What are the criteria that distinguish clitics/particles from affixes?

Question

This question inspired me to finally ask a question that has been bothering me for years: how does one distinguish clitics and/or particles from affixes, especially when those clitics are phonologically enclitics, like Latin -que or arguably the Japanese case markers ?

My question is especially important when one looks at agglutinative languages, where affixes are often added with little to no phonological change to the root or affix. How can one decide whether those affixes are really true affixes or clitics in such languages? Since affixes belong to the morphology of a language, while clitics typically are described as part of the syntax of that language, it's an important distinction to make.

Let me give you an example of why I find this question both extremely interesting and yet difficult to solve: the Basque language is an agglutinative language with a rich declension system (see an example here). The behaviour of those declensions is AFAIK best explained if they are considered affixes, as a typical declension features instances of:

• Suppletion (the definite absolutive plural is an unanalysable -ak);
• Conditional epenthesis and elision: the dative -i and the cases starting with -e- both take an epenthetic -r- when added after a vowel, including the singular article -a, but not after the plural article -e. Moreover, the plural article actually disappears before the case suffixes in -e-, so that the only distinction between the indefinite and the plural is the presence or absence of the epenthetic -r-;
• Other weird behaviours: the local cases are a weird bunch. Unlike the other cases, adding them to the stem directly doesn't form the indefinite but the singular (except the inessive -n which does take the singular article). To form the indefinite, one has to add -ta- between the stem and the ending. And even more strangely, this -ta- element must also appear between the plural article and the local case ending!

Yet the Basque affixes do also behave in ways we expect more of clitics or separate postpositions:

• They only appear on the last element of a noun phrase, or on the last element of a series of coordinated noun phrases. E.g.: aita altu eta haur txikiarekin: with the tall father and the small child (where -a: "the" and -(r)ekin: "with" appearing only on txiki: "small")
• They can be stacked on top of each other, a phenomenon called "surdéclinaison" in French. E.g.: Ponetarekilakoarekin: poneta-a-(r)ekila-ko-a-(r)ekin: beret-ART-COM-GEN.LOC-ART-COM: "with the one who is wearing the beret"

I hope that with this example it's now clear why I'm asking this question. I find the distinction between clitics/particles and affixes nebulous at best, and would love to get some clarification. Thanks in advance for your insights!

Answer 1

I'm not sure why you think that affixes cannot be stacked. Here's an example from English, whose morphology no one will argue has non-affixal properites:

anti- dis- establish -ment -ari -an -ism

We also need to separate case marking from case agreement. A given noun phrase is assigned one case; in languages like Latin that case percolates through the NP by agreement among its members, but that operation is logically separate from the assignment of case. (Your Basque example could be analyzed as a post-position, but a parallel one with an unambiguous case marker is easy to construct: aita eta amak "dad and mom-ERG"*.)

Clitics and affixes often share many features (phonological dependence on host/root, grammatical function), so it can be difficult to distinguish them from each other a priori. Furthermore, the development from (independent word ->) clitic -> affix is quite common historically, so at any given time there is no strong guarantee that a particular item in a particular language is not in the process of shifting from one category to another and thus displaying a mixture of properties. As a result of this fact, the categories "clitic" and "affix" have not been rigorously distinguished in the literature, leading to vagueness about what their defining properties are.

There is one kind of clitic that is unambiguously non-affixal: second-position (also sometimes called Wagernackel) clitics.
1. These are always placed after the first word/phrase (there can be variation between the two options) regardless of its identity.
2. Another diagnostic of pure phonological clitic-ness is syntactic movement: if a clitic's host is moved, does the clitic follow (affixlike behavior) or remain behind (phonological clitic-like behavior).

1.1 and 2.1 below are examples of not in English showing behaviours 1 and 2 respectively above. 2.1 is the non-affixing variant, which is more formal and indicates that it is the more conservative option:

(1.1) Didn't John kiss Mary?
(2.1) Did John not kiss Mary?

---

*The issue of distributing case over conjoined elements is interesting, given what we (don't) know about the syntax of conjunctions; it could provide a diagnostic of case vs. prepositionhood. It has been claimed that case cannot distribute over a conjunction whereas prepositions can. In the case of Basque, we should avoid calling things like the dative and (especially) ergative prepositions, so the status of this generalization is not clear.

Answer 2

The distinction is actually quite nebulous. Indeed, people are often led more by tradition to classify things as clitics/affixes than by actual linguistic argumentation. I think that many people go back to Zwicky's work on the topic (1977), though Spencer (2000) argues that better typologies of "clitic-hood" exist. Chapter 9 'Clitics' of Spencer's book 'Morphological Theory' discusses these issues a bit. I refer you to Spencer for some background.

The assumption, and the source of much of the clitic-related confusion, is that clitics will not trigger stem allomorphy or different phonological processes on the stem. (Indeed, the last poster was arguing against this interpretation.) Linguists have, I think, wrongly assumed that the syntactic status of clitics renders them unable to interact with the phonology or morphology. It is probably better to get outside of the "affix/clitic" dichotomy and think instead of syntactic dependence and conditioned allomorphy as independent properties. This requires really separating the syntactic behavior of elements from their morphophonology.

Answer 3

I think you've misunderstood the claim that "clitics don't trigger suppletion." It's perfectly consistent to have an unanalyzable clitic like -ak which spells out more than one feature. The usual claim is that clitics don't trigger suppletion of the lexical word that they attach to. So for instance, if the definite article -a is a clitic, then you shouldn't be able to have a word foo that unexpectedly becomes bar-a rather than foo-a when the definite article is attached.

I also don't think your epenthesis and elision data call for -a to be analyzed as an affix. There are two alternate possibilities here. First, you could posit a whole paradigm of clitics: -a for definite absolutive singular nouns, -ak for definite absolutive plural nouns, -ari for definite dative singular nouns, -ei for definite dative plural nouns, and so on. Essentially, these would be like the German determiners (which mark case, number and definiteness), except that they'd be clitics instead of independent words.

The second possibility would be to treat the case markers as separate morphemes (so that e.g. the dative clitics would be underlyingly -a-i and -e-i) and come up with some sort of phonological epenthesis and elision rules that generate the forms you actually see. Note that it's entirely possible to have a conditional epenthesis rule that, for instance, would insert an r between a and i but not between e and i.

I don't know enough about Basque to say which approach would work better. I'm really just commenting on the data you've presented in your post. But based on this data, I'd say the definite article was pretty clearly a clitic (or a whole paradigm of clitics), and your arguments for calling it an affix seem unconvincing.

[Gee, I hope I didn't just do your Syntax homework for you.]