In a recent blogpost, I promised that I’d pay more attention to the nanosyntactic approach if the authors look at more representative samples of the world’s languages, and it turns out that this is not difficult, because the fair open-access journal Glossa regularly publishes papers in this vein. A recent paper is van Baal & Don (2018), on universals of possessive pronouns, based on a sample of 50 languages.
But what is a “possessive pronoun”? The authors start with their own language Dutch, where the genitive (= adpossessive) forms of personal pronouns (e.g. mijn ‘my’, jouw ‘your’) are traditionally called “possessive pronouns” (as in the other Germanic and Romance languages). For these languages, there is actually no reason to distinguish between “possessive” and “genitive” pronouns. Strikingly, none of the 50 languages of their world-wide sample seems to have such a distinction (the two IE languages in the sample are Albanian and Nepali, which are like Dutch). So the first finding of their study is that a contrast between “genitive” and “possessive” pronouns is apparently very rare.
However, Czech is such a language, and Caha’s (2009) foundational thesis on the nanosyntax of case evidently has a strong influence on van Baal & Don’s thinking. They claim that in the universal case-hierarchy, possessive pronouns are higher than genitives in the functional sequence:
(1) nominative > accusative > genitive > possessive
This seems to be based on the contrast that occurs in many Slavic languages between adnominal genitive forms (e.g. Czech naš ‘our’) and adverbal genitive in genitive objects (e.g. Czech nás ‘of us’). In the Slavic (and some other older Indo-European) languages, it seems that the adnominal forms are often closely related to the genitive forms, which is of course not surprising given their semantic similarities – but this formal similarity is not found everywhere (cf. Russian moj ‘my’, contrasting with menja ‘of me’) To what extent this is a general tendency cannot be answered because very few languages have the relevant contrast.
Even more seriously, it is quite unclear how one would even define an “adverbal genitive” in a cross-linguistically applicable way. Adnominal possessive forms can be defined with respect to the core functions of possessive marking (expressing kinship, part-whole, and ownership relations), but there is no general notion of “adverbal genitive”. In the older Indo-European languages (and still marginally in Modern German), it makes sense to speak of “genitive objects”, because the case form used for adnominal possessors can also be used for certain verbal arguments – but by definition, these cases show the same forms for “genitives” and “possessives”. Thus, the contrast between “genitive” and “possessive” in (1) has no coherent definition. It seems that van Baal & Don don’t worry about definitions because they simply assume, along with many others in generative grammar, that grammatical categories are (innately given) natural kinds, and thus need not be defined.
But the paper does present an interesting finding concerning coexpression patterns: Of those 37 languages in their sample that do not have suppletive genitives, 7 have genitive forms that coexpress (“are syncretic with”) dative or accusative forms, and 26 have genitive forms that are based on the accusative/ergative and have additional marking, e.g. Lezgian abur ‘they (NOM)’, abur-u ‘they (ERG)’, aburu-n ‘their’. Van Baal & Don are not really interested in these latter asymmetric cases, so they give only a single example, but from my perspective, this is of course a general coding asymmetry that finds its explanation in frequency-based predictability: Genitives are generally less frequent than accusatives/ergatives, and for this reason they need longer coding. More generally, Blake’s (1994) case scale (made famous by Caha in generative circles) seems to find its easiest explanation in the correlation with frequency of use – cases lower on the scale are apparently used less frequently in all languages:
(2) Blake’s case scale: nominative > accusative/ergative > genitive > dative > locative > instrumental
But frequency-based predictability does not explain coexpression patterns, so for these, there would be no good competitor to Caha’s scale if it turned out to be cross-linguistically valid.
One clear pattern that van Baal & Don find strongly confirmed is that genitives/possessives tend to be similar to (and typically based on) accusatives/ergatives and datives, not to nominatives/absolutives. This is fully in line with the findings by Smith et al. (2016), who noted that “ABA suppletion” is not found in personal pronouns for the sequence “unmarked case – dependent case – oblique case”. Genitives/possessives are one subtype of oblique case, so to the extent that they look at different languages, van Baal & Don have confirmed Smith et al.’s findings. (Smith et al. use Distributed Morphology, while van Baal & Don seem to be closer to nanosyntax, but the difference does not seem to matter much.)
However, there are four languages that seem to contradict the prediction of the functional sequence. I wouldn’t worry so much about them, because the statistical trend seems clear, but van Baal & Don ARE worried, and they try to explain them away in a number of different ways (in one instance by assuming a zero-affix for which there is no other evidence). One of the ways is by noting that the offending “possessive forms” in Chamorro and Atalyal are actually person indexes (“bound pronominal elements”), not free personal pronouns, and that the claim about coexpression based on the scale in (1) is only made for pronouns that belong to the same paradigm. This should of course have been explained at the beginning, but the authors may not have been very clear about the difference between these two types of person forms, which I tried to distinguish carefully in Haspelmath (2013) (in footnote 14, they add another confusing concept, “agreement”, but they do not try to elucidate how it might relate to person indexes and free pronouns).
If the scale in (1) is supposed to be valid not only for free personal pronouns, but also for person indexes (as they suggest in footnote 13), then a study by Siewierska (1998) provides some sobering data: In Siewierska’s sample of 130 languages, there are almost as many languages with (mostly) identical subject and possessive markers as there are languages with (mostly) identical object and possessive markers (29 and 35 languages, respectively). Thus, it seems that coexpression patterns in person forms are rather different for person indexes and free personal pronouns, for reasons that are unknown.
In short, van Baal’s & Don’s paper on possessive pronouns in a sample of 50 languages reflects the great interest that nanosyntacticians take in typology, but also shows that it is difficult to come up with big new insights on fairly well-known phenomena, especially if one does not work with clearly defined concepts but hopes that somehow the existing terms map clearly to comparable phenomena. (And Glossa should require greater standards of reproducibility; the paper lists the languages in the appendix with references, but does not say for each language which type it represents, and there are no page numbers.)