Fauconnier & Verstraete (2014) examine “Differential A Marking” (DAM, where ergative flagging is different in prominent and less prominent nominals), compare it with “Differential O Marking” (DOM, where accusative flagging is restricted to prominent nominals), and conclude that the two are not each other’s “mirror image”. Whatever the explanation of (better-known) DOM, the explanation for DAM must thus be different.
They note that two previous approaches claim that DOM and DAM are mirror-image phenomena, both in terms of observed cross-linguistic patterns and in terms of their explanation. The expectation-based (or “markedness”-based) approach well-known from Silverstein, Dixon, Comrie and Aissen derives the cross-linguistic patterns from the expectation that P arguments have low referential prominence (in terms of person, animacy, definiteness and topicality). The semantic approach of Næss (2007) derives the patterns from the idea that prototypical patients are highly affected, and indefinite and inanimate patients are not as highly affected. Both these approaches see agents (and A arguments) as the polar opposites of patients (and P arguments), so a similar explanation would hold for DAM.
But it has been known for some time that DAM does not show exactly the same patterns as DOM. Specifically, while Special P Marking is common with locuphoric (1st/2nd person), definite, animate and topical P arguments, Special A Marking occurs primarily with allophoric (3rd person) and focal A arguments. It is rare to find special ergative marking for inanimate nominals, and special ergative marking for indefinite nominals is almost never found. Thus, de Hoop & Malchukov (2008) proposed somewhat different explanations for the two kinds of phenomena, and Fauconnier & Verstraete are making this point even more forcefully.
They argue that agents and patients are not each other’s mirror image semantically: Agenthood (consisting of the properties of instigating and transmitting an action) is not really gradable, while patienthood (consisting of the properties of being an endpoint and of being affected) is gradable. Thus, the agent role is not readily compatible with inanimate referents – this would explain the fact that inanimate agents are sometimes marked in a special way, and indeed some languages completely forbid inanimate A arguments (so these languages cannot say ‘The wind broke the window’, and they must resort to ‘The window broke because of the wind’ or similar).
These are very interesting semantic and cross-linguistic observations, and I am sure that there is something to this. And the formal asymmetry between DOM and DAM is indeed intriguing.
However, I felt that two important pieces were missing in Fauconnier & Verstraete’s story. First, they do not consider person forms at all, apparently for no deep reason. But with (locuphoric, sometimes also allophoric) person forms – both independent pronouns and person indexes – the frequent absence of special ergative forms is particularly striking, and it seems to be a fairly good mirror image of the frequent presence of special accusative forms with person forms (also found in English). Since in actual discourse, most arguments are expressed by person forms, the Silversteinian mirror-image explanation thus seems to work fairly well for most tokens of argument expression.
Second, they do not differentiate between “Differential A Marking” and “Special A Marking”, where by the latter I mean a situation where the A has an overt ergative marker in some situations, but no marker in other situations. It is well-known that the vast majority of DOM cases are of this sort – “differential” P marking is overwhelmingly a contrast between an overt accusative flag and zero, as in Spanish Ø vs. a. (Partitive/accusative alternations as in Finnish, or genitive/accusative alternations as in Polish seem to be quite rare, and seem to occur under rather different circumstances than typical cases of Special P Marking.) Fauconnier & Verstraete mention a range of languages that have two different ergative markers for animate and inanimate A arguments. Indeed, these would not be explainable under an expectation-based approach, which can explain a contrast between overt marking and zero (or between short marking and long marking), but not a marking contrast with no length difference.
Looking at the examples that Fauconnier & Verstraete discuss, we see that the zero/overt contrasts (Special A Marking) occur primarily when focality is the trigger for A marking, where animacy plays no clear role. (I found this the weakest aspect of their paper: They note that focality-based DAM is frequent, occurring in 14 of their sample languages, but the semantically-based explanation they offer does not easily extend to information-structure phenomena.)
Another interesting set of phenomena they discuss is the ban on inanimate (or non-motive) A arguments, which is found in 19 of their languages. (Non-motive participants are inanimates that do not possess an internal source of energy; motive inanimates are natural forces like fire, wind, rain.) This is indeed plausibly explained in semantic terms, though I was wondering whether the semantic “clash” that they talk about should be seen in terms of semantic incompatibility (so that violating it would require a “stretch”), or perhaps simply in terms of low likelihood. According to the expectation-based approach, low likelihood can be a source of marking asymmetries. I was also wondering why Fauconnier & Verstraete do not mention the well-known phenomenon that indefinite A arguments are uncommon and in fact not allowed in some languages. They emphasize that special A marking does not occur with indefinite A arguments, but given the emphasis they put on the ban on inanimate A arguments, one would have liked to see them discuss the ban on indefinite A arguments.
Thus, given my own bias toward the expectation-based approach (not in terms of “markedness”, but in terms of frequency-induced expectedness, cf. Haspelmath et al. 2014 and earlier related work), I don’t think that things look so bad for this kind of explanation. Even though Fauconnier & Verstraete are of course right that topicality and focality are not exact mirror images, in terms of actual language use it would seem to be rare for nominals to be both topical and focal, so statistically they are fairly mirror-image-like. And as I noted above, the split between person forms and non-person forms is also fairly mirror-image-like. The main asymmetry between Special P Marking and Special A Marking seems to be that A arguments tend to be avoided (rather than specially marked) when they are inanimate or indefinite. The former has a plausible semantic explanation, as noted by Fauconnier & Verstraete. The latter is still awaiting a good explanation, as far as I am aware.
de Hoop, Helen & Malchukov, Andrej. 2008. Case-marking strategies. Linguistic Inquiry 39(4). 565–587.
Fauconnier, Stefanie & Jean-Christophe Verstraete. 2014. A and O as each other’s mirror image? Problems with markedness reversal. Linguistic Typology 18(1). 3–49.
Haspelmath, Martin, Andreea Calude, Michael Spagnol, Heiko Narrog & Elif Bamyacı. 2014. Coding causal-noncausal verb alternations: a form-frequency correspondence explanation. Journal of Linguistics 50(3). 587-625.
Næss, Åshild. 2007. Prototypical transitivity. Amsterdam: Benjamins.