In 2016, I gave a talk about “Coptic as a language without words” (which is available on YouTube; the handout is here), which was merely an illustration of a point that I had made in my earlier 2011 paper: That there is no general definition of ‘word’ that applies to all languages and that can be used in general linguistics (e.g. for universal claims about the nature of morphology, or for claims about “morphological typology”). The “word” notion is deeply entrenched in people’s consciousness because of our (Western) spelling habits, but it does not seem to be supported by evidence from morphosyntax of the world’s languages.
This general “word-skeptic” point can be illustrated nicely by Coptic, whose phonological unit “bound group” (Layton 2004; Haspelmath 2015b: 125) includes all kinds of elements that would not be regarded as words. An example from my presentation is the following:
This includes two bound groups (with two stresses), and the second element is a prepositional phrase (‘see’ takes a prepositional object in Coptic), but is the first element an inflected verb? The last three morphs (n-tn-nau [NEG-1PL-see] ‘we do not see’) look like an inflected verb, but this form is preceded by a relativizer, a plural definite article, and a preposition – these elements can hardly all be treated as verbal inflections.
So it is not clear how “words” should be defined in Coptic, and in fact, this notion has not been important traditionally in Coptic linguistics (and there are different conventions of dividing the scriptio continua into words by linguists). However, this conclusion is less radical than it might appear, because such difficulties arise in most languages – they are simply not so visible in European languages with well-established spelling systems.
Now there are three ways in which one could react to this situation:
(i) One could simply ignore my critical paper from 2011 and do business as usual.
(ii) One could draw the conclusion that we should try and avoid using terms such as “word” and “morphology” in general linguistics, because they are not well-established.
(iii) One could try and provide a definition of “word” after all, because we need retro-definitions of well-known terms that everyone uses anyway.
The first approach is of course taken by many linguists who are not aware of the paper or do not see its relevance; it is also conspicuously taken by Aikhenvald et al. (2020) in their paper “The essence of ‘word’”, where they talk about grammatical words as if their existence as a general phenomenon of grammar were self-evident and had not been challenged.
The second approach was the one that I took until fairly recently (which is reflected, for example, in my 2015a paper on clitics): I tried to avoid talking about “words” and “morphology”, and started using the words “form” and “expression” more often (which are often perfectly sufficient in general and comparative contexts).
But more recently I became interested in retro-definitions (the term first appears in my 2021a paper on terminological standardization). By this I mean definitions of traditional terms that most people use without having a particular definition in mind. Terms such as affix and morph (informally called “morpheme”) are terms of this sort – we use them all the time and assume that we understand each other, but we would be hard put to define them if we were asked to. Textbooks introduce them by exemplification but do not give definitions. Since these terms are used very widely, by tens (or hundreds?) of thousands of linguists and language teachers around the world, I felt that it was worth the effort to propose clear and relatively simple definitions. In my 2020 paper, I define a morph as a minimal form, and in my 2021b paper on bound forms and affixes, I define an affix as a non-promiscuous bound form that is not a root (as also discussed in this blogpost). The latter is not a particularly natural concept, and if one were to start linguistics from scratch, one would probably not have a term for such a concept. But we are sitting on a huge earlier tradition that comes with widely used terminology which will not go away, so it is best to have definitions of these terms that conform very largely to the earlier usage. This is the rationale for retro-definitions, and it applies to “word” and “morphology” as well.
So how could one define “word” in such a way that the definition largely corresponds to our intuitions? Note that I no longer require that it fully corresponds to our intuitions (let alone to the spelling conventions of particular languages) – but it should still correspond in the bulk of the cases (80% or more).
So here is a concrete proposal:
(1) A word is a free morph, or a nonroot bound morph that is not an affix, or a combination of a root or a compound with its affixes.
This seems straightforward enough, and the main reason I didn’t envisage this simple definition in 2011 was that I had no clear way of distinguishing between affixes and clitics (the most influential paper on the topic, Zwicky & Pullum (1983), doesn’t provide hard and fast criteria). Moreover, I had no good way of distinguishing between compounds and phrases. Now, in the meantime, I adopted the retro-defintion of an affix as a non-promiscuous bound form (as mentioned above), which gives a few counterintuitive results but avoids the numerous problems of phonologically-based definitions. This tells us right away that English Genitive ’s is not an affix (because it is promiscuous, as in the boy I met yesterday’s bag, where it doesn’t attach to a noun), and it tells us, for example, that German infinitival zu is a prefix (e.g. zu gehen ‘to go’), unlike English infinitival to: It always occurs on verbs, as German does not allow splitting infinitives (‘to boldly go’ translates as mutig zu gehen, not *zu mutig gehen).
But what is the difference between a compound and a phrase? In connection with my 2018 review of the polysynthesis handbook, I suggested a definition of “incorporation” that can be adjusted to give a general definition of compound:
(2) A compound is a combination of two roots which occur next to each other and which cannot be expanded by nominal or adjectival modifiers.
This definition thus excludes expressions such as French une tasse de porcelaine ‘a china cup’, because it can be expanded, e.g. une tasse de porcelaine française ‘a cup of French china’. For some reason, expansion by modifiers does not seem to be an important criterion in the literature on compounds (cf. Lieber & Štekauer 2009), but it appears to be the only one that can be applied to languages of all types.
Thus, I have now proposed a retro-definition of the term word, and with this definition in hand, we can go back to Coptic. Of course, Coptic does have words according to this definition, like all other languages. Let us consider this example again:
As I noted above, the last three morphs (n-tn-nau [NEG-1PL-see] ‘we do not see’) in the first morph sequence are an inflected verb, i.e. a word – but the bound elements preceding it are not words, but clitics similar to English Genitive ’s. The relativizer ete- can introduce not only verbal relative clauses such as n-tn-nau, but also nonverbal ones and then it does not combine with a verb. The plural definite article n- can precede not only nouns, but also relative clauses, so it, too, is promiscuos and hence a clitic, not a prefix. The same goes for the coordinator (or preposition) mn- ‘and, with’.
Coptic seems to have an unusual number of preposed clitics that contain only consonants and reduced vowels (or no vowels at all), but that’s all. It may appear unusual only because these proclitics are not written separately, so its orthographic words do not look like orthographic words in languages like Latin or Russian.
But the main point is: Despite what I thought a decade ago, it is sometimes possible to provide retro-definitions of traditional grammatical terms that are undefined and tend to be identified by test batteries (or worse, by cherry-picked criteria that suit the author’s purposes). Definitions are important because we do not know in advance whether the general terms we use correspond to something in the reality of human languages – so we need to be clear about which parts of our work are fixed (by terminological fiat) and which parts are our empirical discoveries. We have not made general discoveries about syntax and morphology being different – but we can still have a definition of ‘word’ that is independent of discoveries.
[P.S. In the original version of this blogpost, I had forgotten the second possibility: “a nonroot bound morph that is not an affix”. I was focusing on the distinction between phrase and word, not the one between affix and word which I had dealt with in my 2021 paper on bound forms and affixes. It’s hard to believe that the definition of “word” should be disjunctive and so complex…]
[P.S.2, November 2022: While writing up the definition for a short paper, I realized that there was a problem: The definition I provided in this informal blogpost fails to account for German verbs like geb-en, where the root geb- is not a word. In the written version, I had to make the definition more complicated by adding a notion of “required affix”; see the first draft version on Academia.edu here: https://www.academia.edu/91869514/Defining_the_word]
Aikhenvald, Alexandra Y & Dixon, R. M. W. & White, Nathan M. 2020. The essence of “word.” In Aikhenvald, Alexandra Y. & Dixon, R. M. W. & White, Nathan M. (eds.), Phonological word and grammatical word: A cross-linguistic typology, 1–24. Oxford: Oxford University Press.
Haspelmath, Martin. 2011. The indeterminacy of word segmentation and the nature of morphology and syntax. Folia Linguistica 45(1). 31–80. (doi:10.1515/flin-2017-1005)
Haspelmath, Martin. 2015a. 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. 2018. The last word on polysynthesis: A review article. Linguistic Typology 22(2). 307–326. (doi:10.1515/lingty-2018-0011)
Haspelmath, Martin. 2021a. Towards standardization of morphosyntactic terminology for general linguistics. In Alfieri, Luca & Arcodia, Giorgio Francesco & Ramat, Paolo (eds.), Linguistic categories, language description and linguistic typology, 35–57. Amsterdam: Benjamins. (DOI: 10.1075/tsl.132.02has, https://ling.auf.net/lingbuzz/005489)
Haspelmath, Martin. 2021b. Bound forms, welded forms, and affixes: Basic concepts for morphological comparison. Voprosy Jazykoznanija 2021(1). 7–28. (doi:10.31857/0373-658X.2021.1.7-28)
Layton, Bentley. 2004. A Coptic grammar: With a chrestomathy and glossary (Sahidic dialect). 2nd edition. Wiesbaden: Harrassowitz.
Lieber, Rochelle & Štekauer, Pavol. 2009. Introduction: Status and definition of compounding. The Oxford handbook of compounding, 3–18. Oxford: Oxford University Press. (10.1093/oxfordhb/9780199695720.013.0001)
Zwicky, Arnold M. & Pullum, Geoffrey K. 1983. Cliticization vs. inflection: English n’t. Language 59(3). 502–513.
[Disclaimer: I am a mere interested amateur]
“A compound is a combination of two roots which occur next to each other and which cannot be expanded by nominal or adjectival modifiers”
This seems to imply that in the Oti-Volta languages, there are _no_ compounds, which is much at variance with the way these languages have usually been described (though, as it happens, I actually agree with the conclusion …)
[Examples from Kusaal, but the principles are the same for at least the other Western Oti-Volta languages, and for Oti-Volta in general at least as far as adjectives are concerned.]
The Oti-Volta languages express number in nouns by paired singular/plural sets of flexional suffixes. They all show extensive compounding (to call it that pro tem) within NPs: it is, for example, the regular way of construing a noun with a modifying adjective: the noun precedes, as a stem form, followed by the adjective, which inflects to show the number of the head noun [in most Oti-Volta languages, it also agrees in noun class, but agreement has been lost in Kusaal, as in most of the Western Oti-Volta group]; thus
bʋsʋŋ “good goat”
bʋsʋma “good goats”
kugsʋŋ “good stone”
kugsʋma “good stones”
More remarkably, _demonstratives_ (and other dependent pronouns) behave like adjectives:
bʋkan “this goat”
bʋban “these goats”
kugkan “this stone”
kugban “these stones”
You can add demonstratives after noun-adjective compounds:
bʋsʋŋkan “this good goat”
bʋsʋmban “these good goats”
You can elicit forms with two adjectives (though they are rare in the wild)
bʋwɔkpiel “white long goat”
bʋwɔkpiela “white long goats”
More to the point, nouns can also be _pre_modified, for example by a mass noun expressing a substance:
salima bʋtiŋ “gold cup”
and this is compatible with adding adjectives and/or demonstratives:
salima bʋtiŋsʋŋ “good gold cup”
salima bʋtiŋkan “this gold cup”
pu’anya’aŋkʋdkaŋa “this old woman here”
Kusaal also has modifier-first compounds of the more familiar sort; the first element again appears as a stem:
lɔk “item of goods”
la’amaan “maker of goods”
Again, the first element can be premodified:
salima la’amaan “maker of gold items, goldsmith”
salima nɛ anzurifa la’amaan “gold- and silversmith”
In a nutshell, although there are pragmatic limitations on how far you can do this with individual “compounds”, in principle the first component of any kind of Kusaal compound, head-first or dependent-first, is susceptible of being modified.
The reason why I started by saying that I actually _agree_ with the conclusion, though, is that although every account I have ever read of Western Oti-Volta languages treats these sequences as compounds (as is also reflected in the word division in the standard Kusaal orthography I’ve used above), I think that what I have called “stem” forms above should in fact be regarded as “words.” They are unequivocally _bound_ forms (even the ones that look segmentally like singulars in fact differ systematically in tone and/or tone sandhi behaviour from singulars); but there are other “bare stem” words in Kusaal which are _not_ right-bound (notably, verb perfectives) and there are _no_ distinctive phonological features confined to noun “combining forms” alone.
There are other arguments for the wordhood of these combining forms, even though they remain a definitely distinct part of the noun paradigm from singulars and plurals, with no sign that they are becoming obsolescent; combining forms as heads are often remodelled after singulars segmentally (never tonally), and there are some actual errors (probably) in the current Bible translation like
saŋgbauŋ nɛ teŋgbauŋ paal “a new sky and new earth” Rev 21:1
beside the expected construction seen in
arezana paal nɛ dunia paal “a new heaven and new earth” Isaiah 65:17
[The separation of the combining form from the adjective paal “new” is normal (though essentially mistaken) in the standard orthography whenever the preceding combining form happens to look like a singular as written.] The significant thing about saŋgbauŋ “sky” is that it is one of the relatively few words where the combining form is the same as the singular _tonally_ as well as segmentally. So it is possible for a speaker to mistake a _spoken_ combining form for a singular
here, which is presumably why the translator (and others checking the text) didn’t notice that a bound combining form has got “stranded.” But the very possibility of such a mistake seems to suggest a feeling that combining forms are “words” rather than word fragments (as, of course, does the orthographic convention I just mentioned.)
Many thanks, that’s really interesting!
You state that
The last three morphs (n-tn-nau [NEG-1PL-see] ‘we do not see’) look like an inflected verb, but this form is preceded by a relativizer, a plural definite article, and a preposition – these elements can hardly all be treated as verbal inflections.
I’m wondering, is this meant to be interpreted as relativizer, definite article, and preposition *can’t* be verbal inflections? Or is it more of an intuition for why it makes sense that a rigorous definition of verbal inflection would end up excluding these?
Basically, if you had an (admittedly unlikely) situation where in some language, definite articles can only occur directly next to verbs (so they are non-promiscuous, but attach to verbs rather than nouns as expected) would you treat such things as verbal affixes? Or are they excluded outright from being verbal affixes?
Yes, you are right – I didn’t specify this in the blogpost (which would have become too long otherwise), but indeed: relativizers can be verbal affixes (then they are often called participial markers), just as articles can be nominal affixes (as in Swedish, where one type of article only occurs after nouns), and adposition-like elements can be affixes: then they are generally called case affixes. Next question: Can definite articles or adposition-like elements (flags) be restricted to verbs? This is not possible, I would say, because these two are *defined* as occurring with nouns. An adposition-like element that only occurs on verbs would be called a subordinator, not a flag. (I don’t know what an article-like element on a verb would be.)
Interesting article, as usual!
A question: you provide the following definition of compound:
(2) A compound is a combination of two roots which occur next to each other and which cannot be expanded by nominal or adjectival modifiers.
And you state:
This definition thus excludes expressions such as French une tasse de porcelaine ‘a china cup’, because it can be expanded, e.g. une tasse de porcelaine française ‘a cup of French china’
What is meant by “expansion” here?
Would the English expression “blackboard” not count as a compound, because “blackboard” can be modified by adjectives, as in “large blackboard”?
Alternatively, is the fact that in the French example, “française” modifies “porcelaine” rather than all of “une tasse de porcelaine”, while in my English example, “large” modifies “blackboard” as a whole, rather than either “black” or “board”, a significant difference? [This is probably what you meant by “expanded by nominal or adjectival modifiers”, now that I think about it, but I just wanted to make sure.]
Yes, in “large blackboard”, it’s the compound as a whole that is modified, not one of the roots. In “tasse de porcelaine française”, it’s the root “porcelaine” that’s modified (one could probably also modify “tasse”: “tasse française de porcelaine”). One difficulty here is that in some languages such as English, the modifying element in a compound can itself be a compound, e.g. “[flower pot] saucer” – here one might say that it’s not a compound because the first root can be modified. In my 2018 article, I said that it had to be modifiable by an adjective, but this only applies to compounds with a nominal nonhead. Here I was looking for a more general definition of “compound” that encompasses other types of compounds (e.g. V-V compounds).
I would like to point out that one complication, basically unsolvable in my view, is that at least in some languages one can find phonological domains (i.e. domains in which certain phonological processes such as harmony apply, while they do not apply outside of that domains) which are roughly the size of ‘words’, but often they do not seem identical to it. Similarly for prosody: some languages have ‘word stress’ which roughly means: one stress per ‘word’, where this word is roughly but not always precisely identical to morphosyntactic words.
(The kinds of things where phonological words and morphosyntactic words are not exactly the same tend to be phonological clitics (obviously), and certain types of morphology, such as derivation and compounding.)
Yes, that’s a “complication” that I leave out of consideration, because there is already a broad consensus that (1) word-like phonological domains are often somewhat different from grammatical domains; (2) that the domains may be different for different phonological regularities of the same language; and (3) that “phonological wordhood” cannot be constitutive of the syntax-morphology distinction. But of course, it’s an extremely interesting question to what extent phonological domains tend to correlate with morphosyntactic patterns – and while there’s a lot of anecdotal evidence from individual languages, it seems that little is known (from broad surveys) about general patterns. Or is this a wrong impression? Where should I look?
I’m surprised not to see Damaris Nübling’s work (since 1992 or so) mentioned. Precise definitions for affixes and several distinguishable kinds of clitics are proposed there. The general conclusions seem to be the same as yours.
Oops, I see you’ve cited her 2005 book chapter in your 2015a paper. Her 1992 book is on Google Books (one of those annoyingly truncated previews, but there’s enough there to work with).
Thank you for pointing out the relevance of this old book. I know about that work by Nübling, but I hadn’t looked at it in recent years. This is very helpful.