How does animacy affect word order in a VOS language? Takuya Kubo1, Hajime Ono2, Mikihiro Tanaka3, Masatoshi Koizumi4, Hiromu Sakai1 1Hiroshima
University, 2Kinki University, 3Showa University, 4Tohoku University 1Corresponding author:
[email protected]
Experiment
Abstract Although language production mechanism is often assumed to be universal, the range of languages investigated so far is quite limited. We conducted a picture description task to clarify how conceptual accessibility affects word order selection and how incremental language production is in Kaqchikel which is Object-Subject order language spoken in Guatemala. Although Kaqchikel speakers tend to produce passive sentence with animate patient, a tendency to mention animate entity before inanimate entity was not observed. These results support a weakly incremental view on language production.
Method Participants: 60 native Kaqchikel speakers. Task: Participants verbally describe the target pictures in a simple sentence. Design: 1 × 3 (patient animacy: Human, Animal, and Object). Materials: 24 target pictures manipulating patient animacy. 18 fillers depicting intransitive event.
Introduction p Conceptual accessibility • The ease with which the referent of a sentence can be retrieved from memory ([1]). • Previous works have found that more accessible entity tends to be mentioned earlier than less accessible entity as follows(e.g., Animacy: animate > inanimate, Givenness: given > new).
1. Direct hypothesis • Conceptual accessibility correlates only with the serial order among constituents ([2]). • More accessible entity tends to be assigned an earlier position in a sentence.
2. Indirect hypothesis • Conceptual accessibility correlates with grammatical function ([1]). • More accessible entity tends to be assigned a higher grammatical function along with Noun Phrase Accessibility Hierarchy (NPAH: SU > DO > IO… ). (e.g. more passives are produced with animate patient).
p Incrementality and conceptual accessibility • Radically incremental view ・More accessible entity tends to be lexicalized and produced earlier and speaker
Human-Human
Human-Object
Filler
Prediction p Direct hypothesis predicts … ・Animate patient will tend to be mentioned earlier than inanimate patient. ・VOS order would be dis-preferred with Object patient.
p Indirect hypothesis predicts … ・Animate patient will tend to be assigned higher grammatical function. ・Passive structure: Human patient > Object patient.
Result
gradually construct syntactic structure ([3]).
• Weakly incremental view ・Conceptual accessibility drive speakers to choose a syntactic structure in which
SVO VOS VSO Passive Anti-passive Other Total
accessible concept can serve as syntactic subject ([4]).
Research Questions ü Does OS language have a tendency to mention more accessible entity before less accessible entity? ・Typical subject and object are animate and inanimate respectively. ・Does Animate entity tend to follow inanimate entity in OS language?
Human-Animal
Procedure: Experiment was conducted by a native Kaqchikel speaker in a quiet room in Antigua, Guatemala. Analysis: Mixed effect logistic regression.
HumanHuman 175 (70.6%) 26 (10.5%) 3 (1.2%) 11 (4.4%) 22 (8.9%) 11 (4.4%) 248 (100%)
HumanAnimal 202 (75.4%) 47 (17.5%) 5 (1.9%) 6 (2.2%) 3 (1.1%) 5 (1.9%) 268 (100%)
HumanObject 156 (59.1%) 100 (37.9%) 1 (0.4%) 2 (0.8%) 1 (0.4%) 4 (1.5%) 264 (100%)
Total 533 (68.3%) 173 (22.2%) 9 (1.2%) 19 (2.4%) 26 (3.3%) 20 (2.6%) 780 (100%)
p Voice • More passive structure with human patient than object patient (p<.05).
ü What is the relation between conceptual accessibility and the word order in OS languages? ・Pullum (1977) proposed principles for linearization in world’s languages [5]. Fundamental principle: NPs are linearized along with NPAH. Additional principle: Subject NP is assigned the final position in VOS languages.
p Constituent Order • More SVO order than VOS order (68.3% vs. 22.2 %). • More VOS order with inanimate object than other two conditions (p<.001).
*
100% 75%
Kaqchikel (OS language)
50%
p Advantage of Kaqchikel ・Word order alternation is independent of voice alternation.
0%
p Grammatical function
Human-Human
・Head-marking and ergative language ・Grammatical person and number of NPs agree with prefixes of verb.
p Constituent order ・Canonical: VOS [6] ・Most frequent: SVO ・Limited: VSO
p Voice system
Ø Examples a. X-∅-u-ch’öy
ri ak’wal ri xtän (VOS)
COM-erg.3s-abs.3s-slapped
the boy
the girl
The girl slapped the boy.’ the girl
x-∅-u-ch’öy
COM-erg.3s-abs.3s-slapped The girl slapped the boy.’
ri ak’wal (SVO) the boy
Acknowledgement: The authors thank Yoshiho Yasugi, Lolmay P. García, Feliberto P. Majzul, Juan E. A. Sián for their support for conducting experiments in Guatemala, and other project members for valuable comments. This research was supported by Grant-inAid for Scientific Research (S) #22222001 (PI: Masatoshi Koizumi). Reference:[1] Bock & Warren (1985). Cognition, 21, 47–67. [2] Branigan & Feleki (1999). In Proceedings of the 21st Cognitive Science Society Conference, Vancouver. [3]Levelt (1989). Speaking: From intention to articulation. Cambridge, MIT Press. [4] Christianson & Ferreira (2005). Cognition, 98, 105-135. [5] Pullum (1977). Syntax and semantics: Vol.8, Grammatical relations, New York, Academic Press, 249-277. [6] Matzar, Cotzajay, & Tuiz (1999). Gramatica del Idioma Kaqchikel. Proyecto Linguistico Francisco Marroquin. La Antigua Guatemala. [7] Tanaka, Branigan, McLean, & Pickering (2011). Journal of Memory and Language, 65, 318-330.
Human-Animal
Human-Object
Discussion p Significant tendency to assign human patient to subject function. ・More passive use with human patient than object patient.
p We did not find crucial tendency to mention animate entity earlier than inanimate entity in Kaqchikel.
・active, passive, anti-passive. ・active and passive voice is relevant in our experiment.
b. Ri xtän
VOS SVO
25%
・Although SVO order is most frequently in active sentence, VOS order is more often when patient is object. ・This result does not support direct effect of conceptual accessibility. ・This tendency follows Pullum’s assumption of linearization.
p Difference among SO language and OS language. ・Recent work on Japanese (SO language) suggests indirect effect also influences the serial word order in a way such that animate entity tends to be assigned not only subject function but also earlier position ([7]). ・Kaqchikel speaker has a tendency to assign more accessible entity with higher grammatical function, but not in the earlier position of the sentence.
What distinguish Kaqchikel from SO languages?
・ Since Kaqchikel is different from SO language, basic word order of OS language may not be compatible with NPAH directly. ・ NPAH show directly impact on grammatical function assignment but only indirectly influence linear order assignment ・ Our results support weakly incremental view of sentence production. 25th Annual CUNY Conference on Human Sentence Processing, Poster Session 3, Friday, March 16