A few weeks ago, a new paper of mine on bound forms and affixes was published by the journal Voprosy Jazykoznanija. This has long been the most important Russian linguistics journal, and it now publishes articles in English as well. Here’s an introduction to some of the key points of the paper (Haspelmath 2021a).
Since my 2011 article on on the problems with a general “word” concept, I have been aware of the great difficulties with defining the widely used term affix in such a way that the criteria are the same for all languages. But this is required of a general (= comparative) concept in linguistics: If we apply different criteria in different languages, we do not use the term in the same way, and we end up comparing apples with oranges. Affixes are generally thought to be different from clitics, but these have the same problem: There are no clear and widely accepted criteria for identifying clitics and distinguishing them from affixes (Haspelmath 2015).
In the 2011 paper, I realized that “test batteries” as they are well-known from the famous article by Zwicky & Pullum (1983) are not the solution – if we find enough “arguments” for a particular classificatory decision, we may persuade our readers, but the method is not rigorous (and often subjective). Different criteria may give different results, and then the temptation is great to cherry-pick those that fit our preconceptions.
So can we do linguistics without a general “affix” concept? Actually, I do think that this would be possible (for example, my efficiency theory of asymmetric coding in grammar does not need it, see Haspelmath 2021b). But the term affix is deeply ingrained, and linguists will use it anyway, so we want to have a clear definition for it.
In earlier work, I had made use of the Bloomfieldian term “bound form” (especially in Haspelmath 2013; see also my 2019 paper on flags as bound markers of semantic roles), which I found was a good concept. A bound form is simply a form that does not occur on its own – this is a single criterion that is in principle applicable in all languages. All languages have some forms, such as the English article the or the Mandarin perfective marker le, which MUST occur together with some other form. Note that boundness is not a phonological criterion – even though many bound forms are phonologically deficient (e.g. English the, with its schwa), there are many others that are not deficient. Thus, English -ing is not deficient (cf. ring, sing, …), but it cannot occur in isolation. It is an affix not because of any phonological properties, but because it is bound, and it occurs only on verbs.
Many linguists seem to have the intuition that affixes are somehow “phonologically attached” to their base or “fused” with it, but these metaphors have never been made precise. I have long suspected that the reason why these vague notions are compelling and have rarely been questioned is the influence of our word-based writing systems: If we didn’t write some elements together and some separately (as in Chinese and Thai), the metaphor of “attachment” wouldn’t occur to us. Chinese linguists have traditionally distinguished between xūcí 虚词 (empty words) and shící 实词 (full words), but have not said anything about “attachment”. Another factor may have been the heterogeneity of nominal and verbal affixes in Latin: Traditionally, Western linguists started out by learning different “declension” and “conjugation” paradigms, where the affixes were not separated by the roots and seemed inextricably linked to them. The notion of a morph is surprisingly young (it did not catch on until the 1920s, under the label morpheme), and there is still much confusion around it among linguists. (Many are still arguing that we should describe “morphology” in a word-based way, even though we cannot readily distinguish words from phrases.)
But of course, phonological alternations frequently happen at morph boundaries, and we want to understand what constrains them. For example, are the kinds of bound morphs that we normally call “affixes” more likely to exhibit such alternations than other kinds of bound morphs? Recall that the English article the is a non-affixal bound morph, and it alternates (th[ə] tree vs. th[i] apple) – while the English affix -ing does not. Phonological alternation is thus orthogonal to affixhood. Another nice example of this is the contrast between the English genitive marker ’s (which is widely thought to be a clitic, e.g. Anderson 2008), and the German genitive suffix -s. The latter is phonologically invariant (e.g. Kants Theorie ‘Kant’s theory’), but the English genitive marker is variable: the queen of Denmark’s husband vs. the king of France’s wife (-[s] vs. -[əz]).
This has led me to propose a new term for a bound morph that shows phonological interactions with its host: welded form (also discussed in an earlier blogpost). This new term drives home the point that bound forms need not be “phonologically attached” (and that a notion of “phonological boundness” makes no sense – we should talk about [phonological] weldedness).
After all these prelminaries, here’s the definition of affix that I propose – first the short version:
An affix is a non-promiscuous bound form that is not a root.
Roots can be bound forms (for example, the German verb root helf- ‘help’ must occur with some inflection: helf-e ‘I help’, helf-en ‘to help’, etc.), so clearly affixes must be defined as non-roots. But the relatively new part here is that affixes are different from non-affixes in that they are non-promiscuous – in other words, they only occur on roots of a single class. This makes English -ing an affix (it only occurs on verb roots), and it makes German genitive -s an affix (it only occurs on noun roots). However, English genitive ’s is not an affix because it occurs on verbs as well (Anderson gives the example: Every man I know’s taste in wallpaper is appalling, where ’s occurs on a verb).
As the term “promiscuous” is not so widely known (outside of morphology) and sounds a bit odd, the real definition of “affix” is as follows:
An affix is a bound morph that is not a root, that must occur on a root, and that cannot occur on roots of different root classes.
In the paper, I discuss a number of limitations of this definition – in particular, the fact that elements which occur both on nouns and adjectives (e.g. Spanish libro-s nuevo-s ‘new books’) are not included in the definition, although intuitively, some of them are affixes. Clearly, the definition does not correspond fully to everyone’s intuitions. But if it corresponds to our intuitions 80% of the time, I would say that it’s worth it – after all, nobody thinks that our intuitions must necessarily correspond to a coherent scientific concept.
By contrast, it is clear that our scientific terms should have clear definitions, so if we want to use affix as a scientific term (something that linguists often seem to be doing, or trying to do), we need a clear definition. If we have to sacrifice either intuitiveness or clarity, the choice should be clear.
Anderson, Stephen R. 2008. The English “group genitive” is a special clitic. English Linguistics 25(1). 1–20. (doi:10.9793/elsj1984.25.1)
Haspelmath, Martin. 2011. The indeterminacy of word segmentation and the nature of morphology and syntax. Folia Linguistica 45(1). 31–80.
Haspelmath, Martin. 2013. Argument indexing: A conceptual framework for the syntax of bound person forms. In Bakker, Dik & Haspelmath, Martin (eds.), Languages across boundaries: Studies in memory of Anna Siewierska, 197–226. Berlin: De Gruyter Mouton.
Haspelmath, Martin. 2015. Defining vs. diagnosing linguistic categories: A case study of clitic phenomena. In Błaszczak, Joanna & Klimek-Jankowska, Dorota & Migdalski, Krzysztof (eds.), How categorical are categories? New approaches to the old questions of noun, verb, and adjective, 273–304. Berlin: De Gruyter Mouton.
Haspelmath, Martin. 2019. Indexing and flagging, and head and dependent marking. Te Reo 62(1). 93–115. (doi:10.17617/2.3168042)
Haspelmath, Martin. 2021a. Bound forms, welded forms, and affixes: Basic concepts for morphological comparison. Voprosy Jazykoznanija 2021(1). 7–28.
Haspelmath, Martin. 2021b. Explaining grammatical coding asymmetries: Form-frequency correspondences and predictability. Journal of Linguistics. (doi:10.1017/S0022226720000535)
Zwicky, Arnold M. & Pullum, Geoffrey K. 1983. Cliticization vs. inflection: English n’t. Language 59(3). 502–513.
I’m not a linguist, but only an interested and mildly-educated observer. But I wonder about your view that English genitive -s is promiscuous because it occurs on verbs (as in “every man I know’s taste”) as well as nouns. Intuitively I would have said that the -s in “every man I know’s taste” occurs not on the verb “know” but on the noun phrase (which includes the dependent clause “I know,” modifying “man” ) “every man I know.” Might one not define the domain of English genitive -s as “noun or noun-phrase” and thus render it non-promiscuous? The role of “Every man I know” in the larger sentence is after all exactly analogous to the role of “Every man” in the sentence “Every man’s taste in wallpaper is appalling,” just as it plays the same role in the sentence “Every man I know has terrible taste” as “Every man” plays in the sentence “Every man has terrible taste.” So the noun / noun-phrase distinction doesn’t seem functionally crucial. I’m also not sure how one distinguishes a noun from a noun phrase without a word concept. (I’m agnostic on the necessity of a word concept itself; if we give up on “word,” I’d happily call “Every man I know” a noun instead of a noun phrase, so as to account for its noun-like roles in sentences.)
Thanks for this comment – and indeed, many linguists have said that the English possessive ‘s occurs on the noun phrase. The reason I do not do this is that in general, we have no good way of identifying “phrases”. Noun phrases are normally easy to identify, but most other phrases (e.g. “verb phrase”, and most others) are not well-defined. So I make use only of the concept of “root”, which is unproblematic.
I see! Thanks for answering.
”in other words, they only occur on roots of a single class. This makes English -ing an affix (it only occurs on verb roots), and it makes German genitive -s an affix (it only occurs on noun roots)”
It’s somewhat difficult to understand how you distiguish ‘verb roots’ from ‘noun roots’ without making use of the concept ‘word’ which you reject. Am I missing something?
Is the root of, say, German opferte a ‘verb root’ or a ‘noun root’? What is the root of German pausieren?
I’d like to say that action-denoting roots are verb roots and object-denoting roots are noun roots. In the case of “opfer(-te)”, the root is a verb root (and also a noun root; or maybe a noun-verb root). In the case of “paus(-ieren)”, it’s neither because “Pause” does not denote an object or an action. – You may find this answer unsatisfying, because the comparative concepts are not exhaustive, but this is thbe nature of synchronic comparison: Only certain aspects of languages can be compared in a rigorous way.
So I am wondering how the new definition applies to Modern Greek, and also to Icelandic, where there are a considerable range of traditional ‘affixes’ that appear on both traditional ‘nouns’ and ‘adjectives’ (promiscuous) but also subject to semantically arbitrary declension class features (non-promiscuous). Putting nouns and adjectives into one class might be viable in Greek, but not in Icelandic (for the same kinds of reasons that apply in German, such as strong vs weak forms).
Or perhaps the definition of ‘promiscuous’ needs to be a bit more elaborate, considering both syntactically significant word classes and arbitrary inflectional classes. Spanish plural -s would remain a non-affix, but it also has some wierd properties iirc, such as ‘se les di’ being able to mean ‘I gave them to him’ in some varieties of Mexican Spanish (I hope I’m remembering this right), suggesting -s as more of a free-floating plural marker (non-projecting word in the LFG framework?)
Yes, there are quite a few intermediate cases. Indeed, in Greek we would basically end up with non-affixes (as in Spanish), while Icelandic (like German) would have case *affixes*. This is unintuitive, but I’m no longer worried by this, because I decided that it’s better to have a clear definition that sometimes gives unintuitive results than not to have a definition. (Our intuitions are typically based on stereotypes, not necessarily on coherent concepts.)
I just reread your your 2021 paper, “Bound forms, welded forms, and affixes:
Basic concepts for morphological comparison”, and it was really interesting! However, I noticed that in this paper, you refer readers to your 2020 paper, “The morph as a minimal form” for the definition of the terms “morph” and “form”.
I just wanted to point out that there’s a bit of ambiguity in that paper regarding the definition of “form”.
In that paper, you state that “The term morph should be defined as in (1), as a minimal linguistic form (i.e. a minimal pairing of syntacticosemantic content and a string of phonological segments)”
This seems to imply that the meaning of the term “form” is “a pairing of syntacticosemantic content and a string of phonological segments”.
However, later you state:
A form is a recurrent pairing of content and segmental shape: “Every language consists of a number of signals, linguistic forms. Each linguistic form is a fixed combination of signaling-units, the phonemes …We assume that each linguistic form has a constant and definite meaning” (Bloomfield 1933: 158)
This paragraph seems to provide two definitions for the term “morph”, which are not clearly equivalent.
One possible definition mentioned here is “a recurrent pairing of content and segmental shape”.
The second possible definition is “a fixed combination of signaling-units, the phonemes, which has a constant and definite meaning”.
These are not clearly equivalent, because the first refers to “segmental shape”, while the second refers to “phonemes”. I am not sure how Bloomfield defines the term “signaling-units”, or the term “phoneme” (here it seems that they are equivalent). However, I do know that in some approaches to phonology, a “phoneme” is an equivalence class of segments. In such approaches to phonology, a definition based on “segmental shape” cannot be equivalent to a definition based on “phonemes”.
I hope that this post doesn’t seem too negative. I’m really interested in your important work on providing clear definitions for these important linguistic concepts, and I just wanted to point out this small area for potential ambiguity so your definitions could be even better! 🙂
Thanks! Well, I agree with Bloomfield on a surprising number of points, but not on his use of “phoneme” – I generally prefer “segment” (also because it is modality-neutral, I hope). Bloomfield had the hope that this classificatory notions would turn out to be universally applicable, and that class terms like “phoneme” would turn out to be definable in a consistent way. However, I think that phonemes are language-particular, just like syntactic roles (“grammatical relations”) and word classes. Dryer (1997) has a good discussion of this.
Thanks for your reply!
I’m still confused as to what the actual definition of the term “linguistic form” is. I think it’s “a pairing of syntacticosemantic content and a string of phonological segments”? But just wanted to make sure. 🙂
Yes, I think so. However, I have not thought about it deeply. Maybe “form” will have to be a “semantic primitive”. It’s impossible to define all terms – for some basic ones, we have to rely on some kind of previous understanding.
Personally, defining “linguistic form” as “a pairing of synacticosemantic content and a string of phonological segments” makes sense to me.
Phonological segments are already defined in phonology, while “syntactico-semantic content” seems like a true primitive concept. 🙂
This is an interesting post! I’m glad you’re still working on providing clear definitions for widely used terms like “affix”, as this seems to be a prerequisite for linguists to understand each other’s work.
I know that you’ve been working on terminological issues for quite some time now, so your definitions are scattered across your work. I was wondering: do you have a document where you can see all of these definitions in one place? Such a document could be very convenient for people who are interested in using your proposed definitions.
Relatedly, I was wondering: do you have any ideas for how to get linguists to actually start using standardized definitions? In an earlier post of yours, https://dlc.hypotheses.org/989 you mentioned the idea of setting up a nomenclature committee for linguistics for this purpose, or organizing sessions at big linguistics conferences to discuss terminology, both of which seem like very reasonable suggestion to me. Do you know if there’s been any progress towards setting up such committees or conferences since that post?
As usual, thanks so much for reading! 🙂
These are good questions. The first one: Yes, I’m working on such a list of definitions – this will be a website called “Grammaticon” (see https://osf.io/3yck8/). The second one is a more distant dream, but I talked about this briefly in the last section of this paper: https://ling.auf.net/lingbuzz/005489
I’m surprised not to see a mention of the work of Damaris Nübling, perhaps in particular this book which presents different kinds of clitics and their distinction from affixes (with various sorts of borderline cases) in considerable detail.