A Quantitative Approach to the Morphological Typology of Language

in meaning. Another characteristic would generally be agreed on. Every word must have at least one root morpheme. Hence in a one-morpheme word, that morpheme is necessarily a root. In contrast, derivational and inflectional morphemes need not occur, and there are some languages, so-called root or isolating languages, in which derivational and inflectional morphemes are rare or perhaps do not occur at all. Derivational morphemes may be defined as morphemes which, when in construction with a root morpheme, establish a sequence which may always be substituted for some particular class of single morpheme in all instances without producing a change in the construction. If the class of single morphemes for which the derivational sequence may substitute contains one of the morphemes in the derivational sequence itself, we call the sequence endocentric; if not, then it is exocentric. For example, duckling in English is a derivational sequence, since it may be substituted anywhere for goose, turkey, etc. wi hout change of constructional meaning. Since duck is included in this class of single morphemes for which duckling may substitute, -ling is here an endocentric derivational morpheme. Singer is an exocentric sequence, since the class of single morpheme sequences for which singer may substitute consists of single-morpheme nouns only, and does not include the verb sing. Hence -er is an exocentric derivational morpheme. We can now define the inflectional morpheme simply as a nonroot, nonderivational morpheme making the three classes exhaustive and mutually exclusive. Inflectional morphemes, like derivational, need not occur at all in any particular language. When it is part of a word pattern, however, its appearance in the appropriate position is compulsory like that of the root. One member of the class is frequently zero. In these instances, the absence of an overt phonemic sequence shows itself as significant because the word in this form has definite syntactic limitations on its uses; e.g., the nominative singular in Turkish or the noun singular in English. We finally reach what is in some ways the most difficult problem, the definition of the word unit. It is clearly fundamental to the purpose of the present study inasmuch as all the indices involve the number of words. In most instances this is explicit; sometimes it is tacit, as in the index of agglutination, in which the number of morpheme junctures is always one less than the number of words. There is at present no general agreement on this topic. Some deny the validity of the word as a linguistic unit. Others admit it, but deny that it need be taken into account in the description of a 191 This content downloaded from 157.55.39.123 on Tue, 18 Oct 2016 06:10:03 UTC All use subject to http://about.jstor.org/terms INTERNATIONAL JOURNAL OF AMERICAN LINGUISTICS particular language. Some say the word is definable only for each language in a separate ad hoc fashion. Some define it in phonological, others in morphological terms. In practice, however, the word continues to be the key unit of most actual language descriptions. Of the two basic types of overall definitions of the word, the phonological and the morphological, the former is clearly insufficient for the purposes of the present study. In phonological definitions, we define the word in terms of some single phonological characteristic, or through a combination of characteristics which serve as markers. These markers are usually stresses or boundary modifications of phonemes; i.e., junctures. Besides the fact that the use of phonological markers to define the word sometimes leads to the isolation of individual units which we should not wish to call words on other grounds, many languages do not have such phenomena, so that a phonological procedure cannot lead to a universal definition. The other basis has been called morphological, since it is based on the distribution of meaningful elements. Of definitions of this kind, Bloomfield's characterization of the word as the minimum free form is the most satisfactory in that it is universally applicable and points in the right direction to freedom or absence of freedom, that is, bondage as the basic criterion. The actual test of freedom, ability to occur in isolation, is, however, difficult to apply in practice and leads to unusual results. For example, the in English would not be a word by Bloomfield's test. The procedure adopted here can only be briefly outlined. It has led to results satisfactory for this study in the relatively few doubtful cases regarding the existence or nonexistence of a word boundary in the languages under consideration. Instead of asking whether a particular minimal form is bound or free in general, as is usually done, the present treatment is in terms of morphs in particular contexts. This allows us, for example, in Latin to make ab, "from," a free form as a preposition but a bound form as a verb prefix in abduco, "I lead away." What is specified as bound or free is not a morph as such but a contextually determined class of mutually substitutable morphs. Such a class is here called a morph substitution class (MSC). This notion is expanded to include a sequence of morpheme substitution classes which may in all circumstances be substituted for a particular MSC and none of whose members are identical in membership with it.3 It is convenient to use the term nucleus to cover both individual MSC's and such substitutable sequences. Having broken up an utterance into nuclei in this manner, we now test each nucleus boundary to see if it is a word boundary or not. A nucleus boundary is a word boundary if it is possible to insert an indefinitely long sequence of nuclei. If it is an intraword boundary, either no nucleus can be inserted or a fixed maximum number can be. For example, in the sentence "the farmer killed the ugly duckling" there are nine morphemes: (1) the (2) farm (3) er (4) kill (5) ed (6) the (7) ugly (8) duck (9) ling; seven nuclei: (1) the (2) farmer (3) kill (4) ed (5) the (6) ugly (7) duckling; and six words: (1) the (2) farmer (3) killed (4) the (5) ugly (6) duckling. There is an intraword boundary at "kill-ed" because no nucleus may be inserted. 3 This is necessary in order to exclude endocentric phrases in which a sequence of words can always be substituted for the head or chief number. A sequence of adjectives followed by a singlemorph noun would be a nucleus were it not for the proviso of dependence among its members. Adjectives are not bound to nouns in English, for example, because they occur in predicative adjective constructions also. For the basic ideas of the MSC and the derivational sequence, I am largely indebted to the stimulus of the writings of Zellig S. Harris and Rulon S. Wells. The resemblance to the notions of focus class and expansion of the latter writer is particularly close. See especially R. Wells, 1947, pp. 81-117. 192 VOL. XXVI This content downloaded from 157.55.39.123 on Tue, 18 Oct 2016 06:10:03 UTC All use subject to http://about.jstor.org/terms NO. 3 QUANTITATIVE APPROACH TO MORPHOLOGICAL TYPOLOGY OF LANGUAGE