ИСТИНА

Kibrik et al. (1977) suggests that in Archi, Lezgic, assignment of nouns to the non-human genders 3 and 4 is at least partly based on referent size. The idea has been echoed in Corbett (1991) and Corbett & Fedden (2018) who suggest, more specifically, that, in Archi, big entities are assigned to Gender 3. For Lak, another East Caucasian language, in his discussion of the reconstruction of the proto-Lak gender assignment, Zhirkov (1955) proposes that historically Gender 3 was assigned to all animals, natural phenomena, round-shaped and also large objects. To our knowledge, for other Lezgic languages that are genealogically related to Archi and show a similar four-way classification (i.e. have two non-human genders in addition to feminine and masculine), no such effects have been reported. It is a priori unclear whether this is because size effects in these languages are weaker or absent altogether. The aim of this talk is to statistically test the hypothesis of referent size effects for Lak, Archi, Rutul, Tsakhur, Kryz, Budukh and Khinalug. What we want to see is whether Archi is indeed so different in this respect from its sister Lezgic languages, and whether it is similar to Lak, its more distant cousin and neighbour. We start with a theoretical discussion that helps us to put forward a falsifiable hypothesis. We argue that three different types of size effects can be expected, including absolute size effects observed in the lexicon at large (as formulated by Corbett and Zhirkov), categorial size effects observed within specific conceptual domain (such as animals, as argued by Kibrik) and, finally, flexible gender assignment based on referent being conspicuously smaller or larger than expectations (see e.g. Aikhenvald 2012, Di Garbo 2014), the latter functionally akin to diminutives and augmentatives. Next, we will propose a method to detect such effects in a statistically meaningful way. Unlike categories of shape ('is.round') or kind (e.g. 'is.human'), size is not based on a (nearly) categorical judgment, but is a relational scalar category based on judgments like ‘is.bigger’ or ‘is.smaller’. It is not immediately clear how to manually annotate referent size or establish thresholds for entities to be judged big or small in an absolute way. We build several datasets based on our own experimental data and the data from McRae et al. (2005) and Binder et al. (2016). To test for correlations between being small and Gender 4 and being large and Gender 3, we mapped the concepts of the datasets onto nominal vocabularies of the four languages. Our results are as follows. We observe size effects in gender assignment in Archi and probably Lak, confirming observations by Kibrik and Zhirkov, but not in other Lezgic languages, for which indeed no such observations have been made. In particular, we observe categorial size effects in Archi for some but not other tested conceptual domains. We conclude by suggesting that our methodology (including our datasets themselves) allows checking for size effects not only in the seven languages in the analysis, but also, in principle, for any language.