Explaining special patient and agent marking: Is there a serious challenge to the theory of efficient coding? (A discussion of Chappell & Verstraete 2019)

Many languages have special patient marking when the patient is referentially prominent (definite, animate, or a personal pronoun), and special agent marking when the agent is non-prominent. Typologists have been aware of this since the 1970s (e.g. Silverstein 1976; Moravcsik 1978), and the first type of phenomenon (typically called differential object marking, or DOM) is very widely known, as it occurs in well-known languages such as Spanish, Russian and Hindi-Urdu.

I have long thought that the explanation of these tendencies is also well-established: Other things being equal, languages tend to reserve special coding for unpredictable meanings, because this yields an efficient tradeoff between speaker needs (saving energy) and hearer needs (clarity) (I summarize the general theory of efficient coding very briefly here).

But this theory has been challenged in a number of ways in recent years. Some generative linguists have provided accounts of DOM based on innate knowledge, and have acted as if the theory of efficient coding did not exist (as discussed earlier here and here). I found this work deeply puzzling, but in the present post, I would like to discuss a few other challenges.

First, briefly, there is the challenge by Bickel et al. (2015), who claimed that their data provided “typological evidence against universal effects of referential scales on case alignment”. However, this is not at all clear, because their database of 435 languages contains very few cases of languages that do not conform to the referential-scale predictions, and many languages that conform to them perfectly. The choice of their statistical methods is non-obvious (to say the least), and in a subsequent study, Schmidtke-Bode & Levshina (2018: 531) come to the opposite conclusion:

The global structure of the classic hierarchies (pronouns > nouns) and all high- low prominence distinctions (animacy, definiteness, topicality, kinship) are almost without exception, and while there is more variation within the pronominal domain, a closer look at the data reveals that the number of counterexamples is not significant enough to override the strong support that the referential hierarchy … receives from our statistical models. Therefore, our conclusion is the opposite …, namely that there is evidence for universal scale effects on case marking.

Bernard Comrie had made very similar comments in a plenary talk at the 2012 SWL conference in Dubrovnik (as discussed briefly here).

Thus, while further research is definitely warranted, I do not think that there is reason to abandon the Silverstein-Moravcsik generalizations for empirical reasons. Sinnemäki (2014) found that differential object marking is widespread on all continents, and while differential agent marking is less common (and particularly well-respresented in Australia, in New Guinea and in inner Eurasia), the patterns are so strikingly similar (in a mirror-image way) that an analogous functional explanation seems very likely.

Now earlier this year, another very interesting typological paper was published: Optional and alternating case marking: Typology and diachrony”, by Hilary Chappell and Jean-Christophe Verstraete. These authors do not present a full challenge to the efficiency theory, but they claim to highlight “problems with the classic model of case marking based on markedness reversal” (where by “markedness reversal”, they presumably mean the efficiency theory in terms of a tradeoff between speaker effort and clarity).

First, they distinguish between three types of differential role coding: optional coding (where an accusative or ergative marker can be absent without affecting the grammaticality, in the same syntactic construction), alternating coding (where two different accusative or ergative markers occur under different conditions), and split coding – either referential splits (where the coding depends on animacy, person or nominality) or constructional splits (such as tense/aspect splits, negation-conditioned splits). Such coding splits are well-established in the literature, but they are often lumped together with optional and alternating marking under “differential case marking” – and Chappell & Verstraete object to this. They prefer to treat Mandarin Chinese ba-marking of objects, and Umpithamu optional ergative marking (e.g. ama(-mpal) umpaniluungku yuku [man(-ERG) broke tree] ‘The man knocked down the tree’) as different from split coding. After all, there are no clear lexical or constructional conditions for this variability, and there is no valency alternation involved either.

In my own work (especially in my forthcoming paper, Haspelmath 2020a, about the connection between referential prominence and argument flagging), I followed the tradition and discussed both split coding and “optional coding” under the general heading of differential coding – because they are apparently motivated equally by the need for efficient coding. In another, more general paper (Haspelmath 2020b), I have extended the term differential coding to all kinds of other grammatical phenomena which are known to show systematic coding asymmetries in the world’s languages. Thus, plurative and singulative marking of nouns, or special coding of alienable possession, or reflexive marking on introverted vs. extroverted verbs show very similar patterns: Special coding tends to be limited to situations where the meaning in question is least predictable. In all these domains, we sometimes find patterns of strict conditioning, and sometimes patterns of more loose conditioning. The two types can be distinguished, but they can also be seen together in order to explain them together.

Chappell & Verstraete (2019: §2.1.2) recognize that the term “optional” might suggest free variation, which they do not want to imply, because the use or non-use of the accusative/ergative marker is often conditioned in a clear way, e.g. by definiteness or focus (for instance, this is the case with accusative marking in Hebrew, which is fully obligatory with definite nominals, and impossible otherwise). But in other cases, the conditioning is not very clear, and linguists need to consider a long list of variables that are probabilistically associated with the occurrence or non-occurrence of the marker (as in Japanese, where subject and object marking is generally optionally, but still not randomly distributed). So one might distinguish additionally between discourse-conditioned coding (as in Hebrew), and probabilistically variable coding (as in Japanese). It seems that Chappell & Verstraete primarily have the first in mind.

They do recognize that split coding often occurs under similar conditions as optional coding, and that in many languages, the coding may obey two different kinds of regularities simultaneously. In the Australian language Umpithamu, for example, the ergative marker -mpal (in the example above) is optional only with animate nominals, but obligatory with inanimate ones – exactly as expected on the efficiency theory. But they still want to distinguish optional and split coding. There is nothing wrong with this, but I did not understand what the distinction buys them – what kinds of generalizations they can capture only in this way. One could equally, it seems, make the cut between (Japanese-style) probabilistically variable coding on the one hand, and strictly conditioned asymmetric coding on the other (whether discourse-conditioned, lexically conditioned or constructionally).

One generalization that Chappell & Verstraete completely miss is the finding that when there are two different markers, the longer marker tends to be used for the less predictable situation. They actually mention two languages with two different ergative markers (Warrwa and Nêlêmwa), and in both cases, the distribution is just as expected (Warrwa has -na for the ordinary ergative, and -nma for the focal ergative; and Nêlêmwa has a+ for the nimate ergative, and ru+ for the inanimate one).

Thus, I do not see a serious challenge to the efficiency theory. Chappell & McGregor seem to focus on the somewhat diverse language-particular conditioning factors, rather than on the over all patterns. Language-particular factors may involve ingredients with little cross-linguistic generality, such as affectedness (which plays a role in conditioning the Mandarin ba-construction, as well as ‘take’-derived object markers in West African languages). But greater affectedness of the patient is not generally associated with more marking – on the contrary, it is sometimes associated with less marking (as in the well-known English contrast between shoot someone and shoot at someone).

When one focuses on the truly general aspects of differential role coding, one finds that more coding of the P and less coding of the A is very strongly associated with what I call referential prominence (in Haspelmath 2020a) – whether inherent prominence (animacy, pronominality, 1st/2nd person) or discourse prominence (definiteness, givenness, non-focusedness). The predictions of the efficiency theory of role coding thus cover general patterns in both “optional coding” and coding splits in the same way.

A final possible challenge that is mentioned by Chappell & Verstraete (in their §3.2, where they talk about an alleged “recalibration” of the field) is Cristofaro’s (2013; 2019) claim that

“hierarchies which look like a valid generalization synchronically may in fact be composite diachronically, with different parts deriving from quite different historical sources, and often involvingporinciples that are quite different from the functional principle supposed to underlie the hierarchy” (p. 19)

I discussed Cristofaro’s claims in my 2019 paper on mutational constraints, arguing that multi-convergence of sources shows that the resulting patterns are not due to constraints on change, but that it is the other way round. Chappell & Verstraete include a long section (§4) on diachronic origins of optional ergative and accusative markers, and their discussion nicely shows that the sources are very diverse. What I found particularly striking is the finding that obligatory markers may become optional, e.g. obligatory instrumental markers may become optional ergatives, and obligatory dative markers may become optional accusative markers. There seems to be nothing about these diachronic sources that predicts the resulting optionality. On the contrary, it is the resulting functions (marking agent and patient) that make the optionality (or variability) possible. So I see the diachronic sources as indirectly providing more support for the efficiency theory of variable agent and patient marking.


Bickel, Balthasar, Alena Witzlack-Makarevich & Taras Zakharko. 2015. Typological evidence against universal effects of referential scales on case alignment. In Ina Bornkessel-Schlesewsky, Andrej Malchukov & Marc D. Richards (eds.), Scales and hierarchies: A cross-disciplinary perspective, 7–43. Berlin: De Gruyter.

Chappell, Hilary & Jean-Christophe Verstraete. 2019. Optional and alternating case marking: Typology and diachrony. Language and Linguistics Compass 13(3). e12311. doi:10.1111/lnc3.12311.

Cristofaro, Sonia. 2013. The referential hierarchy: Reviewing the evidence in diachronic perspective. In Dik Bakker & Martin Haspelmath (eds.), Languages across boundaries: Studies in memory of Anna Siewierska, 69–93. Berlin: De Gruyter Mouton.

Cristofaro, Sonia. 2019. Taking diachronic evidence seriously: Result-oriented vs. source-oriented explanations of typological universals. In Karsten Schmidtke-Bode, Natalia Levshina, Susanne Maria Michaelis & Ilja A. Seržant (eds.), Explanation in typology, 25–46. Berlin: Language Science Press. http://langsci-press.org/catalog/book/220.

Haspelmath, Martin. 2019. Can cross-linguistic regularities be explained by change constraints? In Karsten Schmidtke-Bode, Natalia Levshina, Susanne Maria Michaelis & Ilja A. Seržant (eds.), Competing explanations in linguistic typology, 1–23. Berlin: Language Science Press. http://langsci-press.org/catalog/book/220.

Haspelmath, Martin. 2020a. Role-reference associations and the explanation of argument coding splits. to appear. https://ling.auf.net/lingbuzz/004047.

Haspelmath, Martin. 2020b. Explaining grammatical coding asymmetries: Form-frequency correspondences and predictability. to appear. https://ling.auf.net/lingbuzz/004531.

Moravcsik, Edith A. 1978. On the case marking of objects. In Joseph H Greenberg (ed.), Universals of human language, vol. 4: Syntax, 249–289. Stanford: Stanford University Press.

Schmidtke-Bode, Karsten & Natalia Levshina. 2018. Assessing scale effects on differential case marking: Methodological, conceptual and theoretical issues in the quest for a universal. In Ilja A. Seržant & Alena Witzlack-Makarevich (eds.), Diachrony of differential argument marking (Studies in Diversity Linguistics 19), 463–489. Berlin: Language Science Press. http://langsci-press.org/catalog/book/173

Silverstein, Michael. 1976. Hierarchy of features and ergativity. In Robert M. W. Dixon (ed.), Grammatical categories in Australian languages, 112–171. Canberra: Australian National University.

Sinnemäki, Kaius. 2014. A typological perspective on Differential Object Marking. Linguistics 52(2). 281–313.




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.