Since Plank (1991), syncretic patterns in inflectional paradigms have been studied in a sustained way, fuelled also by the increasing popularity of Distributed Morphology (e.g. Siddiqi 2019) and nanosyntax (Caha 2009; Baunaz et al. 2018).
Consider a German inflectional paradigm such as the following:
Many linguists would say that the first and third person plural here have a (systematically) syncretic form (-en), while the -t suffix of the 3SG and the the -t suffix of the 2nd person plural are accidentally homonymous (i.e. not systematically syncretic).
Here I’d like to make a few comments on such patterns, on the occasion of reading Caha (2019), a very accessible introduction to some of the main issues. Caha’s own research focus has been on “contiguous case syncretisms” in Russian and other languages, and the nanosyntax approach offers an ingenious solution for them (see Caha 2009). But in the 2019 paper, Caha begins by discussing accidental vs. systematic syncretism (where only the latter is relevant for grammatical analysis), and this discussion made me wonder whether inflectional syncretism is ever systematic. Maybe it is always “accidental” from the perspective of a particular language?
In fact, one might propose the following as a general heuristic:
(1) Anti-syncretism Principle: Difference of function implies difference of form.
This would be the opposite of Müller’s (2005: 236) “Syncretism Principle” (Identity of form implies identity of function; see also Hein & Weisser 2021: §2), and it would lead us to suspect that there are two different forms with the shape -en in German: 1PL -en and 3PL -en.
How can we tell which approach is right?
The problem is the same as what semanticists have discussed under the heading of “polysemy vs. homonymy”, and the consensus seems to be that there is no way for linguists to draw the distinction in a general way (e.g. Geeraerts 2001). Just think of animal names such as crane, ram, bat, hawk, and mole – all these can be used with a non-animal meaning (crane, ram and bat for tools, and hawk and mole for people), and for all of them, a more or less salient similarity can be discerned. So are they all polysemous? Or are they all homonymous pairs? Or are some polysmous and some homonymous? It is impossible to tell. (If we adopt a “Polysemy Principle”, we will be inclined to emphasize the similarities, and if we adopt an “Anti-polysemy Principle”, we will be inclined to emphasize the differences. But we cannot tell which analysis is the right one.)
Caha’s paper on syncretism is remarkable because it is so strongly typologically oriented (like much of the nanosyntax literature). This becomes clearest in his discussion of the Greek and Arabic case systems. On the one hand, it might seem that Modern Greek and Standard Arabic both have three cases: nominative, genitive and accusative. However, the case called “genitive” in both languages has different functions: it can code adpossessor and recipient in Greek, but only adpossessor in Arabic. Thus, if we distinguish GEN (= adpossessor) and DAT (= recipient) for Greek, we can map Greek onto Arabic in a more satisfactory way:
Now one might object that GEN and DAT are really semantic roles here, not case labels, but this is a necessary move in Caha’s typological approach – because case features can be equated across languages only on the basis of their meanings. (Indeed, Caha defines syncretism as a situation where “a single marker [appears] with multiple meanings”, not as a situation where an exponent realizes multiple (possibly nonsemantic) feature values.) Basically, this means that these case labels are comparative concepts, specifically comparison meanings. “Inflectional syncretism” is thus a special case of the coexpression patterns that have been discussed by typologists under the heading of semantic maps, as Caha notes in §2 (more recently, I proposed that they are better called coexpression diagrams; Haspelmath 2023).
Now what about the distinction between systematic and accidental syncretism? In practice, researchers only provide explanatory accounts of cases that they judge as “systematic” (= as “polysemous” rather than homonymous), as was noted by Hein & Weisser (2021: §2):
“Nonetheless, often it remains quite unclear whether a given case should be counted as accidental homonymy or as systematic syncretism and it must be admitted that, in practice, it often reduces to an analytical choice. Cases which lend themselves to straightforward accounts by means of theoretical concepts like underspecification or decomposition (which will be discussed below in more detail), will be analyzed as instances of systematic syncretism whereas cases that do not will often be treated as accidental.”
But if there is no clear way of distinguishing between systematic and accidental syncretism, the whole enterprise of general theories of syncretism becomes problematic. Caha (2019) devotes five pages to discussing the difference, but he ends up saying, quite honestly: “Discrepancies of opinion are bound to persist. The reason is the clash between various types of considerations and perspectives…” I discussed the problem of subjective decisions in grammatical analysis elsewhere, and it seems that the phenomena of syncretism are another example where these problems suggest an alternative approach – an alternative to the generative mainstream represented by Müller’s “Syncretism Principle”.
The alternative is in fact mentioned by Caha, who cites Cysouw (2007: 233) as saying that “the difference between common and rare [syncretisms] seems to be a much more important fact to be modelled than the difference between rare and non-existing [syncretisms].” On this approach, there is no need to distinguish between “systematic” and “accidental” syncretism, and in some sense, syncretism is always “accidental” (from the perspective of the possible meaning distinctions that languages could make).
One might alternatively suggest that syncretism is always systematic (which would amount to the same, because there is likewise no difference between two kinds of syncretism), but if “systematic” means “part of a language-particular system”, this is not so easy to implement, because for some syncretisms, describing them as semantically related leads to very far-fetched analyses. (For example, some authors have suggested that the nominative plural -s and the genitive ’s in English is not accidental; Leiss 1997; Caha 2016.)
Positing syncretism in all cases when there is a (perceived) difference in function (as per the Anti-syncretism Principle suggested above) is easier to implement in practice, but whether there is a difference in function in any given case is of course subjective (as we saw), so how do we make our theoretical work more objective?
Caha’s (2019) paper contains a remark that points in the direction of a solution. With respect to the systematicity of the GEN.SG–NOM.PL syncretism, he notes (emphasis mine):
“Within the confines of a single language, the GEN.SG—NOM.PL syncretism is certainly systematic and has to be accounted for. On the other hand, if one is interested in patterns that are common in a large and diverse sample of languages (e.g., in the one used by Baerman et al. 2005), the GEN.SG—NOM.PL syncretism appears insignificant against that sample… However, it seems fair to say that both types of systematic patterns need to be accounted for, regardless of whether their systematic nature can be observed in a single language only or more broadly.”
In general, linguists need accounts for two types of systematic patterns: language-particular patterns (p-linguistics) and cross-linguistically general patterns (g-linguistics). It is typical of generative grammar (including nanosyntax) that one attempts to account for both types of regularities by means of the same primitive concepts, hypothesized to be natural kinds that are part of the innate language blueprint (or UG). But this is not necessary: Alternatively, many linguists would say that language-particular categories are distinct from comparative concepts (e.g. Dan Slobin, from the perspective of language acquisition), and that we do not need natural kinds in linguistics for the time being.
On this view, g-syncretism can be treated quite differently from p-syncretism.
For the former (i.e. cross-linguistically general patterns of syncretism), we need general explanations, presumably in terms of semantic similarity (as originally noted in Haiman 1974, the first paper on cross-linguistic tendencies in grammatical coexpression patterns that I am aware of). We do not need to distinguish “systematic” and “accidental” syncretism, as we can conflate rare and non-existent syncretisms (as per Cysouw 2007). All syncretism patterns that occur in relevant numbers in the world’s languages are treated as relevant, regardless of what their language-particular status is. Interestingly, Caha (2019) notes that some intriguing coexpression patterns (of the *ABA type) are relevant at the cross-linguistic level, but are represented only by isolated cases in many languages (specifically, Caha refers to Bobaljik’s (2012) work on comparative-superlative marking). On this view, the theoretical relevance (or “systematicity”) of coexpression patterns arises at the cross-linguistic level, and we do not need to identify “systematic syncretism” in particular languages. G-syncretism is relevant because it is found in the world’s languages, regardless of possible theoretical accounts.
By contrast, language-particular syncretism (or “p-syncretism”) can always be treated as accidental, and this would be the default by the Anti-syncretism Principle. In practice, of course, many descriptive grammars do present inflectional paradigms in this redundant way. Even though all theorists agree that German verbs have only two distinct inflected forms in the plural, most descriptive grammars treat the 1PL–3PL syncretism as accidental. Instead of the paradigm that I gave at the beginning of this post, they give a paradigm such as the following, with the usual sequence of the three persons (1 > 2 > 3) and the syncretism uncaptured (= treated as accidental):
Caha (2019) notes that capturing language-particular syncretisms is in general a goal of p-analyses, because we want to have economical analyses:
“The purpose of a descriptive grammar is to list the minimal number of distinct forms and to display the paradigms as compressed as is possible.” (§2, below (14); boldface added)
If we describe Greek as in (14) above, we do not minimize the Greek cases, so many grammarians would not like to adopt such descriptions (also because they might appear to be biased by the Ancient Greek system, which did distinguish between genitive and dative, a contrast lost in the modern system). However, such descriptions are surely possible, and sometimes we might want to adopt them (as Caha demonstrates in his discussion of a similar phenomenon in Estonian).
Many grammarians would say that only the minimal description of a language is adequate, but it is difficult to say in general what exactly counts as “minimal” (Chomsky used to think of this in terms of an evaluation metric (Goldsmith 2011; Ermolaeva 2023), but this approach was abandoned by most generativists). In general, it seems difficult or impossible to avoid some amount of subjectiveness in language-particular analyses. For p-descriptions, the primary criteria of adquacy are correctness and completeness, not simplicity. Thus, the Anti-syncretism Principle (leading us to describe all syncretic patterns as accidental) does no harm with p-syncretism either.
I found Caha’s (2019) paper very transparent in the way it discusses syncretisms and ways of thinking about them, and while Caha would probably not endorse everything I say here, he does make it clear that p-analyses and g-analyses are potentially different, citing my 2003 paper on semantic maps as an approach that is related to his. Caha’s (2009) work on case inflection starts out with the observation that case syncretism is sometimes (or often) based on contiguity in paradigms (see also Hein & Weisser 2021: §4), and this is of course similar to the implicational map of indefinite-pronoun markers of my 1997 book (Haspelmath 1997). Implicitly, Caha’s paper shows that nanosyntax features are comparison meanings, and the feature trees that are used in this approach are perhaps just another kind of coexpression diagram (Haspelmath 2023).
References
Baerman, Matthew & Brown, Dunstan & Corbett, Greville G. 2005. The syntax – morphology interface: A study of syncretism (Cambridge Studies in Linguistics). Cambridge: Cambridge University Press.
Baunaz, Lena & De Clercq, Karen & Haegeman, Liliane M. V & Lander, Eric (eds.). 2018. Exploring nanosyntax. New York: Oxford University Press.
Caha, Pavel. 2009. The nanosyntax of case. Tromsø: University of Tromsø. (PhD dissertation.) (https://munin.uit.no/handle/10037/2203)
Caha, Pavel. 2016. GEN.SG = NOM.PL: A mystery solved? Linguistica Brunensia 64(1). 25–40.
Caha, Pavel. 2019. Syncretism in morphology. Oxford Research Encyclopedia of Linguistics 2019. (doi:10.1093/acrefore/9780199384655.013.552)
Cysouw, Michael. 2007. Building semantic maps: The case of person marking. In Miestamo, Matti & Wälchli, Bernhard (eds.), New challenges in typology: Broadening the horizons and redefining the foundations, 225–248. Berlin: Mouton De Gruyter.)
Ermolaeva, Marina. 2023. Evaluating syntactic proposals using minimalist grammars and minimum description length. Journal of Language Modelling 11(1). 67–119. (doi:10.15398/jlm.v11i1.334)
Geeraerts, Dirk. 2001. The definitional practice of dictionaries and the cognitive semantic conception of polysemy. Lexicographica 17. 6–21.
Goldsmith, John. 2011/2013. The evaluation metric in generative grammar. 50th anniversary celebration for the MIT Department of Linguistics. (http://people.cs.uchicago.edu/~jagoldsm/Papers/singh.pdf)
Haiman, John. 1974. Concessives, conditionals, and verbs of volition. Foundations of Language 11(3). 341–359.
Haspelmath, Martin. 1997. Indefinite pronouns. Oxford: Oxford University Press. (https://library.oapen.org/handle/20.500.12657/30819)
Haspelmath, Martin. 2003. The geometry of grammatical meaning: Semantic maps and crosslinguistic comparison. In Tomasello, Michael (ed.), The New Psychology of Language, volume 2, 211–243. New York: Lawrence Erlbaum. (10.5281/zenodo.831410)
Haspelmath, Martin. 2023. Coexpression and synexpression patterns across languages: Comparative concepts and possible explanations. Frontiers in Psychology 14. (doi:https://doi.org/10.3389/fpsyg.2023.1236853)
Hein, Johannes & Weisser, Philipp. 2021. Syncretism – Recurring patterns (for Wiley-Blackwell Companion to Morphology 2023). LingBuzz. (https://ling.auf.net/lingbuzz/006677)
Leiss, Elisabeth. 1997. Synkretismus und Natürlichkeit. Folia Linguistica 31(1–2). 133–160. (doi:10.1515/flin.1997.31.1-2.133)
Müller, Gereon. 2005. Syncretism and iconicity in Icelandic noun declensions: A Distributed Morphology approach. In Booij, Geert & van Marle, Jaap (eds.), Yearbook of Morphology 2004, 229–271. Dordrecht: Springer. (doi:10.1007/1-4020-2900-4)
Plank, Frans. 1991. Of abundance and scantiness in inflection: A typological prelude. In Plank, Frans (ed.), Paradigms: The economy of inflection, 1–40. Berlin: Mouton de Gruyter. (doi:10.1515/9783110889109.1)
Siddiqi, Daniel. 2019. Distributed Morphology. In Audring, Jenny & Masini, Francesca (eds.), The Oxford handbook of morphological theory, 143–165. Oxford: Oxford University Press. (https://doi.org/10.1093/oxfordhb/9780199668984.001.0001)
OpenEdition suggests that you cite this post as follows:
Martin Haspelmath (January 11, 2026). Is inflectional syncretism ever “systematic”? Some comments on Caha (2019). Diversity Linguistics Comment. Retrieved February 15, 2026 from https://doi.org/10.58079/15hdb
Doesn’t your distinction boil down to repeating vs. rare/unique patterns? Of course people can disagree over where there is the boundary between repeating and rare.
Yes, that is true – but it is a well-known issue in all kinds of quantitative research, and there are well-known statistical techniques for dealing with the issue. So I would say that we can probably stop agonizing over the systematic vs. accidental nature because we can reduce the problem to a well-known other problem that we need to address anyway.