Is “markedness” still alive? On Kiparsky, Dixon and some others

The idea that inflectional features like tense and number often have two contrasting values which are somehow systematically asymmetric goes back to Jakobson (1932; 1957), and his idea that often one value is marked and the other unmarked has been very influential, especially in structural/generative linguistics. The term “markedness” has also been used in many other ways, leading to a lot of confusion, and I recommended doing without it entirely (Haspelmath 2006).

References to morphosemantic markedness seem to have declined over the last years (maybe my paper, which was widely cited, has had an effect), but there are still a few authors who are taking this notion very seriously. Grimm’s (2018) recent paper on countability distinctions (discussed recently on this blog) contains much that I agree with and anticipated in my own work (Haspelmath & Karjus 2017), but Grimm also says:

“Following Jakobson (1957) and Horn (1989), I take morphological markedness to be rooted in semantic markedness… In particular, I assume that a grammatical number category’s coding preferences are linked to the semantic markedness patterns…” (2018: 551)

What does it mean for morphological markedness to be “rooted in” semantic markedness? This remains vague in Grimm’s paper. I have proposed an alternative explanation for many of the effects that have been attributed to morphosemantic markedness, which makes reference to frequency-induced predictability and efficient coding. I keep asking whether there are serious competitors to my explanatory theory, because it has not been widely adopted yet.

Grimm refers to Kiparsky & Tonhauser (2012), who discuss a somewhat wider range of phenomena than Grimm – not only nominal number, but also person, gender, case, and evidentiality. They say that they follow Jakobson in claiming that “morphological markedness is grounded in semantics, but has consequences in syntax, morphological form, and even in phonology” (2012: 2071). They posit a “convergence/correlation” between formal and semantic features, and argue in particular against the view, fairly widespread among formal semanticists (e.g. Sauerland et al. 2005), that the plural is the unmarked number value. This goes against Jakobson’s (and Greenberg’s 1966) view, and Kiparsky & Tonhauser side with Jakobson. But the argumentation is very complex, and requires settling on a particular way of distinguishing between implicatures and encoded (semantically spoecified) semantic features. If the English plural denotation includes singular entities (and is in this way unmarked/unspecified), then (1) should be infelicitous if Lee saw a single bear:

(1) Lee didn’t see bears.

Kiparsky & Tonhauser claim that (1) is actually OK in single-bear contexts, because one can say Lee didn’t ever see bears, but he often saw single ones. This example gives one a sense of the degree of subtlety involved in this discussion among semanticists, which has been largely limited to English.

A (2011) paper by Bale, Gagnon & Khanjian (not taken into account by Kiparsky & Tonhauser) broadens the scope a little in that it includes some facts from Western Armenian, where the counterpart of “Do you have children?” (bəzdig-ner uni-s? [child-PL have-2SG]) cannot be answered positively if the person has only a single child (unlike in English, where the dialog Do you have children? Yes, one boy is quite normal). So is the plural in Armenian marked, while it is unmarked in English? Unlike Kiparsky & Tonhauser, who follow Jakobson in making sweeping claims (or sweeping suggestions) about all categories and all languages, Bale et al. are very careful in their interpretations. They consider two possibilities (or “hopes”): (i) that there may be a systematically positive correlation between morphological and semantic markedness (if semantic features are interpreted as restrictions on denotations), and (ii) that there may be a systematic negative correlation (if features are interpreted as augmenting functions). They find that the plural in “most languages” (French, German, Italian, Spanish etc.) is consistent with the latter view (thus agreeing with Sauerland et al. 2005), but they admit that they do not know how widespread the opposite pattern (exemplified by Western Armenian) is. Since plural marking is very often optional in the world’s languages (as can be seen in my 2005 WALS contribution on nominal plurals), I would venture the hypothesis that the Armenian pattern is at least as common as the English/French pattern.

Bale et al. also discuss problems for their interpretation of the English plural, and end on a laudably cautious note:

“In fact, it remains a strong possibility that there is no connection between subset relations among denotations in semantics and the distributional features in morphology.” (2011: 220)

If my theory that refers to frequency-induced predictability is correct, then indeed we would not expect any clear correlation between the forms and meanings of inflectional constructions. (I should mention one limitation: frequency-induced predictability makes clear predictions about overt coding, but not necessarily about syncretisms and restricted occurrence of marked features; however, we know very little about the latter types of “markedness” effects in the world’s languages, so they may not be very general.)

Kiparsky & Tonhauser also mention gender and evidentiality, but they say nothing about evidentiality and admit that “establishing semantic markedness for gender is inconclusive” (which I do not find surprising because the genders are not generally asymmetric in frequency). For person, they claim (as expected) that third person is unmarked both semantically and morphologically, but the argumentation that markedness applies in the same way to inclusive, first (exclusive) and second person is complex and dense (they do not consider the fact that the third person also has the highest frequency and thus its behaviour is also expected from an efficiency perspective). For case, the argumentation in based on the Kiparsky-Wunderlich theory of syntactic functions, where nominative case is the least specified (the most unmarked), and dative case is the most specified (the most marked), with accusative/ergative in between. This is in line with my frequency-based predictions (because nominative is the most, and dative the least frequent), but predictions diverge when it comes to locative cases, which Kiparsky & Tonhauser try to assimilate to the grammatical cases: They say that allative and ablative are equally marked, whereas in fact languages very often show a goal-source asymmetry such that source is marked in a special way (as is expected because spatial sources are rarer in texts than goals).

Another well-known author who has talked about markedness is Dixon (2010: §5.7), and he also situates his discussion in the old Trubetzkoy/Jakobson tradition, though he explicitly says that morphological markedness is quite different from Trubetzkoy’s phonological markedness, and accuses Jakobson of “overusing” markedness (thus agreeing with some key points of my 2006 paper). Dixon distinguishes between formal markedness and functional markedness, where the former refers exclusively to overt coding, while the latter refers to both semantic unrestrictedness and distributional unrestrictedness (the latter is subsumed under morphological markedness by Bale et al. – so the terminology is really confusing!).

There are two aspects of Dixon’s discussion that I liked particularly (in addition to its exemplary clarity of presentation): First, Dixon does not make sweeping claims, and recognizes that “formal and functional markedness do not always correlate” (2010: 238), as illustrated by Portuguese gender, where masculine is functionally (distributionally) but not formally unmarked (it is not zero-marked). Second, he talks not only about overt coding, but also mentions the greater frequency of the unmarked term, and in particular, he notes that “the realization of the unmarked term is typically a shorter or simpler phonological form than the realization of marked terms” (giving the example of Warrgamay tense-mood marking, where the functionally unmarked category has the suffix -y, contrasting with -ma/-gi/-lagu for the marked categories; 2010: 239). So the formal asymmetries are not just about the presence or absence of something (as in Jakobson’s binary thinking) – they are about coding length, which suggests an substantive explanation in terms of coding efficiency (rather than a structuralist explanation).

Dixon says cautiously that “the notion of markedness can be useful in linguistic description and explanation, so long as it is defined and applied with care, and not overdone (as, unfortunately, it often is)” (2010: 235). I do agree that the Trubetzkoyan and Jakobsonian notions of markedness have inspired a lot of thinking in linguistics (it seems that is was a genuinely novel idea that had no antecedents in 19th century linguistics). But the contemporary verdict seems clear: There is no coherent “markedness” concept with explanatory value, and we do not need markdness for description either – on the contrary, the term markedness continues to confuse many linguists. To the extent that truly general phenomena exist in the world’s languages, they require explanations that have few connections with Jakobson’s binary abstract specifications.


Bale, Alan, Michaël Gagnon & Hrayr Khanjian. 2011. On the relationship between morphological and semantic markedness. Morphology 21(2). 197–221. doi:10.1007/s11525-010-9158-1.

Dixon, R. M. W. 2010. Basic linguistic theory. Vol. 1. Oxford: Oxford University Press.

Greenberg, Joseph H. 1966. Language universals: With special reference to feature hierarchies. The Hague: Mouton.

Grimm, Scott. 2018. Grammatical number and the scale of individuation. Language 94(3). 527–574. doi:10.1353/lan.2018.0035.

Haspelmath, Martin. 2005. Occurrence of nominal plurality. In Martin Haspelmath, Matthew S. Dryer, David Gil & Bernard Comrie (eds.), The world atlas of language structures. Oxford: Oxford University Press.

Haspelmath, Martin. 2006. Against markedness (and what to replace it with). Journal of Linguistics 42(1). 25–70. doi:10.1017/S0022226705003683.
Haspelmath, Martin & Andres Karjus. 2017. Explaining asymmetries in number marking: Singulatives, pluratives, and usage frequency. Linguistics 55(6). 1213–1235. doi:10.1515/ling-2017-0026.

Horn, Laurence R. 1989. A natural history of negation. Chicago: The University of Chicago Press.

Jakobson, Roman. 1932. Zur Struktur der russischen Verbums. Pražskỳ Linguistickỳ Kroužek.

Jakobson, Roman. 1957. Shifters, verbal categories, and the Russian verb.

Kiparsky, Paul & Judith Tonhauser. 2012. Semantics of inflection. In Claudia Maienborn, Klaus von Heusinger & Paul Portner (eds.), Semantics: An international handbook of natural language meaning, 2070–2097. Berlin: De Gruyter Mouton.

Sauerland, Uli, Jan Anderssen & Kazuko Yatsushiro. 2005. The plural is semantically unmarked. In Stephan Kepser & Marga Reis (eds.), Linguistic evidence: Empirical, theoretical and computational perspectives. Berlin: Walter de Gruyter.



2 thoughts on “Is “markedness” still alive? On Kiparsky, Dixon and some others

  1. In my native language, if you have a male friend, a group of male friends, or a mixed group, you use one word. If you have a female friend, or an all-female group of friends, you use another word (which is the first word + a morpheme).

    What do you suggest, how such a feature should be called?

  2. One aspect of abolishing markedness is perhaps not so innocent, which is that you then need some kind of language-specific typing in linguistic structures (as assumed in HPSG), so that for example an NP-representation in a language with case features (as required if there is any kind of case-agreement) always needs to have a case-value. So then the syntax will have to be elaborated in a fairly dumb-looking way to dump a default case value (nominative in Icelandic, Turkish and Kayardild, perhaps Modern Greek as well) on all NPs that don’t get case via some more specific environment.

    Whereas, in a theory such as LFG that does not have language-specific type specifications, the default case-value can be no case feature, with this referenced by the morphology of nominative (or ‘absolutive’, if that looks like the default) by a specification that says that the case-value must be absent. I find this significantly more attractive than the typed system (for case; gender and number raise different issues), although this appearance could be an illusion waiting to be overthrown by more & better typology and analysis.

    In the case of number, I think a plausible view is that an unmarked form would be neutral between singular and plural, with the possible existence of features specifying both plurality (‘I properly contain an atom’ in formal semanticalese), and singularity (‘I do not properly contain an atom’). Frequency would then plausibly be a major factor in controlling actual marking. About gender, I think it’s probably wisest to say nothing.

Leave a Reply

Your email address will not be published. Required fields are marked *

This site uses Akismet to reduce spam. Learn how your comment data is processed.