[…] if we match utterances, we obtain some individual difference between the members of each particular pair of utterances; that is, we obtain discrete elements each of which represents some particular inter-utterance difference. By the method of chapter 5, fn.3 [Unit length], these differences may be expressed as combinations of a few basic differences: the difference between some utterances is exactly one basic difference (e.g. 'tack-tap'); the difference between others is some particular sum of particular basic differences (e.g. 'tack-tip'). We thus obtain discrete elements which can be combined together. These elements are phonemic distinctions, rather than phonemes; i.e. they are the difference between /k/ and /p/ (more exactly, between 'tack' and 'tap', between 'sack' and 'sap', etc.) rather than being /k/ and /p/ themselves. However, for convenience, we will set up as our elements not the distinctions, but classes of segments so defined that the classes differ from each other by all the phonemic distinctions and by these only. These elements are obtained by summing over all distinctions: [k]—[p], [k]—[l] ('pack—pal', 'sick—sill'), [k]—[s] ('pack—pass'), etc.; [l]—[t] ('still—sit'), [l]—[s] ('pal—pass'), etc. In this way we define /k/ to represent all the paired distinctions in which [k] was a member, /l/ to represent all the distinctions in which [l] was a member, and so on. These classes, or phonemes, are thus a derived (but one-one) representation for the phonemic distinctions.
- Harris (1951), a pag.34