In my (2011) paper on the indeterminacy of word segmentation, I despaired of the task of distinguishing between “words” and “affixes”, but in the meantime, I have become more optimistic. I am now ready to answer the above question (“What is an affix?”) by a proposal for a definition.
It should be noted that in general, when a linguist asks “What is an X?” (where X is a general concept), this means (or should mean) “what is the best general definition of the term X?”. This is because the notion of X cannot be presupposed to exist in the world, and all we know is that there is a well-known term that we want to continue to use and whose definition is unclear.
In 2011, I was resigned to giving up the terms “word” and “affix” (and in 2015, the term “clitic”), but of course, linguists continue to use these terms. So I kept wondering whether one can find a rigorous definition that corresponds to at least 80% of their usage. This matters, because many general claims depend on the notion of affix, and we want to be able to test such claims and develop them further. Two striking examples are the claim that there is a suffixing preference (e.g. Himmelmann 2014, also discussed here) and that “grammatical complexity” is correlated with social factors (Lupyan & Dale 2010).
A year ago, I started working on a review of the “Oxford Handbook of Polysynthesis” (published as Haspelmath 2018a), and while the concept of “poly-synthesis” was poorly defined in the volume, I wondered again whether at least the simpler concept of “synthesis” could be given a clear definition. The latter seems even more important, also because there are many claims about regular developments from synthetic to analytic and back again (dubbed the “anasynthetic cycle” in my 2018b paper).
My current answer (contrasting with my agnostic attitude in 2011) is yes – the term affix can be defined as follows:
Affix: An affix is a non-promiscuous bound form which is not a root.
It is not possible to characterize affixes semantically, but roots can be defined as “morphs that denote things, actions or properties” (Haspelmath 2012), and by saying that affixes are non-roots, we come fairly close to capturing what they can do semantically (basically, everything else). One may object that some affixes denote people (= a subtype of thing), and some denote actions (e.g. causative affixes), and diminutive affixes may denote the property ‘small’ – but I said earlier that we should be content with 80% overlap with traditional usage.
An affix is also characterized as a bound form, because free forms (= forms that can occur in isolation) would never be considered affixes. Upon first hearing hi Dad!, the analyst might suspect that hi- is a vocative prefix, but once it is recognized that hi! can occur on its own, such a description becomes impossible.
Now crucially, an affix differs from a clitic in that it is non-promiscuous, meaning that it always occurs on roots of the same root-class – either always on noun roots, or always on verb roots, or always on adjective roots (these are the three root-classes, by definition). It is true that traditionally, the term “clitic” is thought of as being somehow phonologically special (needing a host to “lean on” to be pronounced), but linguists have not been able to give a general phonological characterization of this notion (as is made quite clear in the concluding chapter of Spencer & Luís 2012). Traditionally, affixes have often been characterized as somehow phonologically “more attached” than clitics (Zwicky & Pullum 1983), but nobody has been able to give clear criteria for this that can be applied to all languages. (There is a widespread feeling that languages have “phonological words”, but if these are identified differently in different languages, there is no rigorous way of testing such claims.)
The traditional use of the term “affix” is heavily influenced by linguists’ experience with Indo-European (and Semitic) languages, and in these languages, elements such as tense-aspect markers, plural markers and comparative markers are generally restricted to verbs, nouns, and adjectives respectively – and I think that this is what makes us call them affixes. If a bound aspect marker is not restricted to occurring on verbs (like the Mandarin perfective marker le, which may occur after the verb, or after the object, etc.), it will not be called an affix. Likewise, if an object person index always occurs on the verb (as in Arabic and Hebrew, e.g. עֲזַבְתָּ֑נִי ʕazab-ta-ni [forsake-2SG.SBJ-1SG.OBJ] ‘you have forsaken me’), then it is regarded as an object person suffix. By contrast, Portuguese object person indexes sometimes occur on the verb (vejo-a ‘I see her’) and sometimes on an element preceding the verb (que=a vejo), so they are promiscuous, and hence they are clitics.
This also gives the right results for articles (“right” in the sense that they tend to agree with what people have always said). The English article a(n) has two different phonological shapes (a and an) depending on the context, so by a phonological characterization, it looks more like an affix. But it need not occur on a noun (as in an apple) – it may also occur on an element preceding the noun which is not a noun (a green apple). By contrast, the Swedish definite article -en (e.g. kung-en ‘the king’) always occurs on a noun, and is thus regarded (and written) as a suffix. Bound prepositions in European languages may occur directly on the noun (French à pied ‘on foot’), but they may also occur on other pre-nominal elements, and thus they are not prefixes. By contrast, Egyptian Arabic prepositions always occur on nouns – because there are no prenominal modifiers – and may thus be regarded as prefixes (but in Standard Arabic, demonstratives precede the noun, so that the prepositions are not always adjacent to the noun).
I should add that the definition involves a notion of “occurrence of a bound form on another element” (its host) that I have not characterized clearly yet. A bound form cannot occur on its own, and in most cases, it is immediately clear what the host of a bound form is. The host of the English Genitive clitic ’s is the immediately preceding element (a noun or some other element in a postnominal modifier), and the host of the Portuguese clitic =a is the verb or some preverbal element. However, some bound forms do not have a very clear host. German has a set of discourse particles (e.g. wohl, ja, doch) that cannot occur on their own and cannot occur clause-initially, but it is not clear that their host is either the element preceding them or the element following them (e.g. sie hat es ja gesehen [she has it PCL seen] ‘she saw it (as you know)’ – here it is unclear whether the particle ja “occurs on” es or on gesehen – it seems that it simply needs to be in a particular clause-medial position). However, it is not actually crucial for the definition of affix that we identify the host of any bound form – since affixes always occur on roots of the same class, there is never any question which is their host.
So it seems that the new definition of affix (as a non-promiscuous bound form which is not a root) works quite well, and I again acknowledge the contribution of Harald Hammarström in leading my thinking in this direction. But how happy are we with such a definition? Is there anything deep about affixes?
Of course not – but that was not the point. I still do not think that such a definition gives a justification for a deep division between a morphological and a syntactic component, and the definition does not get us any closer to understanding why languages have the kinds of forms and rules that they have. All it could do is give us a better understanding of what a colleague means when she talks about “affixes”, assuming that linguists use the term in the same way. It is thus a contribution to a possible terminological standardization – not more than this, but also not less (because without standard terminology such as the IPA, we will keep talking past each other even at the lower levels).
“An affix is also characterized as a bound form, because free forms (= forms that can occur in isolation) would never be considered affixes. Upon first hearing hi Dad!, the analyst might suspect that hi- is a vocative prefix, but once it is recognized that hi! can occur on its own, such a description becomes impossible.”
OK, so let’s take an example from a fairly known language, namely Croatian/Serbian, where there’s a whole set of words (in the traditional understanding of “word”) which contain ni-: nikad “never”, nitko “nobody”, nigdje “nowhere”. These words are connected to kad “when”, tko “who”, gdje “where”. So nikad “never” = ni + kad “when”.
Is ni- an affix? There morpheme ni can appear before many other words, expressions, including nouns, verbs, prepositional phrases (and it’s spelled separately then). But it cannot appear completely alone, i.e. there’s no question where an answer can be just “ni”.
Even more interesting, some combinations of prepositions and words above show rearrangement of morphemes:
od “from” + nikog “nobody (genitive)” = ni od kog (spelled as three separate words). Of course this is just a spelling, but such rearrangement of morphemes show ni is not strongly bound to the morpheme after it…
So, is it an affix or not? What is it, if not an affix?
It’s not an affix (despite the spelling) because it does not always occur on elements of the same class.
Chris Lucas commented on Facebook:
“Is the postnominal position of demonstratives in Egyptian Arabic really sufficient to make prepositions in that variety affixes? Like standard Arabic, Egyptian has a definite article and quantifers which can intervene between prepositions and nouns.
… The definite article in Arabic is an interesting case. Does it count as promiscuous or not, given that it is also prefixed to adjectives and postnominal numerals when these agree with a definite noun? I.e. in these cases it seems to be an agreement marker rather than the locus of expression of definiteness, as in your English and Swedish examples.”
My answer: Yes, the definite article in Arabic is a case whose treatment according to my definition is not immediately obvious. Since it can occur not only on nouns, but also on adjectives, that makes it promiscuous, and hence not an affix. If it is not a prefix, then prepositions cannot be prefixes either, because prefixes must occur on roots (= minimally separated from them by other prefixes). However, if one thinks that the article on adjectives is a different element, then the non article would become a prefix. In general, one needs to know whether two elements are “the same”, and I don’t have a general answer to this. I realize that this problem makes the criterion hard to apply in many cases, but at least it seems to be an independent problem that needs to be solved somehow anyway.
David Gil comments as follows (on Facebook):
“I continue to welcome your engagement with these issues, in the face of other people whose inclination is just to stop worrying and “get on with it” …
In that spirit, though, I still have a problem with your definition, or actually with your prior definition of root as “morphs that denote things, actions or properties”. You acknowledge that this definition only works in “80%” of the cases, and I would suggest that some of the members of the residue (especially person markers on verbs) actually denote “important”, or “non-peripheral” cases, of the kind that we can’t be happy about ignoring.
But my objection here is a more general one. We’ve had this conversation before, but we still disagree on whether semantics constitutes a firmer foundation for etic comparisons (as you believe) or actually a shakier one (as I am inclined to believe). I simply have no confidence in my ability to determine whether something is a thing, an action, or a property. To take just one example: I have asked a classroom of students whether “destruction” denotes a thing or an activity — and the verdict was split: some were sure it was a thing, others equally sure it was an activity. One can construct obvious “accounts” for each of these two competing judgements, but the fact remains that we do not have access to a clear apriori distinction between semantic categories such as these upon which we can confidently rest potential definitions of comparative concepts.”
My answer is: Yes, there are many cases where it’s not really clear whether some expression denotes a thing or an action or a property. The example of “destruction” is not a typical one, because here, if there is a root, it would be “destruc-“, which is even harder to have intuitions about. But even nouns like “wave” are not clearly things, and verbs like “know” are not clearly actions (aren’t they more like properties?). But the point is: There are (in all languages) enough roots like “tree”, “pen”, “moon”, “shoe”, “bird” which are very clearly noun roots, enough roots like “run”, “take” and “help” which are very clearly verb roots, and enough roots like “old”, “small”, “good”, “white” which are very clearly adjective roots – and it’s because of their MEANING that we call the corresponding form classes in all (or most) languages nouns, verbs and adjectives. And my very strong intuition is that when we call a morph a “root”, the clearest cases are verb roots, noun roots, and adjective roots – in fact, all other possible roots are questionable (sure, one could say that there is a “quantifier root” “all” in English, but in such contexts, we generally don’t use the term). So I think that the definition works, and I don’t know any alternative.
Martin, I am wondering about your statement that “affixes always occur on roots of the same class”. To start with, I would like to note in connection with your Swedish example that a number of Scandinavian non-standard varieties allow for definite articles to be suffixed to adjectives, when these are used on their own. I am not mentioning this as a counterexample, however, since it may perhaps be accounted for by saying that in those contexts, the adjective has the function of a noun. But there are other cross-linguistically frequent cases which I think are problematic. It is common for agreement markers to have the same shape as markers on the word they agree with or as agreement markers from other word classes. In a Russian example such as *on-a byl-a bol’n-a* ‘she was sick’, the suffix *-a* marks feminine gender on a pronoun, a copula verb, and an adjective. Like in the case of the Arabic definite article, one could of course suggest that these are three different suffixes, but since the phenomenon appears to be pretty general, it is in need of a general solution, and I am not sure that this can be done without ending up in circularity. Two more points: one is that a distinction should be made between the ability to occur on different word classes and the ability to move around to different hosts in a sentence. The latter is a pretty sure indication of a clitic. But even clitics may be a bit picky in their choice of hosts. The genitive -s clitics in English and Scandinavian are less likely to attach to, say, an adverb than to a noun. The other point is that it would appear that the question of whether two elements are “the same” belongs to the description of individual languages, and thus any generalization that depends on that question would involve language-particular considerations – which may cause problems for the demarcation line between language description and language comparison that you have argued for. (I think we may have discussed this issue before.)
Yes, Östen, these are all very good points, and they weaken my claim that my definition has come rather close to the traditional use of the term “affix”. If I have to say that the Russian element -a in your example is not a suffix, that’s very much against the traditional way of talking. But you offer no alternative, so the only conclusion can be that we don’t know what an “affix” is after all… (Incidentally, the “moving around” criterion for clitics does not work for second-position elements – they are always in “the same position”, but they are not normally regarded as affixes.)