Length effects in an OV language with Differential Object Marking and mixed head direction

Length effects in an OV language with Differential Object Marking and mixed head direction
Pegah Faghiri (Sorbonne Nouvelle Paris 3), Barbara Hemforth (CNRS/Paris-Diderot) & Pollet Samvelian (Sorbonne Nouvelle Paris 3)

pegah.faghiri@univ-paris3.fr
The “long-before-short” (LbS) preference observed in SOV languages, e.g. Japanese, Persian

and Basque, challenges the universality of availability-based theories of word order preferences supported by incremental models of speech production, put forward to account for the “end-weight” preference observed namely in German and English. Based on data from Japanese, Yamashita & Chang (2001) propose a unified but language-specific accessibility-based account for both preferences. Their rationale is that 1) longer constituents are lexically richer and hence more salient than shorter ones, while being more costly to process and plan; 2) language-specific properties may modulate the sensitivity of a production system to conceptual vs. formal factors. With respect to word order, Japanese speakers focus more on conveying meaning than on sequencing forms, contrary to e.g. English speakers, given that Japanese has fewer syntactic constrains (e.g. flexible word order & null pronouns) and that the shift does not occur in the postverbal domain where the verb is shown to exert strong influence.

Recent studies Japanese (Tanaka et al. 2011) and Basque (Ros et al. 2015) reject a conceptual account of LbS in favor of Hawkins’ dependency-based distance-minimizing model (1994, 2004), which predicts mirror-image preferences in (strictly) VO and OV languages. Contributing to this discussion, we provide data on Persian, another SOV language with flexible word order and null pronouns, but head-initial in NP, PP and CP. Persian has verbal agreement with subject and Differential Object Marking (realized by =r ) triggered by definiteness.

Previous corpus and experimental studies on the relative order between the DO and the IO in the preverbal domain (Faghiri & Samvelian 2014, Faghiri et al. 2014) have found a clear LbS preference, not predicted by Hawkins’ model (due to Persian’s mixed head direction). Moreover, they show that the effect of length depends on the degree of determination (or definiteness) of the DO: r - marked (definite) DOs prefer OD-OI-V regardless of length, (existential) indefinite DOs prefer OD-OI- V even more with OD≥IO, and bare DOs show a less strong preference for OI-OD-V with OD≥IO. It is argued that these preferences conform to the conceptual account, given that the fact that the effect of length and definiteness converge.

Our data on the relative order between the subject and the DO support these results. Firstly, our corpus data reveal that the degree of definiteness of the DO remains highly relevant: only r - marked DOs occur in OSV order (at the overall rate of 4.6%). Secondly, we manipulated the relative length and animacy (of the subject) following a 2x2 design in a controlled experiment with r -marked animate DOs (20 items, ex.1), using a web-based sentence completion questionnaire (56 participants). We found (cf. Fig.1) a highly significant effect of animacy (p<.001), OSV order was more frequent (12% vs. 4.4%) for inanimate subjects (1b) than for animate ones (1a), but no effect of length contrary to dependency-based length-minimizing accounts predictions (e.g. Gibson 2000, Hawkins 2004). Meanwhile, the conceptual hypothesis can provide a satisfactory account of these preferences: 1) animacy is a known conceptual factor to influence ordering preferences (e.g. Branigan & Feleki 1999, Kempen & Harbusch 2004, Branigan et al. 2008); 2) it is highly likely that the joint effect of definiteness and animacy undermines the
contribution of length to the conceptual
accessibility of the DO.