NEVIS AND JOSEPH EXPOSED:

AN APPLICATION OF ZWICKY & PULLUM'S CRITERIA CRITICALLY ANALYSED

by

A. Ashbaugh, S. Clarke, & S. Micklewright

This paper is the follow-up to our presentation which focused upon the distinctions which can be made between clitics and affixes. We will look at the criteria formulated for this purpose by Zwicky and Pullum, and why they found it necessary to devise such criteria. We will then examine how Nevis and Joseph use these in their 1992 article, "Wackernagel affixes: Evidence from Balto-Slavic". The bulk of the paper, however, will show that Nevis and Joseph's article is flawed in its approach, confused about its terminology, and weak in some of its theoretical assumptions regarding both motivations and explanations for the observed behaviour of the Lithuanian reflexive.

Clitics and affixes are notoriously difficult to define. Indeed, this difficulty has led a number of linguists to abandon any attempts to do so, resulting in their simply regarding clitics as a descriptive cover term. This makes Zwicky and Pullum's 1983 article, "Cliticization vs. Inflection: English n't" particularly ambitious. In this article the authors explicitly set out to draw distinctions between the two bound forms, leading them to devise a six-fold set of criteria to do just that. They then go on to apply these criteria to the contracted negative form n't. This form, which has traditionally been viewed as a clitic, is in fact shown to be a negative inflectional affix as a direct result of the application of the criteria. This result exemplifies the fact that one can often make new grammatical evaluations, leading to the reformulation of previously well-defined grammatical categories.

At this point, it would be useful to note the salient differences between clitics and affixes. Clitics can be subdivided into simple clitics and special clitics. A simple clitic is a contracted version of a full form word which can be replaced by that full form word in the same syntactic position. Clitics lack main word stress, usually because they do not contain a vowel. Hence, post-lexical phonological rules have to be applied to ensure that they are attached to a host, making the clitic part of a larger phonological domain. This is necessary because clitics are prosodically deficient and can never occur in isolation. This larger domain is referred to as the clitic group, consisting of the clitic plus its host (or anchor) word. Special clitics, on the other hand, have no corresponding full form. They only occur as bound morphemes which can be adjoined to their host in specific syntactic contexts. Overall it can be said that clitics do share certain properties of their full forms, but, as they have to be phonologically attached to a host, differ in that they have none of the independent qualities found in words. Clitics are also quite problematic in that words which have become simple clitics through fast speech can easily be reanalysed as special clitics, precisely because they have no corresponding full form. In addition, if a special clitic always attached to a host of a specific syntactic category then it could be described as being an inflectional affix.

Affixes are also bound to their stem word but unlike clitics they become an integral part of that stem phonologically, syntactically, and semantically. Therefore, an affix plus stem can be seen to form one functional unit. Affixes can also be sub-divided into inflectional and derivational affixes. Inflectional affixes are used to indicate grammatical relations. Amongst such inflectional affixes in English are the markers -ed for past tense, -(e)s for plurality, and -est for superlative adjectives. Inflectional affixes have no effect on the class of the word to which they attach. Conversely, derivational affixes not only create a new word but also a new class. These descriptions of clitics and affixes, although seemingly quite well defined, are still rather indistinct from one another. This is the principal reason why Zwicky and Pullum devised their six criteria. These consist of distinguishing characteristics displayed by both clitics and affixes, thereby giving a very easy method of identifying each. The criteria are listed below:

In applying the criteria, it is essential to remember that in order for an item to be seen as either a clitic or an affix, it must fulfil more than one of these characteristics. In addition, an item may still well be a clitic or an affix, despite the fact that it may not adhere to every distinction given for either category. Nevis and Joseph use these criteria in their 1992 article, as they believe this to be the best method there is for distinguishing words from non-words and affixes from non-affixes. They focus their attention on the behaviour of the mobile reflexive marker -s(i)- in Lithuanian. This reflexive marker has traditionally been referred to as an affix, but because it shares qualities of both affixes and clitics, Nevis and Joseph found it crucial to put it to the test. Firstly, emphasis is made on the "mobility" of -s(i), which is more characteristic of clitics than of affixes. Nevis and Joseph state, "mobility of affixes is not unprecedented" (p.93), and "mobility alone cannot be taken as sufficient evidence for demonstrating clitic status for a given element" (p.94). The general principle concerning the mobility of -s(i)- is that if the host word to which -s(i)- attaches has no prefixes then it attaches to the end. However, if the host has one or more prefixes, then -s(i)- attaches in second position. This then conforms to "Wackernagel's Law" which describes the common placement of clitics in second position, therefore suggesting that this element in Lithuanian could be a clitic. Nevis and Joseph find other examples of this type of occurrence both in clausal units (Russian question marker li) and phrasal units (Macedonian definite article). However, if -s(i)- were a clitic, it would fail to comply with the Lexical Integrity Principle, which states:

This therefore implies that no language will have word-internal placement of clitics, or endoclitics, contradicting the Wackernagel behaviour of -s(i), as summarised above. Nevis and Joseph attempt to get round this by suggesting that "-s(i) would be an instantiation of a new type of mobile affix� one that moves within a word, in keeping with a version of Wackernagel's Law at word level" (p.97). They reinforce this view by applying a set of criteria to s(i). These selected criteria are a representation of the Zwicky and Pullum criteria but are "repeated and elaborated upon" (p.98) so as to state the most relevant tests applicable in this particular case:

Nevis and Joseph proceed to exemplify each criterion with reference to Lithuanian -s(i). In their view, all four criteria point towards an affixal analysis. For example, they note that the reflexive only attaches to verbs and deverbal nominals, not to regular nouns or adjectives, and therefore assume that the crucial stem selection is [+verb]. With regards to b), they offer a paradigm containing irregular gaps when the reflexive is attached to the end, even though declension is otherwise regular with non-reflexive deverbal nouns, and if -s(i)- is after a preverb. They continue to say that criterion c) is also met, as no syntactic rules affect -s(i)- independently. Finally, Nevis and Joseph present criterion d) which provides their strongest evidence. They find several examples of morphophonological idiosyncrasies, including the raising of low vowel <e> or /æ/ in the 1st and 2nd person plural to a higher mid-vowel <>. This raising typically applies when -s(i)- immediately follows the vowels, but it does not necessarily apply in all cases. A selection of their data are given below:

Nevis and Joseph explain such exceptions by suggesting that this 'morpheme-specific process' of raising occurs only if -s(i)- immediately follows 1st person plural -me or 2nd person plural -te. They also describe -s(i)'s effect on word stress as being similar to that of other affixes:

They conclude from their findings that Lithuanian -s(i)- is best taken to be an affix despite its mobility, and account for its variable position by referring to it as a word-level instantiation of Wackernagel's Law.

After close examination, however, several questions arise: How plausible are their methods, how valid are their findings, and how reliable is their evidence? For although the article arrives at an arguably acceptable conclusion about the element in question, this is not due to the careful analysis of data on the part of the authors. Nevis and Joseph appear to be ignorant of the proper use and application of Zwicky and Pullum's criteria, as evidenced by their reformulations and elaborations. This is exacerbated by their preconception that the element is definitely an affix, and the consequent influence which this bears over their interpretation of the data. The theoretical foundations for their conclusions are often lacking in substance, a fact which they themselves mention in the occasional end note. Moreover, the authors seem to have a general misunderstanding of several of the characteristics of clitics described above.

Nevis and Joseph get off to a bad start in the first paragraph of their paper and are subsequently plagued throughout its length by their confusion over clitic behaviour:

By singling out affixes as prototypical bound elements, N&J ignore the fact that clitics are also bound. This usage continues in the second paragraph, where they take "affix" and "bound element" to be synonymous. This may seem to be a trivial point, but this misunderstanding does go on to form the basis for their entire article: the Lithuanian reflexive, ignoring these ordering restrictions, is supposedly behaving like a clitic. The ordering constraint which they refer to in fact applies to both affixes and clitics, and can only help to distinguish these elements collectively from free words. More importantly, as a result of their imprecise use of terms, N&J falsely regard mobility as a criterion for clitic status, and in so doing identify clitics more with independent words than they perhaps ought to be. This in turn causes at least one of the defects in their testing of the element.

In order to demonstrate their intentions regarding -s(i), N&J present data from a similar endeavour in Greek, in which they argue that the weak pronouns are in fact affixes and not clitics. Restricting ourselves to analysing the information presented in the paper in question, the evidence they offer for their finding is the following: the Greek pronouns show fixed ordering in relation to one another, occur in interior positions, are not deletable under identity, and show "a variety of idiosyncrasies characteristic of affixes" (p94). This suffers from the same ailment discussed above, in that the authors are unable to recognise the boundedness of clitics and what that entails. Affixal idiosyncrasies aside, the only evidence they give that can distinguish the pronouns as affixes is their occurrence in interior positions. That they obey rigid ordering constraints and are not deletable under identity merely demonstrates they are not independent words, a superfluous finding if the question is whether they are clitic (as had been the general consensus) or affixal. Granted, we do not have all the data, but that which is given here is less than completely convincing. N&J (ironically) claim that this is exactly the sort of operation that needs to be carried out upon the Lithuanian marker, using the Z&P criteria.

Zwicky (1985) makes several comments related to the use of linguistic tests in general, and adds that Z&P's criteria are, for the most part, stated as tendencies, and none are bi-directional. Thus, someone wishing to test a given element against such criteria must be both careful in intent and faithful to the original formulations. N&J fail in this respect at least half the time. They initially offer a list of paraphrased criteria, part of which is repeated below, as suggested "criteria for affixal status" (p.93):

With respect to the direction of the inference, d., f., and g. above are fine. However, the phrasing of c. implies a result that is backwards to that of the original, i.e. if there is high selectivity, you have an affix (unlike Z&P's, if you have an affix, it probably exhibits high selectivity). In any event, not one of these is stated as a tendency. The formulation of e. is far worse, however. Z&P's criterion (E), above) states that the combination of [word+affix] can be treated as a syntactic unit, but not the combination of [word+clitic]. N&J, on the other hand, are saying that words and clitics individually can be treated as syntactic units, but not affixes. This is a complete misinterpretation of the criterion given by Z&P, and is a blatant example of N&J's carelessness. This test in particular will be discussed further below. Finally, one must assume that N&J's h. above is the equivalent of Z&P's F), or a characteristic of "closure" described by Zwicky (1985), that certain inflections "close off" words to further affixation. The latter, however, is again used to distinguish bound elements from free. The criterion as formulated by N&J, though reasonably following from that of Z&P, is not itself the same criterion. We can see, then, that N&J have already misunderstood the use and nature of these criteria, before they even get to the data at hand.

This is not the only initial problem with the article, though. The authors say that -s(i)- has "never been subjected to rigorous testing... to determine its affixhood" (p.94). Thus they set out, not to determine what type of element the Lithuanian reflexive might be, but rather to prove that it is really an affix. By looking for affixal properties, they ignore a crucial characteristic of clitics. This approach causes at least one fatal flaw, to be discussed in greater length below.

To begin, we will briefly mention N&J's second criterion, which they state in almost exactly the same way as Z&P do, i.e. as a tendency, with the inference in the right direction. The results are consequently valid and we need not discuss it any further.

Next, we shall consider N&J's fourth criterion, which, as stated above, was given as the strongest evidence for -s(i)'s affixal status. It is similar to Z&P's phrasing, except that it is less tentative than the original. Again, the results are reasonable, but N&J still need to remember that there are numerous examples of demonstrably clitic elements which exhibit idiosyncratic behaviour. Hence they may have placed too much importance on their data, especially in light of other results to be discussed. This is not a problem with the use of the test itself, however.

The authors do have a problem though when they argue that -s(i)'s behaviour with respect to stress is further evidence of their conclusion, saying, "if -s(i) were a clitic, it might well be expected to be invisible to lexical stress assignment..." (p.100). This idea seems thoroughly unfounded, and one need only look to Macedonian for a clear counterexample. Macedonian stress almost always falls on the antepenultimate syllable, or the first syllable of a disyllabic word. This stress rule treats dependent clitics (definite article and pronouns) as part of the word, and can therefore cause the stress to shift:

In addition, Kanerva (1987) presents data from Finnish where both clitics and affixes behave identically with respect to stress assignment. In Finnish, primary stress falls on the first syllable and secondary stress falls on the third or fourth, then every other syllable. No word-final syllable may receive stress. Clitics count as parts of the word for stress assignment, since secondary stress can fall on a syllable immediately preceding them, and they themselves may also bear secondary stress:

Thus, although there do exist languages where clitics can be "invisible" to stress (e.g. Italian clitic object pronoun, the addition of which causes an apparent violation of stress-assigning rules), the phenomenon does not constitute an absolute grammatical principle. There is sufficient evidence to show that some clitics do affect stress in divers ways, meaning that there is little basis for N&J's claim.

So far, the tests have not been exceedingly problematic; however, N&J remedy this situation with the application of their next criterion. First, one must be wary of the way in which they state this criterion, both at the beginning of their paper (as already discussed), and in the elaborated version. The Z&P test proposes that if a given combination of elements can be affected by syntactic rules, then it must form a syntactic constituent, and therefore must be an affixed word [word+affix], and not a clitic group [word+clitic]. N&J's test, on the other hand, states that syntactic rules can affect [word+affix] combinations (as Z&P say), but then confuses the second half of this criterion with one from Zwicky (1985): "A word can serve as a syntactic constituent, and therefore can be subject to syntactic processes; a clitic, however, is only a proper part of a word-like construct, and should be immune to such processes" (p.288). An affix, as a proper part of a word, will therefore be immune to syntactic processes. While true, this does not help N&J in their quest to determine the nature of -s(i), as the criterion which they are using does nothing more than highlight similarities between affixes and clitics as bound elements, and thereby to distinguish these two from free words. As such, the criterion does not apply to the investigation at hand. This failure is another result of N&J's misidentification of affixes alone as being bound. Even if the N&J criterion were applicable, as used it would not yield any results. N&J state that they "know of no syntactic rules that affect the reflexive verb without also affecting -s(i). Nothing in syntax breaks up the combination of verb + reflexive morpheme, moving the reflexive out of the domain of its verb for instance" (p99). This sounds similar to how Z&P demonstrate their criterion on known English clitics, but must be read carefully: "no syntactic operations exist which treat a word combined with one of the clitics 's or 've as a unit" (Zwicky & Pullum 1983, pp.505-6). In keeping with their own criterion, Z&P say that [word+'s/'ve] is not manipulable by the syntax. N&J, on the other hand, say [word+s(i)] is manipulable by the syntax, but to their knowledge -s(i)- itself is not. The implication on which they're working must be that if syntactic rules affect it, then it must be a clitic. However, that no syntactic rules affect it does not mean that it isn't a clitic, or even that it is. No inference can be made. The test, as applied, is useless.

Finally, the first criterion which N&J apply to -s(i)- is one of selectivity. N&J argue that -s(i)- attaches only to stems which are [+verb]. This analysis is riddled with problems, resulting in large part from the authors' skewed approach to the task. As phrased, N&J's affixes and clitics both show varying degrees of selection with respect to "hosts", whereas Z&P's affixes select for stems. This is a deceptively minor point which leads N&J astray. Affixes have only one host to worry about: the stem to which they attach. By assuming that -s(i)- is an affix, then, N&J take host and stem to be synonymous, missing the crucial distinction involved with clitics. Klavans (1985) points out that clitics, unlike affixes, are constrained in more than one way. Clitic positioning is a matter of the occurrence of a clitic in a particular syntactic or morphological structure, while clitic attachment is the phonological liaison with an adjacent word. She calls the structure which determines the clitic's placement the "host phrase" and the word to which the clitic phonologically attaches the "host word". Clitics need to know about both these hosts. In their discussion of the "affix-and-word Wackernagel Law" parallel to that for clitics and their respective domains, N&J are drawing an analogy along the lines of:

i.e. affixes will attach to their words in the same way that clitics attach to their phrases or sentences. This is fine as far as it goes, but N&J neglect the clitics' other dimension, making their analogy incomplete. They must consider the following as well:

i.e. an affix-like element exhibiting clitic behaviour must attach to an adjacent morpheme in the same way that known clitics liaise with, or phonologically attach to, an adjacent word. Z&P actually say, "The degree of selection between the clitics and the words preceding them is low" (p.504, emphasis ours). That is, a clitic's promiscuous behaviour is with respect to its host word, meaning that N&J's so-called Wackernagel affix will have to pay attention to (or not, as the case may be) the immediately preceding morpheme. Looking at N&J's own data, we are confronted with distinctly clitic-like behaviour, as the reflexive attaches to all manner of different morphemes:

This test, worse than being non-applicable as was the previous one, is actually a counter-argument to the conclusion the authors put forth. Had they not so passionately desired -s(i)- to be an affix, they might not have overlooked the important distinction in the terms "host" and "stem", and its implications. Thus the final score is two points in favour of -s(i)'s being an affix, and one in favour of its being clitic� less than overwhelming evidence.

In addition to these tests, though, the authors also give various theoretical grounds to motivate and explain both the need for testing to be done, and for their conclusion. N&J appeal to the Lexical Integrity Principle and its corollary precluding endoclitics (see above), which, they say, a clitic -s(i) would violate. N&J use the LIP as yet another argument for -s(i)- to be an affix, but mention, "It may be that Lexical Integrity will turn out only to be a tendency and not an absolute; [there are] some troubling cases from Fox in which full noun phrases, adverbials, etc. can intervene between the combination of inflectional agreement prefix plus preverb and the rest of the verb" (Note 23). The principle itself follows from a linear theory of the organization of grammatical components, in which all morphology is completed before any syntax. There are alternatives to this type of view, however. Sadock (1991) for one rejects a purely hierarchical model, arguing that it cannot account for the behaviour of many a clitic. He states, "bound morphemes can shade almost imperceptibly from prototypical clitics... to normal derivational morphemes... and to inflectional morphemes, ...a state of affairs that is utterly incompatible with a theory whose components are arranged [hierarchically and linearly]" (p.54). This can hardly be considered an explanatory mechanism then. The bar on endoclitics is not without exception either. Clitics can be inflected, as in Icelandic (data from class notes), or can intervene between the verb stem and tense/aspect markers, as in Portuguese (Spencer 1991):

N&J themselves recognise the potential weakness of Lexical Integrity, in another note where they mention some dialects of Lithuanian that can place -s(i)- inside the personal endings, e.g. sùka-si-m 'we spin' (as opposed to standard sùka-m-s), and which would therefore show it to be an affix by their account. However, they go on to say, "positioning... provides an argument only if Lexical Integrity is vindicated as a grammatical principle" (Note 11). Recalling the Greek data which were offered at the beginning of the paper to demonstrate rigorous testing and offer evidence of mobile affixes, only position was able to support their assertions as given. N&J's claims are again weak and inconsistent.

The authors also justify the notion of a Wackernagel affix by saying such a thing "is actually expected on theoretical grounds, given claims that have been made, e.g. by Baker (1985) with his 'mirror principle', of parallels between syntactic organization and morphological organization..." (p.97). Baker's Mirror Principle states, "Morphological derivations must directly reflect syntactic derivations (and vice versa)" (p.375). That is, any constraint on one system will automatically entail a constraint on the other. Thus, one might think that a law stating that a certain type of element (a clitic) prefers to stay in a certain syntactic position would constitute a morphological constraint with respect to a given morpheme (affix) within a word. Baker goes on to say, however, that "[clitics are] elements whose distribution is (roughly speaking) fixed syntactically rather than morphologically, but which become part of a 'host' word phonologically. Since their distribution is not fixed morphologically, they are strictly speaking not part of the morphological derivation of a form, and hence the mirror principle does not apply to them" (p.400). If the Mirror Principle does not apply to clitics, how can it effect a constraint on another part of the grammar based on their behaviour? This is an extremely unstable position from which to argue for the existence of an element.

And one has to wonder exactly how Wackernagel -s(i)- is at all, when one looks at evidence buried in Note 6 of N&J's article. There are grammatical morphemes which can sometimes occur before the first preverb, or between the first preverb and -s(i), e.g. negative ne, or the strengthening particle -gi-:

NJ argue that because of the nature of these morphemes, they may be able to fuse with the preverbs, and so only count as one prefix as far as -s(i)- is concerned. However, they still remain as counterexamples to the principle supposedly governing -s(i)'s behaviour. This again demonstrates the shaky foundation on which the authors base their article.

Therefore, we have shown that the "rigorous testing" of the Lithuanian reflexive by Nevis and Joseph was sloppy at best, and resulted in an utter misinterpretation of the data at worst. By assuming that -s(i)- is an affix, they missed an extremely important attribute of clitics, in one case causing their data to contradict their conclusion. They misunderstood a crucial distinction in terms (stem, host word, host phrase) in addition to their confusion over clitics as bound elements. Their reformulations of Z&P's original criteria also led them astray, rendering one of their tests irrelevant, and perhaps leading them to overemphasise the strength of another. The evidence in favor of a clitic -s(i)- is just about even with, if not marginally stronger than, the evidence against.

The authors' theoretical foundations for their views were equally found to be lacking in substance. In commenting about the nature of linguistic tests in general, Zwicky (1985) remarks that criteria such as those of Z&P are "more analogous to medical diagnosis.... The tests point to characteristic symptoms of a linguistic state of affairs..." (pp.284-5), and in the case of such terms as "word" and "clitic", only lists of symptoms are available. Because of this, no single criterion is sufficient to classify an item one way or another; rather there must be a balance of data strongly in favor of one conclusion. With this in mind, Zwicky also suggested a "metacriterion" which seems stunningly appropriate in the given circumstances: "In the absence of clear evidence classifying an item one way or the other, we should assume that the item is a word (or an affix)" (p.289). Thus N&J's conclusion is ultimately vindicated, not by their application of Z&P's criteria to Lithuanian -s(i)-, but by application of Zwicky's metacriterion to Nevis and Joseph's testing.

References

Baker, M. 1985. The Mirror Principle and Morphosyntactic Explanation. Linguistic Inquiry 16, 373- 416.

Carstairs-McCarthy, A. 1993. Current Morphology. London: Routledge.

Crystal, D. 1991. A Dictionary of Linguistics and Phonetics. Oxford: Blackwell.

Giegerich, H.J. 1992. English Phonology: An Introduction. Cambridge: CUP.

Kanerva, J. 1987. Morphological Integrity and Syntax: the Evidence from Finnish Possessive Suffixes. Language 63, 498-521.

Katamba, F. 1993. Morphology. Basingstoke: Macmillan.

Klavans, J. 1985. The Independence of Syntax and Phonology in Cliticization. Language 61, 95-120.

Matthews, P.H. 1991. Morphology, 2nd ed. Cambridge: CUP.

Nevis, J. & B. Joseph. 1992. Wackernagel Affixes: Evidence from Balto-Slavic. In G. Booij and J. van Marle (eds.), Yearbook of Morphology 1992, 93-111.

Sadock, J. 1991. Autolexical Syntax. A Theory of Parallel Grammatical Representations. Chicago: Chicago University Press.

Spencer, A. 1991. Morphological Theory. Oxford: Blackwell.

Zwicky, A. & G. Pullum. 1983. Cliticization vs. Inflection: English n't. Language 59, 502-13.

Zwicky, A. 1985. Clitics and Particles. Language 61, 283-305.

Back

Morphological Theory