How to talk about role coding: cases and index-sets, nominative/accusative, subject/object, agent/patient

As an Athabaskanist working in Russia, Andrej Kibrik has long been interested in the head-marking/dependent-marking typology: Athapaskan languages mark the major clause roles by person prefixes on the verb, whereas the languages of Russia (Uralic, Turkic, Nakh-Daghestanian, etc.) almost all show extensive case-marking and limited person-indexing. In Kibrik (2012) (a paper that was originally presented at the LENCA conference in Kazan), he now urges linguists working on head-marking languages to draw the consequences from the parallelism between head and dependent marking and to use terminology familiar from case systems, i.e. “nominative/accusative” and “ergative/absolutive”, rather than “subject/object” or “agent/patient”.

Thus, the example in (1) is OK, but the examples in (2) and (3) are not good.

(1) Navajo (p. 220)
‘I carry him.’

(2) Lavukaleve (Terrill 2003: 227)
Aira la ali na a-o-le.
woman(F) art.F man(M) art.M 3SG.M.OBJ-3SG.F.SUBJ-see
‘The woman saw the man.’

(3) Lakota (Pustet & Rood 2008: 336)
‘You guys hear me.’

I think that Kibrik is right that case-labels as in (1) would be entirely appropriate for distinct person indexes (see Haspelmath 2009: 510). But what exactly is the difference between case labels like “nominative/accusative” and argument-class (“grammatical-relation”) labels like “subject/object”? According to Kibrik (p. 225), “grammatical relations” are notions that “serve to capture the behavioral properties of arguments”, whereas case labels are more appropriate for role coding. He says that it would be “odd to suggest that Latin case desinences mark grammatical relations” (rather than semantic roles), and that nobody would talk about the “Latin subject case” or the “Latin object case”. But here I am not so sure. The “subject” notion in particular correlates very highly with the nominative case, more so than the “agent” notion does. Latin cases first point hearers to broad argument-classes, and only in a next step can they infer the roles (also on the basis of the verb meaning). So saying that “subject” and “object” are relevant only for behavioural properties is not quite right.

I find the generalized notion of role alignment (e.g. Bickel 2011) highly useful: The extremely diverse roles of verbs are grouped into a small number of broad classes not only by coding elements such as cases and person-indexes, but also by behavioural processes. So we can talk about accusative alignment of case-marking, but also about accusative alignment of relativization, and both give us reasons to talk about argument-classes (or “evidence for grammatical relations”). Subject and object are just two other traditional labels for the argument-classes {S, A} and {P}, much like nominative and accusative. Thus, I do not think that it really matters which labels one uses. Kibrik emphasizes that grammatical relations are language-specific, but of course so are the argument-classes as evidenced by coding elements (cases and person-indexes). Personally I would prefer the case-derived labels, because such labels exist not only for S/A vs. P alignment (accusative alignment), but also for S/P vs. A alignment (ergative alignment), and also because the terms sound more technical and precise. But this is also a matter of taste.

The reason why people often use “subject/object” for person-indexes is of course that “nominative/accusative” are far better known as case-labels than as generalized alignment labels, and informally they have long been used as equivalents of “agent/patient”.

But the main reason why I don’t think we can legislate a particular terminology is that the argument-classes are quite different across languages, and even though similar terms for similar phenomena are transparent, they should not mislead us to think that the phenomena are identical. We can call the Japanese case-markers -ga and -o “nominative” and “accusative”, but in Japanese double-nominative constructions they do not behave as expected. Such idiosyncrasies are normal, and they remind us that the Japanese Nominative is not the same case as the Russian Nominative.

The most surprising aspect of Kibrik’s paper for me was that he chose Navajo to illustrate nominative/accusative person indexing, because in Navajo, the actual shapes of the two kinds of person forms are mostly not different – they differ in position. This is similar to Bantu languages (e.g. ex. (4)), and quite different from, say, Arabic.

(4) Swahili
a. tu-li-wa-ona
‘we saw them’
b. wa-li-tu-ona
‘they saw us’

(5) Arabic
a. ra’ay-ta-haa
‘you saw her’
b. ra’a-t-ka
‘she saw you’

Where the nominative indexes are clearly different from the accusative indexes (as in Arabic), I think using the case labels is very intuitive, but in the case of Navajo and Swahili, it is less so. After all, we would not normally speak about “nominative” and “accusative” word-order marking in languages like Mandarin.

In a recent paper (Haspelmath 2012+), I have proposed the term “index-set” for a paradigm of person forms that differ in their shape. So Arabic can be said to have a nominative index-set and an accusative index-set, but Navajo has only a single index-set. Since the S-indexes are in the same positional slots as the A-indexes, we can still say that there is nominative/accusative alignment of indexing in Navajo, but only alignment of index order, not alignment of index form.


Bickel, Balthasar. 2011. Grammatical relations typology. In Jae Jung Song (ed.). The Oxford handbook of linguistic typology, 399-444. Oxford: Oxford University Press.

Haspelmath, Martin. 2009. Terminology of Case. In Andrej Malchukov & Andrew Spencer (eds.). The Oxford Handbook of Case, 505-517. Oxford: Oxford University Press.

Haspelmath, Martin. 2012. Argument indexing: a conceptual framework for the syntax of bound person forms. Ms., Max Planck Institute for Evolutionary Anthropology.

Kibrik, Andrej A. 2012. What’s in the head of head-marking languages? In Pirkko Suihkonen, Bernard Comrie, & Valery Solovyev (eds.). Argument structure and grammatical relations: a cross-linguistic typology, 211-240. Amsterdam: Benjamins.

Pustet, Regina & David S. Rood. 2008. Argument dereferentialization in Lakota. In Mark Donohue & Søren Wichmann (eds.). The typology of semantic alignment, 334-356. Oxford: Oxford University Press.

Terrill, Angela. 2003. A Grammar of Lavukaleve. (Mouton Grammar Library). Berlin: Mouton de Gruyter.

One thought on “How to talk about role coding: cases and index-sets, nominative/accusative, subject/object, agent/patient

  1. Thank you, Martin. I consider Andrej’s proposals even more controversial for reasons both conceptual and empirical. In particular, one might find Andrej’s claim that the function of head-marking and dependent-marking are similar or the same, compelling in case of Navajo, which does not have (core) case marking, but this claim becomes much less evident when the whole picture of cross-linguistic diversity in this domain is taken into account. The view advocated by Andrej is particularly problematic with respect to the “double-marking” languages, especially those which I call “disharmonic”, i.e. where the (broadly understood) “alignment” of cases does not match that of the personal affixes. And even if we limit ourselves to languages without core case marking, I am just wondering how the logic proposed by Andrej would work, e.g., for Yimas and other languages where form and/or position of personal markers depends not (solely) on their role properties, but also on their referential features (the so-called “hierarchical” systems).

    It would be interesting for me to have a look at the published version of this paper. I read it in the manuscript a year ago and sent Andrej my comments expressing sceptisims about many of his empirical and theoretical claims.

    Nevertheless, I find this paper useful in that it draws our attention to conceptual and terminological issues which should be reflected upon with closer scrutiny.

Comments are closed.