypothesis are quite feasible. For example, children should pass a categorization test while adults fail it. Such experiments are beyond the scope of the present work; I hope to carry them out in future research.