¹ In the body of this paper, investigation enters into our argument as though it were a single morpheme; for an argument based on the fact that investigation is a sequence of morphemes, see fn. 2.

² And it is theoretically possible to resolve some alternative analyses: what appears at first to be non-unique may have a unique solution when more data on distribution are brought to bear on the argument. In our bop example, it might be argued that, irrespective of meaning, the prefix re- in English appears with nouns which show secondary derivation from verbs, and hence are composed of at least two morphemes. But bop is a single morpheme and not a sequence of morphemes like investigation. Since the re- in rebop appears in a different distribution (before a single morpheme) from the re- in reinvestigation (before a sequence of morphemes), the phonemic identity of re- in the two instances is fortuitous. Such an appeal to distribution appears to me to be a device of desperation; for if the re-of rebop marked the meaning ‘again’ or ‘back’, one would simply restate the distribution of morphemes which appear after the prefix re-, rather than question the identity of re- in its various distributions.

³ For finding morphemes, semantic evidence relevantly admitted will show (1) that two morphemes are different in meaning; (2) that two different forms are the same in meaning; (3) that some features of meaning are common to many morphemes otherwise different in meaning. The first point is comparable to the aspect of distinctive features in phonemic analysis; but it does not work out so neatly in a morphology involving bound forms, where two morphemes may constitute a minimum sequence. To find the meaning of morpheme A and of morpheme B, we may be forced to deal with the sequences A + X and B + X, and we may often wonder what the morpheme X contributes to the total meanings of these sequences—the meanings from which we expect to abstract the separate meanings of A and B. In (2), the meaning equivalence in certain distributions is the basis for setting up morpheme alternants. In (3), the class meaning common to different morphemes is probably always discovered after the existence of a morpheme class has been proved by distribution alone.

⁴ For our final example we are indebted to Hans Wolff, who has recently returned from Oklahoma, where he worked with the genetically isolated Yuchi language, attempting to find a more satisfactory statement on person markers than that reported by Günther Wagner, Yuchi (extract from Handbook of American Indian Languages, Vol. III; New York, 1934).

⁵ For the solution of this problem, see Hans Wolff, Yuchi Phonemes and Morphemes with Special Reference to Person Markers (in preparation).