The structural quantification derives from the fact that some non-occurrences of phonemes are represented by long components and others merely by the non-occurrence of one component with a particular other component in a position where the first component otherwise occurs. Let us take the non-occurring */sbin/ and */stend/ in English. If the cluster-long unvoicing component is ¯, we may say that 'spin' is /zbin/; the sequence /zb/ (= /sb/) is impossible since ¯ always extends over the whole cluster in which it occurs. On the other hand the general vowel component contained in /e/ occurs after /st/, but only with the particular quality component of /æ/ and not that of /e/: 'stand' but not '*stend'. There is no long component excluding the /e/ quality component from the position after /t/ or before /n/ or between clusters, since the /e/-quality component occurs in those positions: 'tend', 'spend'. Therefore all we have is the fact that while the general vowel component occurs in between /st/ and /nd/, it does not occur there with the /e/-quality component, although it does occur with that component elsewhere. We may then say that forms like */sbin/ are excluded from the phonetic structure as it described by our components, while forms like */stend/ are not excluded.
- Harris (1944), Pag. 204