Björn Hansen, Edyta Jurkiewicz-Rohrbacher & Zrinka Kolaković
The talk aims to show how corpora can be used to study fairly complex phenomena. We will base the discussion on the example of constraints on clitic climbing in Bosnian, Croatian and Serbian (BCS). Descriptively speaking, clitic climbing (CC) “refers to constructions in which the clitic is associated with a verb complex in a subordinate clause but is actually pronounced in constructions with a higher predicate” (Spencer & Luís 2012: 162). An example of CC out of an infinitival complement is given in (1) where the clitical pronoun ga ‘him’ is realised in the second position of the matrix clause (Wackernagel position); in other cases, however, CC does not take place as in (2) where the clitic ih stays in the complement clause.
(1) | Milan | ga2 | mora1 | vidjeti2. |
Milan | him.ACC | must.3PRS | see.inf | |
‘Milan must see him.’ | Stjepanović (2004: 179f) |
(2) | Bojim1 | se1 | testirati2 | ih2. |
afraid.1prs | REFL | test.INF | them.ACC | |
‘I am afraid to test them.’ | hrWaC v2.2 |
Although clitics in Bosnian, Croatian and Serbian (BCS) have attracted considerable attention in the syntactic literature (cf. Franks & King 2000, Browne 2014, or Bošković 2004),the syntactic conditions and constraints for CC are seriously understudied in comparison to e.g. Czech (e.g. Junghanns2002). There are only very few studies on CC in BSC: Stjepanović (2004), Aljović (2004, 2005) mainly deal with theoretical considerations based on a small selection of construed examples.
Jurkiewicz-Rohrbacher et al. (2017a, 2017b), Hansen et al. (2018) are the first descriptions of CC in BCS based on empirical investigations. Basing on the data obtained from massive web corpora {bs, hr, sr}WaC (Ljubešić & Klubička 2014), the raising-control dichotomy of matrix predicates is shown to be a relevant factor of CC. Apart from that, it is found out that reflexivity plays a major role. Kolaković et al. (accepted), on the other hand, tackle the question of register as a relevant factor by comparing results from Forum subcorpus of hrWaC v2.2, Croatian Language Repository (Ćavar&BrozovićRončević 2012) Croatian National Corpus (Tadić 2009) while examining the same types of matrix predicates.
First, the talk presents the results of the corpus based and corpus driven studies mentioned above, discusses in detail the particular steps of a corpus approach, ranging from the formulation of queries, coping with tagging errors, to the statistical analysis of the data. Second, it will show how these results feed into a major psycholinguistic experiment recently carried out in Croatia (7 experiments x 40 participants = 280 participants). The logistic regression mixed models based on data from the speeded yes-no grammaticality judgment tasks with OpenSesame free software provide the additional evidence for constraints on CC.