GEORGE P. FAUST ter which allophone is used, and any sense of difference here is likely referred to the speaker's attitude, not his dialect. Therefore, the allo- never collide, even when they alternate with one another, and they au be iimplement one other in such a way that in sum they take care of all altuations in which their phoneme occurs DISTRI Complementary distribution is one side of the coin, coNTRASTIVE ITION the other. Sounds contrast most obviously when the difference oduces different words. Thus /p/ /f/ contrast because 'pup' not paf/). And one contrast anywhere in the language is enough to ablish separate phonemes everywhere in the language. It is known quite ldliiltely that [pl between vowels is voiced in normal American speech at least as close to [b] ls thus very close to [b] in the same situation assigned by speakers honetleally as to initial [p]. But since [p] are dll'erent phone mes (cf. the contrast between /pet/ and /bet/), the arity is inconsequential. You can test this for yourself. Just nvlto a friend to pat your "rabid pup' and see whether his reactions are like of the friends you have invited to pat your "rapid pup i the resul llero it is necessary to insist that the difference in meaning initial he difference in sound. Strangers who hear my weakly aspirated (un individual peculiarity) quite easily "misunderstand" me and think Inlitance, that my middle initial is B instead of P. sense of "mis uderstand" here is that they have misclassified a sound, with resultant An meaning. extremely important difference be error of this sort points up one ween phonetics and phonemics. In phonetics a sound can be between a and a /b/ the sense that it can have of each, h the voicelessness of initial and the relatively weak aspira of an initial /b/. The phonetician tries to describe the sound actually duced. But in phonemics there are no gradations. A sound is assigned one phoneme or another, and there is no in-between stage. The.linguistic idonee goes to show that we hear in terms of phonemes and listen to only much of a sound as we need to order to assign it to an established me. This is a fundamental reason why phonemic transcriptions don't d to distinguish among allophones. produced is a We should now be ready for some definitions. A sound a fingerprint. l (ON is a unique historical event, in theory as individual as ever get bviously, only a microscopic sample of the phones produced can orded, and yet the patterning of language is such that linguists In riaslly as confident I as though had a statistically large sample. measure, this is because they have occasion to concentrate on only row variations. The subclasses it ch phones are fitted are ALLo- ONEs, ench of which, though complementary distribution, is distinct