What is a “word?”
We have said that, in the prototypical case, clitics have the syntactic status of independent words but the phonological status of affixes. This means that a clitic does not constitute an independent phonological word. It may seem surprising to speak of a difference between phonological words and syntactic words; but consider the traditional description of the “contracted” forms of English auxiliaries in sentences like the following:
(5) a I’m going home now.
b She’ll be coming around the mountain.
In traditional terms we say that forms like I’m and she’ll are contractions, meaning that two separate syntactic words, a pronoun and an auxiliary verb, are pronounced as a single phonological word. The second elements in these contracted forms (the reduced auxiliary verbs, ’m and ’ll) are shortened to the point that they can no longer be pronounced meaning fully in isolation, since they contain no vowels. Thus, they are no longer independent phonological words, and are often analyzed as clitics.
How are we to tell the difference between phonological words and syntactic words? How can we identify the word boundaries of each type? Let us begin by focusing on the phonological word.
First, a phonological word is the smallest possible utterance in the language; speakers do not normally say anything which is smaller than a complete phonological word. As noted above, the reduced auxiliary forms ’m and ’ll cannot be pronounced meaningfully in isolation, because they are too “short” to be independent phonological words. Furthermore, deliberate pauses and intonation breaks occur only at phonological word boundaries, and never in the middle of a phonological word.1
Second, each phonological word is composed of one or more well-formed syllables (i.e. syllables which satisfy the constraints on syllable structure in the language). The contracted forms I’m and she’ll each contain only a single syllable, so they cannot contain more than one phonological word.
Third, stress placement is frequently determined by phonological word boundaries, and each phonological word normally contains only a single primary stress. Notice, for example, the contrast in stress placement indicated in the last few words of (2a) vs. (2b); the acute accent in those examples marks primary stress. In particular, note that the root gel- in (2a) does not receive a primary stress because it is part of a larger phonological word.
Many other phonological rules are also sensitive to phonological word boundaries in various languages. Harmony systems frequently apply only within a phonological word, and fail to apply across phonological word boundaries. Many allophonic rules apply only at phonological word boundaries, e.g. the devoicing of word-final obstruents in various languages. Such rules provide additional tests for identifying phonological words. These rules typically do not treat the boundary between the clitic and its host as a word boundary; they are likely to treat the two together as a single phonological word.
Let us turn now to the identification of syntactic words. In the approach we have adopted, the terminal nodes of a PS tree are always words. Thus, the syntactic word is the smallest possible constituent in the language: the smallest unit which can be moved, replaced, or deleted by syntactic operations, assigned a position by Phrase Structure rules, etc.
In trying to identify clitics, the main problem will be to distinguish phonologically bound forms which are syntactically free(clitics) from those which are syntactically bound(affixes). There are two general principles which can help in this regard. First, the arrangement of affixes in a word is normally quite rigid, while the order of words in a clause or phrase is often more flexible; and second, lexically specific irregularities are more typical of affixation than of syntactic word combinations. Zwicky and Pullum (1983) offer the following specific criteria for distinguishing between clitics and affixes.
a Affixes “select” their stems; that is, they usually appear only on stems of a certain category (nouns, verbs, etc.). Clitics, on the other hand, may attach to a wide variety of word classes.
b Restrictions on co-occurrence, or gaps in the set of possible combinations, are more common with affixes than with clitics.
c Irregular morphophonemic changes and suppletive forms are more common with affixes than with clitics.
d Irregular semantic composition is more common with affixes than with clitics. The meaning of the word + clitic combination is usually predictable from the meaning of each separately, whereas a particular stem + affix combination may take on an unpredictable meaning.
e Clitics can attach outside other clitics (e.g. you’d’ve), but affixes usually cannot attach outside clitics.
f Syntactic rules will re-order affixed words (affix plus stem) as a unit, but generally treat a clitic and its host as separate elements. The following examples strongly suggest that the contracted form should’ve contains two distinct syntactic words, since it cannot be fronted as a unit in question formation. This means that the reduced form ’ve (from have) is a clitic, rather than an affix.
(6) a You should’ve seen it.
b *Should’ve you seen it?
c Should you have seen it?
Notice that all of these criteria are stated as tendencies, rather than as absolute principles. Specific clitics in particular languages may exhibit these properties in various combinations and to varying degrees. Our job as linguists is to find the analysis that best accounts for the whole range of properties of a given form.
1. Of course, hesitations, self-correction, and false starts can and frequently do produce pauses in the middle of words.
