RELATIONS AMONG PHONEMES, and the limitations of distribution of particular phonemes, are not presented in linguistics as an essential part of the individual phonemes. There exists no method which would enable us to say ‘/b/ is phonemic everywhere except after /s/’ or ‘/t/ is a phoneme except after initial /k/, etc.’ Instead we say that /b/ and /t/ are phonemes, and then tack on statements which correct the phonemic list by pointing out that /b/ does not occur after /s/, i.e. that there is no allophone occurring after /s/ which is assigned to /b/. If a number of phonemes have identical distributions, a single statement is devoted to them all. We say, for example, that English /ŋ/ occurs before no consonant other than /g, k/, or that morpheme-medial clusters in English hardly ever include both a voiceless consonant and a voiced one which has a voiceless homorganic counterpart: we get /ft/, /ks/ in after and axiom, but not /vt/, etc. If a phoneme occurs in few positions as compared with other phonemes in the language, as is the case with English /ŋ/ , we say that it is defective in distribution.
- Harris (1944), a pag.183