On double accusative marking

Recently on LINGTYP, a question was raised, which can be paraphrased as such: are there languages in which the P of a transitive clause is marked by two co-occurring flags (‘case-markers’), e.g., ACC1-N-ACC2 (or, presumably, N-ACC1-ACC2 or ACC1-ACC2-N)? After quite a bit of back-and-forth, it turns out that such situations seem to be extremely rare, while the co-occurrence of indexing and flagging seems to be much more common. Based on the discussion, and a closer look at some other sources, it turns out that only three scenarios are known to lead to double flagging. Interestingly (if not surprisingly), they all involve Differential Object Marking (DOM).

The first scenario involves the grammaticalization of an innovative accusative marker in the context of Differential Object Marking, which co-occurs with a pre-existing accusative marker. Such cases include Old Armenian (as pointed out by Wolfgang Schulze):

NOM.PL osker-k` ‘bones’

ACC.PL osker-s [-def]

z-osker-s [+def]

 It seems that the older accusative marker did not mark definiteness, and the innovative accusative developed precisely in this function.

Ge’ez (Semitic) also came up in this context. But matters here are a bit more complicated. The relevant construction involves the dative preposition la– (‘to’). First of all, there were other strategies for overtly marking definiteness, including (1) demonstrative pronouns, (2) possessive indexes, and (3) argument indexing on the verb. In fact, the DOM strategy seems to have almost always entailed differential object indexing (DOI) on the verb as well, which is why descriptive grammars tend to call such constructions ‘proleptic,’ e.g.,

rəʔināhu la-ʔəgziʔəna
see.1PLA.3SGMP to-lord.1PL
‘We have seen our lord’ (lit. ‘We-have-seen-him to-our-lord’).

In such constructions, the direct object typically does not bear an overt accusative case marker (-a), although grammars mention sporadic examples. But at least some of them seem to involve construct state nouns, which look like accusative-marked nouns. So Ge’ez wouldn’t be a very good example of double accusative flagging.

Other instances, which didn’t come up in the discussion, were mentioned long ago by Georg Bossong (1991): DOM markers can fuse with their hosts, in which case, the ‘unanalyzable whole’ can be marked with an unreduced or less reduced form of the same accusative marker. Bossong mentions Upper Engadinian (Romance) ad a-me, and Central/Southern Baluchi accusative –a followed by –ra, ‘a phonetically less reduced for of the same ACC ending.’ Bossong goes on to discuss rather remarkable cases of triple marking which seem to involve the same sort of process. However, I don’t know whether in all these cases the older constructions still marked typical DOM parameters (specificity, definiteness, animacy, or topicality). Giorgio Iemmolo (p.c.) says that in Bossong’s Romance examples, the accusative marker is totally fossilized.

This leads us to the second scenario. In Huangshui Chinese, the new accusative marker xa can co-occur with the old accusative marker pa. The grammaticalization of xa (from topic marker to accusative marker) appears to result from a fairly complex interplay of information-structural and word-order considerations, as well as, possibly, language contact (Dede 2007, Iemmolo 2013+). Importantly, Huangshui Chinese presents us with an instance of a language in which each of two synchronic DOM markers can occur alone or together.

The third and final scenario involves double marking in the context of language contact. Specifically, an L2 accusative marker can occur with an L1 accusative marker. This seems to be the case in a number of Quechuan languages.

Ulcumayo Quechua (Sanchez 2011)

Algo gati-pu-n a un niñu-ta.
Dog follow-DIR-3SG to a boy-ACC
‘The dog follows a boy’

Lamas Kechwa (Sanchez 2011)

Kawa-yka-n a ese niñitu-ta.
Look-PROG-3SG to that boy-ACC
‘(S/he) is looking at that boy’

At first glance, this looks like a rather typical case of ‘double marking’ in bilingual speech (Muysken 2008). But Sanchez shows that such constructions occur in conditions similar to those in which DOM tends to be triggered (+specificity, mainly). This makes it likely that we are not dealing with a simple case of matter replication, i.e., of an accusative marker, but rather of pattern replication, in which DOM applies to the entire repertoire of the bilingual speaker (Matras 2009).

So, all in all, three ways of getting double accusative flagging: 1. an innovative DOM marker alongside an older non-DOM accusative (Old Armenian); 2. an innovative DOM marker alongside an older DOM accusative marker (Huangshui Chinese); 3. as the result of language contact (leaving aside the thorny question of borrowing vs. code-switching) in which the extension of DOM itself (and not just the accusative marker) is extended to other parts of a bilingual repertoire (Quechuan). Interestingly, language contact might be a big part of the story even in the case of (2). Are there other ways?

This is interesting for a few reasons that I can think of offhand. For one thing, it shows how important it is to make a careful distinction between flagging and indexing, since they are distinct phenomena and are distributed differently across languages.

Second, it shows that double accusative flagging is rare, while the combination of distinct types of coding means is not so rare (e.g., flagging + indexing, and flagging + linear order or indexing + linear order also seem to be common enough). That’s reasonable enough, since indexing, flagging, and linear order are different, and particular coding strategies can develop for a number of different reasons. The rarity of double-flagging would seem to stem from fairly obvious reasons of economy, but there’s a catch with this take on things: DOM tend to emerge well after older flags have gone out of existence or have lost their earlier function. So here’s the big question: why aren’t there more languages like Old Armenian?

Third, double accusative flagging, where it occurs, seems to always involve the emergence or re-emergence of DOM, whether by cross-linguistically recurrent pathways of grammaticalization or through language contact. This isn’t surprising, but other scenarios are imaginable. Since both language contact and grammaticalization have well-motivated functional explanations (at least in some people’s views), it means that we have the beginnings of a small step towards understanding why language structures are the way they are.


Bossong, Georg (1991). Differential object marking in Romance and beyond. In: D. Kibbee& D. Wanner (eds.) New Analyses in Romance Linguistics. Amsterdam/Philadelphia: Benjamins. 143-170.

Dede, Keith (2007). The origin of anti-ergative xa in Huangshui Chinese. Language and Linguistics 8(4): 863–881.

Iemmolo, Giorgio (2013+). From topic marker to object marker: a study on the grammaticalisation of Differential Object Markers. Under review.

Matras, Yaron (2009). Language Contact. CUP.

Muysken, Pieter (2008). Functional Categories. CUP.

Sanchez, Liliana (2012). Convergence in syntax/morphology mapping strategies: Evidence from Quechua–Spanish code mixing. Lingua 122(5): 512-528.


6 thoughts on “On double accusative marking

  1. Great Post , I actually found it useful where you started In such constructions, the direct object typically does not bear an overt accusative case marker (-a), although grammars mention sporadic examples. But at least some of them seem to involve construct state nouns, which look like accusative-marked nouns. So Ge’ez wouldn’t be a very good example of double accusative flagging.

  2. Ok, that’s really interesting. I guess it raises questions about the possibility/frequency of DOM markers that don’t develop from adpositions, since such markers (if there are any) might not have a problem co-occurring with pre-existing cases (e.g., if they’re non-adpositional topic markers). I wonder if this is what’s going on in the Masalit data shown by Christa Wolf in “Case in Africa” (p. 65), in which there’s a word “hayir” that co-occurs with accusative person markers and is glossed as “?” She also mentions languages in which case is marked by tonal differences, such as Fur, in which accusative person markers have low tone, but can be co-marked by the accusative suffix -si, which is typically found as an overt accusative marker on nouns.

    I was thinking that what you say about adpositions not occuring with accusative case outside of IE is surprising, since it’s a shibboleth of grammaticalization studies that some adpositions develop from verb to adposition, which would lead one to expect that at least some nouns occuring with such adpositions might be accusative-marked (e.g., V+acc > Adp+acc). But I don’t know if any such constructions became DOM markers (nor have I looked).

  3. Why aren’t there more languages like Old Armenian? The reason seems to be quite simply that there are few languages where adpositions govern a non-genitive case – outside older Indo-European, this does not seem to occur. Older Indo-European languages are quite unusual in that even fairly grammatical prepositions (with basic meanings such as ‘on’, ‘at’, ‘from’ or ‘with’) often combine with nouns in a non-nominative case, and this case may be distinct from the possessive (genitive) case, e.g. it may be accusative. To those of us trained in Latin, Greek, Russian etc., this is perfectly “normal”, but it actually seems to be very rare, and exotic Indo-European feature. Another highly unusual feature of Indo-European, seen in Old Armenian, is that often the nominative case is overtly marked, like the accusative case, so that both are equally long. Again, this hardly occurs outside IE, as far as I can see.

    • I believe that Old Georgian had adpositions governing all the cases.

      • Yes, Old Georgian is uncannily similar to old Indo-European in a number of ways. I think Gamkrelidze & Ivanov were quite justified in speculating about causes for these similarities.

        • I actually wonder what will happen in languages with very impoverished case systems where the genitive and the accusative will be syncretic, so the acc could be assigned by an adposition by virtue of this. But I don’t have any specific examples in mind.

Comments are closed.