One important concept in all discussions of nonverbal clause constructions (or “non-verbal predication”) is the term “copula” – but what exactly is a copula? Here I continue the discussion of Creissels et al. (2016) (following an earlier blogpost), who distinguish between four strategies (“formal types”) of “predicative marking” in such constructions (p. 15):
The contrast between a copula construction and a juxtaposition construction is perhaps obvious, and Creissels et al. (2026) do not provide a definition of copula. However, it seems that by and large they use the same definition of the term as I do in my 2025 paper:
(1) copula: A copula is a form that indicates a stative link between the two argument positions of an equational, ascriptive or locational clause. (Haspelmath 2025: §7)
If a copula is a kind of form, it cannot be zero, so “zero copulas” are not possible (and Creissels and al. seem to agree, as they talk about “juxtaposition” rather than “zero copulas”). Some typologists have used the notion of “zero copula” (e.g. Stassen 2005), but for comparative concepts, typologists need concrete concepts involving forms (rather than zeroes or transformations, as I discussed earlier).
Now what about special forms such as Spanish hay ‘there is’ in existential clauses (which I call existives)? Are they copulas, or are they something else? Consider an example such as (2):
(2) Hay un gato en el árbol. ‘There is a cat in the tree.’ (Creissels 2019: 42)
In Creissels (2019), such elements were called “inverse-locational predicators” (because Creissels does not want to call such sentences “existential”, as I discussed earlier on this blog). However, it was not clear in that paper what a “predicator” was, and in the new paper, Creissels et al. (2026: 40) say:
“As a rule, the essential elements of existential clauses are a predicator (in the sense of role-assigning element) and an NP expressing the role of existent…”
The remark in parentheses makes it clear that a copula is not an “predicator” in the 2026 article, because the reason that clauses with verbal copulas count as “non-verbal predication” is that they do not assign semantic roles. (On p. 16, Creissels et al say: “Copulae need not necessarily be viewed as “semantically void”. What really characterizes them (and distinguishes them from non-copular verbs) is that they do not contribute to the argument structure of the clause.”)
Since existives such as Spanish hay do not assign semantic roles, they are not predicators, and hence a subtype of copula. (Creissels et al. 2026 do not talk about “inverse-locational predicators” anymore, and they refer to existives as “synchronically unanalyzable elements not found in the corresponding plain-locational clauses”, p. 39. This description is fairly long, so it’s useful to have the new term existive.)
So there is quite a bit of agreement between Haspelmath (2025) and Creissels et al. (2026), but I did not adopt their notion of “predicative inflection” (originally proposed in Bertinetto et al. 2019) in my paper. One example of this comes from Erzya Mordva, where predicative person markers can “attach” to nominal, adjectival, locative and verbal predicates, as seen in the following table from Turunen (2011: 149).
These bound person forms are not “inflection”, I would say, because they are not affixes, and inflectional markers are affixes by definition (Haspelmath 2024). They are not affixes because they occur on roots of different kinds: verbs, nouns, adjectives (as well as on adverbial expressions, including presumably adverbs). If a bound form that is not a root is not selective, we call it a clitic (Haspelmath 2023). These forms are thus person clitics that occur juxtaposed to the predicate word.
(One could imagine person forms that only occur on nouns, or only on adjectives, and that signal that the noun or the adjective occurs predicatively. I do not know of such cases, but I would still say that they are “juxtaposed”, because I see no reason to restrict the term “juxtaposition” to the combination of “words” – the “word” notion does not seem to have a privileged status in morphosyntax.)
Bertinetto et al. (2019) mention another type of “predicative inflection” (called “type IIIb” in Creissels et al. 2026), exemplified by Akkadian. In this language, nouns lack a case suffix when occurring predicatively. Thus, šarrāq-um is the word for ‘thief’ (with a nominative “status rectus” suffix), but in predicative function, only the stem šarrāq is used (without a copula):
This is indeed a very special construction, but the type seems to be very rare, so I did not include it in my 2025 paper.
By and large, however, there is substantial agreement between my paper and Creissels et al. (2026). One particularly interesting aspect of their paper is the set of observations about the occurrence of juxtaposition (p. 19):
It seems that in general, juxtaposition is preferred when the construction type is more frequent, in line with the frequency theory of asymmetric coding (Haspelmath 2021). Creissels et al. do not emphasize this (and they do not provide evidence for their claims, beyond citing Stassen 1997), but I find their observations very pertinent and valuable.
References
Bertinetto, Pier Marco & Ciucci, Luca & Creissels, Denis (eds.). 2026. Non-verbal predication in the world’s languages: A typological survey (Volumes 1-2). Berlin: De Gruyter Mouton. (https://www.degruyterbrill.com/document/doi/10.1515/9783110730982-001/html)
Bertinetto, Pier Marco & Ciucci, Luca & Farina, Margherita. 2019. Two types of morphologically expressed non-verbal predication. Studies in Language. John Benjamins 43(1). 120–194. (doi:10.1075/sl.17013.ber)
Creissels, Denis. 2019. Inverse-locational predication in typological perspective. Italian Journal of Linguistics 31(2). 37–106.
Creissels, Denis & Bertinetto, Pier Marco & Ciucci, Luca. 2026. Non-verbal predication: An analytical framework. In Bertinetto, Pier Marco & Ciucci, Luca & Creissels, Denis (eds.), Non-verbal predication in the world’s languages: A typological survey (Volume 1), 1–56. Berlin: De Gruyter Mouton. (https://hal.science/hal-05282984v1)
Haspelmath, Martin. 2021. Explaining grammatical coding asymmetries: Form-frequency correspondences and predictability. Journal of Linguistics 57(3). 605–633. (doi:10.1017/S0022226720000535)
Haspelmath, Martin. 2023. Types of clitics in the world’s languages. Linguistic Typology at the Crossroads 3(2). 1–59. (doi:10.6092/issn.2785-0943/16057)
Haspelmath, Martin. 2024. Inflection and derivation as traditional comparative concepts. Linguistics 62(1). 43–77. (doi:10.1515/ling-2022-0086)
Haspelmath, Martin. 2025. Nonverbal clause constructions. Language and Linguistics Compass 19(2). e70007. (doi:10.1111/lnc3.70007)
Stassen, Leon. 1997. Intransitive predication. Oxford: Oxford University Press.
Stassen, Leon. 2005. Zero copula for predicate nominals. In Haspelmath, Martin & Dryer, Matthew S. & Gil, David & Comrie, Bernard (eds.), The world atlas of language structure. Oxford: Oxford University Press. ((http://wals.info/chapter/120))
Turunen, Rigina. 2011. Parts of speech in non-typical function: (A)symmetrical encoding of non-verbal predicates in Erzya. Linguistic Discovery 9(1). 137–172. (doi:10.1349/PS1.1537-0852.A.387)
OpenEdition suggests that you cite this post as follows:
Martin Haspelmath (October 28, 2025). What is a “copula”, and what is “predicative inflection”? Further comments on Creissels et al. (2026). Diversity Linguistics Comment. Retrieved April 15, 2026 from https://doi.org/10.58079/151xa




[Apologies for posting a question months after the blog entry; I’m slow] I’m curious about the idea that copulas aren’t tied to semantic roles (I’m avoiding “assign roles” here for a reason that I’ll clarify below), although my confusion may just be from the specifics of how ‘semantic’ is defined here. In discussions of copular clauses, everyone makes use of label sets like “identifier…identified” or “token…attribute” and the like, although the number of labels and their actual names vary. Now, I can see that those categories are rather transparently based on speaker intentions and are bound closely to the speech act itself (“thing I want to identify,” “info that lets me identify something”), so they seem more like labels for a role in a statement (more metalinguistic) rather than a role like Agent or Patient in an action correlated to a standard verb–the kind of thing where we can imagine we’re dealing with a description of a set reality independent of the speakers, etc. But Agent and Patient are themselves roles that the speaker is presenting the item being in; everything a speaker says is that speaker’s construction of the event. An Agent is always “thing that I’m presenting as Agent right now” (something that becomes obvious the moment you have an inventive speaker bent on shifting blame for an accident or the like; “The pedestrian impacted the front of my car”). And if I use a copula, hearers are expecting the things that go with it to be on one of a small set of roles, whether we call those “semantic” or not.
How much of the “they don’t assign roles” position is there because it’s required by a particular model architecture (one where semantics can’t make reference to anything pragmatics might have any interest in, and verbs show up with a list of demands, instead of verbs and roles correlating because the selection of both is driven by the selection of an event type, etc.)? If if the distinction between semantic roles and whatever it is that copulas go with is required on model-specific grounds, how absolutely necessary is it for typological discussions? We’d talk about the marking patterns regardless, and those are enough to make having labels like the ones you’re discussing useful anyway.
It would be great to have a truly rigorous and readily comprehensible notion of “semantic role”, but even in the absence of this, I felt that I understood what Creissels et al. (2026) meant. However, in my own definitions of concepts surrounding nonverbal clauses (see my 2025 paper), the notion of “semantic role” plays no role, so I am not defending it.