How can our general understanding of Human Language contribute to making peculiar language-particular patterns more comprehensible? This is what I keep wondering about, and in the recent excellent paper by Peter Arkadiev and Yakov Testelets, we find a really interesting case for discussion: the flag-article suffixes in the Circassian languages (Kabardian and Adyghe, two very closely related languages of the northern Caucasus region). These suffixes (Absolutive suffix -r, Oblique suffix -m) occur only on a nominal only when it is specific, and may be omitted when it is nonspecific:
The Oblique suffix -m occurs on ergative nominals as in this example, but also on dative and locative nominals, as seen in (20) and (22). In all cases, as Arkadiev & Testelets nicely show, the suffix may be omitted if the interpretation is nonspecific.
They note that one way of describing this situation is by saying that case inflection is limited to the article, as is sometimes the case in Germanic languages such as Icelandic, or even German (cf. der Tag/den Tag ‘the day (NOM/ACC)’). So rather than being “case suffixes” with curious restrictions on their occurrence, the markers -r/-m can be considered as “suffixed articles that distinguish case”, as proposed by Kumaxov (1971: Chapter 2) (p. 746). If these forms are (primarily) specific articles rather than case-markers, their distribution becomes completely normal, because they occur when the nominal is specific and do not occur when it is nonspecific, as is expected from specific articles. The peculiarities are (i) that they simultaneously serve as flags (but we also find this elsewhere, e.g. in Romance preposition-article combinations like French au or Italian nel), and (ii) that no flag is present when the article is absent. Arkadiev & Testelets mention a few further languages where the use of all flags is conditioned by specificity or definiteness, but the consistency of this pattern seems almost unique in Circassian (definitely a “typologically rare situation”, p. 745).
But still, there is a temptation to try to understand it in the context of some larger theory of the nature of Human Language, or at least to make it relevant to some larger theory that was proposed earlier, and the authors offer four possibilities:
– the functionalist explanation of DOM (differential object marking)
– generative approaches to DOM
– an analysis in terms of “pseudo-incorporation”
– the distinction between small nominals and large nominals (“NP” vs. “DP”)
After considering these four, they opt for the last one (p. 744), apparently because they feel that it also explains why nonspecific nominals are number-neutral. I will discuss this below, but first, I would like to ask whether the Circassian flag-article suffixes are “a challenge to functionalist theoretical approaches to differential case marking” (p. 744), because I have myself endorsed such a functional explanation in a recent paper (Haspelmath 2021): I have claimed, following a long tradition going back to the 19th century, that the strong universal tendency to flag only referentially prominent objects (P-arguments) when there is differential flagging can be explained by efficiency of coding: P-arguments are frequently referentially non-prominent, so referentially prominent arguments are not expected to be P-arguments and thus need more coding than the more typical P-arguments. (The situation with A-arguments is roughly the mirror image of this.) As Arkadiev & Testelets put it (quite rightly):
“to provide the effectiveness and the economy of communication, languages tend to mark an element whenever it is necessary” (p. 745)
Do the Circassian flag-articles provide a challenge to this functional explanation of a universal tendency? No, for two reasons: (i) because Circassian is just one data point, and individual exceptions cannot challenge a universal claim (see also the sporadic exceptions mentioned by Filimonova (2005)); (ii) because the flag-articles do not fall under the definition of differential argument flagging.
The first is a crucial general point that is often overlooked: Comparative considerations are not directly relevant to language-particular phenomena, and a single language can never challenge a universal tendency (unless one makes strong innatist assumptions; see below). Universal tendencies can only be shown wrong by providing better cross-linguistic data – as in Dryer (1988), where several of Greenberg’s word order universals were shown to have been based on bad sampling. (This also means, of course, that the potential of universal tendencies to illuminate a single language is limited. But this is the reality: Almost all aspects of all languages are social conventions and historically accidental.)
Maybe more interesting is the second point: I would say that differential argument flagging must be defined with respect to “primary flags”. This is not an established term, but I don’t think that it will be controversial that if something can be seen as a “secondary flag”, but primarily has a different meaning, then it will have a distribution governed by its primary function. The Circassian flag-articles are simultaneously articles and flags, but their article function is primary (in that they have a distribution thar is expected of articles). Thus, their distribution is not relevant to their flag function. Compare this situation with that of Turkish, which has a Definite Accusative suffix -i/u/-ü/-ı (there are many other languages with DOM that are similar). This can simultaneously be regarded as a suffixed definite article (a kind of “article-flag”), but its distribution is determined by its primary flag function. One would not say that it is a definite article with a curious “differential” distribution, because it is strictly limited to accusative function.
(Thus, I find it a bit misleading that they call the phenomenon “differential nominal marking”, because there is no differential marking of the article function; this is thus quite different from the “differential definiteness marking” that we find in English, where a preceding adpossessor blocks the definite article: Lee’s house vs. *Lee’s the house; cf. Haspelmath 1999)
Next, Arkadiev & Testelets examine whether the Circassian construction can be subsumed under “pseudo-incorporation” (cf. Borik & Gehrke (eds.) 2015), but they note that while this phenomenon is generally said to involve lack of flagging and number marking, which largely fits the Circassian situation (cf. the number-neutrality of ‘book(s)’ in the first example (b) above), the articleless nominals in Circassian show no positional restrictions and are in this respect quite unlike the Niuean situation (cf. Massam 2001).
But what if one wanted to argue that this is pseudo-incorporation after all? Couldn’t one say that Circassian happens to have the option of scrambling after the relevant structure is established? I don’t quite see how a pseudo-incorporation analysis can be excluded, given the generative methodology (with all its moving parts). Note that nobody has suggested that there is a fixed set of criteria by which pseudo-incorporation is identified, and Borik & Gehrke (2015: 10-11) say explicitly:
“even though the constructions from various languages that have been discussed under the label of pseudo-incorporation display common semantic traits, they can have rather different language-particular characteristics”
And in fact, Dayal (2011) treats a construction with an unflagged object in Hindi-Urdu as “pseudo-incorporation”, even though there is some positional freedom of the object nominal in this language, as in Circassian.
The reason why I am mentioning all this is that Arkadiev & Testelets end up favouring a distinction between “small nominals” (NPs) and “large nominals” (DPs), which is like pseudo-incorporation in that there are some stereotypical cases (e.g. Pereltsvaig’s 2006 Russian subject nominals that don’t control agreement), but there is no clear definition of these notions. It could of course be that the “small/large nominal” distinction is part of the innate grammar blueprint (“UG”) – but if one wants to avoid this assumption, one cannot use the same categories in different languages.
The supposed advantage of the “small/large” distinction (or “NP/DP” distinction, though the role of “determiners” for this distinction is unclear) is that it derives four disparate features of these nominals from their “structurally deficient” nature: (i) they lack an object or subject flag; (ii) they have narrow scope (illustrated in (49) below) (iii) they occur next to the verb; and (iv) they lack number marking. The lack of flagging can be attributed to the lack of “movement” for those languages where there is a word order difference associated with the differential object marking. But this is by no means always the case (as I pointed out in my 2018 review of Baker’s 2015 book “Case”), and Arkadiev & Testelets do not try to argue that the presence of the flag-article is linked to any kind of “movement”. And narrow scope is quite generally associated with indefiniteness or non-specificity (e.g. Ioup 1975), so one does not need a “small nominal” category of general syntax to explain this.
As for number marking, Arkadiev & Testelets note themselves that the correlation is quite imperfect, because in some Adyghe varieties, the articleless nominal can only be interpreted as singular in the absolutive role (p. 732). Saying that there is a “NumP” here (p. 735) amounts to admitting that there is no larger category here that makes any strong predictions. It is well-known that number marking is favoured by specific/definite contexts, but there seem to be good functional reasons for this, so that we do not need a category of the innate grammar blueprint for this.
Thus, as far as I can see, the differential object marking universals (Bossong 1991; Haspelmath 2021) are unaffected, and any innate categories of the generative type are not necessary to describe or explain the Circassian phenomena. Of course, the comparison with other languages has served to inspire the authors to look closely at the Circassian construction (as it often does; cf. Haspelmath 2016: 299), so comparison is useful for language-particular description/analysis. But their findings do not “challenge” the well-established explanation of differential object marking in terms of efficient coding, and it does not provide support for the idea that “pseudo-incorporation” or a “small/large nominal” distinction is (or is not) part of our innate grammatical knowledge. We can do general linguistics with a measurement approach, rather than uniform (innate) building blocks.
At the end, Arkadiev & Testelets suggest that we may eventually discover a scale of syntactic positions (maybe direct object > intransitive subject > transitive subject/nominal adjunct) that may show signs of “pseudo-incorporation”, i.e. lack of flagging, lack of number marking, narrow scope, and fixed position. But their paper has also shown, I think, that these four features should be considered separately (in the sprit of a multivariate approach), because they do not necessarily go together, and it is an empirical question whether they show a quantitative tendency to go together. Lumping them together under an “incorporation” stereotype or an “NP/DP” stereotype can only be the very first step toward more systematic cross-linguistics research that is based on rigorously defined comparative concepts for measurement.
So, lots of food for thought in this really well-researched and well-written article!
References
Arkadiev, Peter M. & Testelets, Yakov G. 2019. Differential nominal marking in Circassian. Studies in Language. John Benjamins 43(3). 715–751. (doi:10.1075/sl.18063.ark)
Baker, Mark C. 2015. Case: Its principles and parameters. Cambridge: Cambridge University Press.
Borik, Olga & Gehrke, Berit (eds.). 2015. The syntax and semantics of pseudo-incorporation. Leiden: Brill.
Bossong, Georg. 1991. Differential object marking in Romance and beyond. In Kibbee, Douglas & Wanner, Dieter (eds.), New analyses in Romance linguistics, 143–170. Amsterdam: Benjamins.
Dayal, Veneeta. 2011. Hindi pseudo-incorporation. Natural Language & Linguistic Theory 29(1). 123–167. (doi:10.1007/s11049-011-9118-4)
Dryer, Matthew S. 1988. Object-verb order and adjective-noun order: Dispelling a myth. Lingua 74(2–3). 185–217.
Filimonova, Elena. 2005. The noun phrase hierarchy and relational marking: Problems and counterevidence. Linguistic Typology 9(1). 77–113.
Haspelmath, Martin. 1999. Explaining article-possessor complementarity: Economic motivation in noun phrase syntax. Language 75(2). 227–243. (doi:10.2307/417260)
Haspelmath, Martin. 2021. Role-reference associations and the explanation of argument coding splits. Linguistics. (https://ling.auf.net/lingbuzz/004047)
Kumaxov, Muxadin A. 1971. Slovoizmenenie adygskix jazykov [=Inflection in the Circassian languages]. Moskva: Nauka.
Pereltsvaig, Asya. 2006. Small Nominals. Natural Language & Linguistic Theory 24(2). 433. (doi:10.1007/s11049-005-3820-z)
OpenEdition suggests that you cite this post as follows:
Martin Haspelmath (March 12, 2020). The peculiar flag-article suffixes in Circassian and general linguistics: Comments on Arkadiev & Testelets (2019). Diversity Linguistics Comment. Retrieved April 21, 2025 from https://doi.org/10.58079/nsvl
Facebook discussion with Yakov Testelets (https://www.facebook.com/martin.haspelmath/posts/10219870971737093):
Yakov: Many thanks, Martin, for your appreciation of our work and bringing this to a wider audience! Facebook isn’t a very convenient place for a discussion, and I’ll write my comments in a personal letter, probably. Here, just a couple of points. 1) I don’t think that the choice between “flag-articles” and DOM is really an “either – or” alternative. Is it necessary to define articles as elements that can occur in all or most grammatical contexts?.. Take some well-known differential object markers e.g. the preposition “a” in Spanish. Why couldn’t we call it an animateness marker restricted to direct objects only? Or the preposition “ʔet” in Hebrew a definite article, but, unlike ha-, restricted to direct objects, too? 2) I see in many languages that nominal modifiers (articles, demonstratives, adjectives…) turn a simple nominal construction into something more complex with different properties, and the DP vs. NP distinction is just one possible model to represent the difference.
Martin: Yes, Hebrew “et” could be regarded as a second article restricted to objects (cf. my parallel example of Turkish). But we don’t normally do this, and I was talking about what we normally do. (But it’s an interesting discussion that should be continued.) – And yes, nominal modifiers (or their absence) sometimes seem to have an effect on the external distribution of a nominal. This needs to be investigated, but probably not by means of the “hidden invariance” approach that is so typical of generative grammar. (I cannot disprove it, but hidden invariance presupposes innateness, which is inherently unlikely. Even many generativists reject innateness, incoherently – maybe because they don’t understand that cross-cultural invariance cannot be hidden unless it is directly rooted in biological factors. For some reason, they never consider convergent cultural evolution as a mechanism for cross-cultural invariance, even though this is widely recognized in other cross-cultral fields.)
Yakov: Re the “hidden invariance” debate, in general and in particular on the two-level DP vs. NP model: I believe that two simple methodological principles can always help: 1) we should strictly distinguish facts from hypotheses and never forget the difference; 2) of hypotheses, only those are really valuable that can potentially grow into accepted theories, i.e. part of the consensus. If there is no gradual growth of consensus in a scientific field, there is probably something wrong with it. I see the main weakness of the generative linguistics in that any problem requires a solution in terms of structure, and it’s impossible to offer a structural solution without proposing a hypothesis. Therefore too many hypotheses have been suggested, and every generative linguist must work as a theoretician – and you can hardly find another science in which every piece of research is theoretical. “Hidden invariance” is a theoretical goal that deserves to be pursued – if we don’t forget it’s a hypothesis that may fail as any other.
Martin: Yes, exactly. This is what I meant when I said that there are “too many moving parts” https://dlc.hypotheses.org/1167
Yakov: Theoretical “moving parts” do not necessarily multiply when a model well suits for some range of phenomena (like Trubetzkoy’s classical model of the phonological system built on oppositions suits for the classification of segments and makes correct predictions about all languages). But if a model is unsuccessful – and I think the constituent tree geometry is not very illuminating about cases – theoreticians begin to multiply the “moving parts” to squeeze some sort of explanation from it at any price – exactly as it happened with Trubetzkoy’s oppositions when Jacobson tried to apply his system for the theory of case. Today, it is impossible to believe how immensely popular and fashionable Jacobson’s wild theory of case was – just because today nobody believes any more that oppositions can account for everything.
Martin: That’s an interesting comparison.
Yakov: Re the exceptions to empirical universals: I agree that, within Greenberg’s tradition, not only are exceptions unavoidable but they are helpful (unless their number is too great), because they often enable typologists to better understand the interplay of factors. Edward Sapir was basically right in claiming that all languages somehow distinguish between “A hunter killed a tiger” and “A tiger killed a hunter”, although Sapir didn’t know that, in fact, a small minority of languages fail to express the difference in a regular way. There are caveats, however: 1) exceptions must not be numerous – i.e. in the case of the “flagging economy” hypothesis, languages like Nganasan when the personal pronouns lack the accusative marker, and the nouns have it, or Yazgulami or Parachi where the personal pronouns have the ergative marker, and the nouns lack it, or the marked case flagging the only argument (S with intransitives), like the absolutive case in Pitjantjatjarra, or the marked nominative case and the unmarked accusative in Kushitic, Nilotic, or Maidu – although there is no need to distinguish it from any other argument, and the like; 2) exceptions must not be diachronically stable – cf. the absence of the “reflexive vs. obviative” contrast in Old English or Old Dutch after they lost the Proto Germanic *sik reflexive – which lasted only several centuries. Then we may hypothesize that there must have been some factor that favours the contrast – although, to my knowledge, there is no historical record that the English or the Dutch of that time complained about their languages’ lack of a contrast that makes speaking more convenient ;-
Martin: I agree with all this – and I’m wondering how common exceptions like Nganasan and Yazgulami are (though the Old English absence of a reflexive vs. obviative contrast is not an exception; an exception would be counter-asymmetric coding, and this merely shows the lack of asymmetry). See Schmidtke-Bode & Levshina 2018 for a quantitative study, based on Bickel’s Autotyp data (in the DOM volume edited by Serzant & Witzlack-Makarevich).
I haven’t seen the last version of this paper, and I do not think I agree with all of its claims. However, I am not sure that Circassian flags should be considered “articles” and not “primary flags”. One piece of evidence is that while unmarked nominals in Circassian must be non-specific, marked nominals can be either specific or non-specific, so the function of these suffixes cannot be associated with marking specificity.
Note that this looks closer to incorporation, as concerns markedness patterns: when you have an incorporating construction reflecting non-specificity and a non-incorporating construction, presumably you will not think of the later as marking anything related to the specificity contrast. (Of course, I do not claim that incorporation is always associated with referential contrasts.)
This might interest you: “The DP Hypothesis in the grammar of Basque”
https://doi.org/10.1075/la.187.04art
It may indeed be relevant under what precise circumstances the “flag-articles” occur, but it would not be surprising if a language had an article that is optional in some of its uses. But it is true that I don’t know of another language with a “specific or nonspecific” article that can be omitted iff the nominal is nonspecific. – On the other hand, I don’t really know what “incorporation” is, because it’s not clearly defined in the literature (see this old blogpost: https://dlc.hypotheses.org/135). I attempted a definition in my 2018 review of the Polysynthesis handbook, but few people will be happy with it, I fear. (It makes no reference to referential properties of the noun, only to lack of modifiability.)