The topic of differential object marking (DOM), or more broadly differential argument marking, continues to be popular in different circles. The journal Linguistics had a special issue in 2014 with 11 papers, there is a recent LangSci volume on the diachrony of differential argument marking (coedited by my Leipzig colleague Ilja A. Seržant), and there is also a steady stream of MGG papers on the topic
– e.g. by Laura Kalin 2018 (dealing with Senaya Neo-Aramaic), by Theodore Levin 2018 (dealing with Palauan), by Coppe van Urk 2017 (dealing with Fijian), and by Pilipp Weisser & Laura Kalin 2018 on coordination and differential object case-marking.
The term differential object marking was coined by Bossong (1982; 1985), and his findings and ideas were beautifully summarized in his 1991 paper.
Basically, Bossong claims that the explanation for the universal tendencies of differential object marking (favouring overt coding of animate, definite and otherwise referentially prominent object arguments) lies in the interaction of the semantic-syntactic role (“slot”) and the referential prominence of the filler (“the specific semantic nature of the NP”; see Bossong’s §2). This explanation is very similar to the explanations given by authors such as Silverstein (1976), Moravcsik (1978), Comrie (1978), Dixon (1979), which all appeal to the special, unexpected nature of referentially prominent objects in one way or another. In a recent (still unpublished) paper (Haspelmath 2019), I have proposed that this generalization, as well as quite a few other generalizations about split and alternating argument coding, can be explained by frequency-induced predictability, like many other grammatical patterns. This basically follows the explanations advanced in much earlier work since the 1970s (and in fact since the 19th century, cf. Filimonova 2005).
In view of the widespread consensus that the DOM tendencies can be explained functionally, it may seem surprising there is a substantial amount of generative work that simply ignores these explanations (e.g. Torrego 1998; Baker & Vinokurova 2010; Ormazabal & Romero 2013). This seems to be based on the tacit assumption that every fact of every language needs to be explained with reference to UG – which in turn is based on the misunderstanding that description and explanation must be the same thing in linguistics. Bossong (1991) emphasizes that to the extent that the patterns of individual languages reflect universal tendencies, they should be explained in general terms – but he also says:
“First of all, typological comparison has to make clear that the phenomenon under study is not isolated but typologically related to other phenomena in many distant and different languages. Only then does it make sense to look for causal explanations” (Bossong 1991: 164).
So the explanation comes after the comparison and the establishment of universal tendencies, not at the same time as the description. Explanation must be in terms of UG if other (more likely) explanations fail, but there is no reason why language-particular descriptions should be in terms of UG vocabulary. To be sure, we need some kind of UG to acquire a language, but we are of course perfectly capable of creating language-particular grammars on the basis of language-particular categories.
So what do we learn from the recent MGG papers on DOM? I may find time to discuss some of the others later, but here I will only deal with T. Levin’s (2018) paper on Palauan. The author argues that DOM in Palauan should be explained as a “last-resort licensing” phenomenon, not as a “morphological” phenomenon (where the different object nominals have the same syntactic features but have different exponents), and not as due to movement either (as in the increasingly popular dependent-case approach). Here is an example (with the accusative preposition er, which only appears on human or singular specific nouns):
|A Sally a menguiu er se el hong.|
|DET Sally TOP read.IPFV ACC that L book|
|‘Sally is reading that book.’ (Levin’s (3a))|
The main argument against the dependent-case approach is that Accusative marking is limited to imperfective constructions in Palauan, while perfective constructions show no object flagging (only object indexing), even though there is no reason to think that the object does not move in perfective clauses. The main argument against the “morphological” account is that it is non-restrictive: It makes it appear completely accidental that DOM is found only in imperfective clauses – whereas this follows from Levin’s analysis, because only imperfective constructions have an extra AspectP that prevents licensing of the object in the usual way. (I did not understand how it is motivated that there is an extra phrase in imperfectives – this seems to be largely theory-internal.)
Here I have three critical questions:
(1) How sure are we that it is NOT an accident that imperfective constructions show special flagging and agreement patterns in a number of languages? As T. Levin notes, progressives (and other imperfectives) often derive from nominalizations, and this of course makes them prone to inherit special argument-marking patterns. Why should this be built into synchronic grammars? And note that perfectives may also arise from special nominalized constructions, as recently emphasized by Anderson (2016) (and briefly discussed earlier on this blog). It is particularly well-known that Russian lost the old Slavic past tense and, as a result, lacks person indexing on its past-tense forms. This is generally seen as an accident, so why couldn’t the Palauan argument-marking split be equally accidental?
(Recall that if this is an accident, then the “boring” morphological account could be right; and in the spirit of Bossong and the other functionalists, one could then attribute the observed regularity to functional-adaptive forces, rather than building it into the formal analysis.)
(2) T. Levin actually adopts the “morphological” analysis himself – for one idiosyncratic aspect of Palauan DOM. The object-marking preposition is not found on plural object nominals. There is no explanation for this – hence this has to be encoded by “brute force”. But if brute force is available anyway (and not, for example, excluded by learnability considerations which are often invoked by Chomskyans), why not use it everywhere? It seems to me that the main justification for this is that otherwise the paper would be too short and too boring (because the basic generalization, that the Accusative marking is found only in imperfective clauses when the object is human or specific and singular, is very easy, and thus very elegant, to state).
(3) The paper has the highly general title “On the nature of differential object marking”, but in concluding section, T. Levin suddenly retreats to a very modest position:
“Of course, just because the present analysis is correct for Palauan does not mean that every language with a DOM system must be analyzed in similar terms. Additional research is needed to determine if other languages can and should be accorded similar analyses or if DOM is better understood as an umbrella term for a number of phenomena.”
But if there is no universal claim, is this still a paper about the “nature” of DOM (rather than merely a paper about a Palauan pattern that could alternatively be analyzed in much more elegant terms)? Moreover, if there is no universal claim, does the argument against the “morphological” analysis (that it is non-restrictive) still go through? It seems to me that in the concluding section, it turns out that T. Levin’s proposal is not restrictive either.
Thus, I do not see a serious competitor here for Bossong’s explanation of the differential object marking universals in terms of the interaction of the semantic role and the referential prominence of the filler nominal (what I call “role-reference associations”). It seems that T. Levin’s paper (like most other papers by MGG authors) has not heeded Bossong’s advice that explanations of universals must come after the empirical establishment of the universals. These authors typically look at a single language and propose an analysis of it in universal terms, hoping that it will somehow extend to other languages, and that the whole enterprise will eventually converge on a single set of natural kinds.