If we now seek a clearer method for obtaining generalizations about the structure of utterances in a language, it should preferably deal with the simplest observables. These are the morphemes, which are uniquely identifiable and easy to follow. Constructs such as ‘morphological levels’ may be useful in particular cases, but there is an advantage in avoiding them if we can achieve the same results by direct manipulation of the observable morphemes.
- Harris (1946), a pag.161

We may say that each morpheme can have only one phonemic form, so that for example the English plural endings /s/, /z/, /Əz/ (as in 'books', 'chairs', 'glasses') constitute three morphemes, and 'am', 'are' constitute two morphemes. Alternatively, we may include each of these sets in a single morpheme, if we say that different phoneme sequence constitute positional variants of one morpheme when they are complementary to each other.
- Harris (1946), a pag.162

In some languages, relatively few morphemes occur in exactly the same environments as others: 'poem' occurs in 'I’m writing a whole—this time', but 'house' does not. Both morphemes, however, occur in 'That’s a beautiful—'.
- Harris (1946), a pag.163

[…] morphemes having slightly different distributions are grouped together into one class if the distributional differences between their environments correspond to the distributional differences between the morphemes. That is, if 'poem' and 'house' differ distributionally only in the fact that 'poem' occurs with 'write' and 'house' with 'wire', and in comparable differences, and if 'write' and 'wire' in turn differ only in that 'write' occurs with 'poem' and 'wire' with 'house', and in comparable differences, we put 'poem' in one class with 'house', and simultaneously put 'write' in one class with 'wire'.
- Harris (1946), a pag.164