Filler-gap dependencies in non-canonical word orders in Japanese

Edson T. Miyamoto¹ and Shoichi Takahashi²
etm@is.aist-nara.ac.jp, fwgg5639@mb.infoweb.ne.jp
1 Nara Institute of Science and Technology
2 Kanda University of International Studies

We propose that the processing of scrambled, non-canonical, word orders in Japanese involves determining a filler-gap dependency for each scrambled constituent and its canonical position. In support of this proposal, the results of a self-paced reading experiment, a corpus count and three probe recognition experiments are reported.

The present data share similar characteristics with previous results on filler-gap processing in English. First, there is greater difficulty to process gaps that are far away from their antecedent (Gibson, 1998; Just and Carpenter, 1980). In a self-paced reading experiment, the slow-down at the gap was more accentuated when its distance to the filler (the scrambled NP-accusative) was increased by the intervention of an adjunct XP -- the slow down at NP-dative in (1d) was greater than in (1c), whereas no difference was found between the canonical accusative NPs in (1a) and (1b) (interaction: F1(1,31)=7.95, P<0.01; F2(1,19)=5.03, P<0.05). Moreover, the NP-dative in (1d) was read significantly more slowly than the NP-accusative in (1b) (F1(1,31)=6.63; P<0.05; F2(1,19)=4.65; P<0.05). There were no significant differences at the ditransitive verb (Fs<1) contrary to predictions by head-driven models.

(1)

a.

Canonical: Subject XP NP-dative NP-accusative ditran-verb aux.

b.

Canonical: Subject NP-dative XP NP-accusative ditran-verb aux.

c.

Short scrambling: Subject XP NP-accusative NP-dative <gap> ditran-verb aux.

d.

Long scrambling: Subject NP-accusative XP NP-dative <gap> ditran-verb aux.

The result cannot be explained by the relative frequency of the constructions compared given that a corpus count conducted in two daily Japanese newspapers (Mainichi Shinbun 1991-1995; Nihon Keizai Shinbun 1990-1996) yielded no interaction between the frequencies of the four conditions above. The ratio between sentences of types (1a) and (1b) was the same as the ratio between (1c) and (1d). (Corpus frequencies including only sentences containing the verbs used in the behavioural study were as follows: (1a) 1202, (1b) 46, (1c) 565, (1d) 21.)

Second, in accordance with previous results in English (Bever and McElree, 1988; MacDonald, 1989), the antecedent of a scrambled constituent in Japanese is reactivated facilitating its recognition at the end of the sentence. A previous study argued against this facilitation (Nakayama, 1995); however, Nakayama did not control for the position of the probed word in the sentence. In a series of three experiments, we replicated Nakayama's experiment controlling for the probe position and we found a facilitation after the scrambled transitive sentences. Moreover, the facilitation is independent of the fact that scrambled constituents are in focus.

References

Bever, T. G., and B. McElree. (1988). Empty categories access their antecedents during comprehension. Linguistic Inquiry, 19, 35-43.

Gibson, E. (1998). Linguistic complexity: Locality of syntactic dependencies. Cognition, 68, 1-76.

Just, M. A., and P. A. Carpenter. (1980). A theory of reading: From eye fixations to comprehension. Psychological Review, 87, 329-354.

MacDonald, M. C. (1989). Priming effects from gaps to antecedents. Language and Cognitive Processes, 4, 35-56.

Nakayama, M. (1995). Scrambling and probe recognition. In R. Mazuka, and N. Nagai (Eds.), Japanese Sentence Processing. Mahwah: Lawrence Erlbaum.