2005 Fiscal Year Final Research Report Summary
The effects of argument structure frequency on verb prediction
| Project/Area Number |
16520229
|
|---|
| Research Category |
Grant-in-Aid for Scientific Research (C)
|
| Allocation Type | Single-year Grants |
|---|
| Section | 一般 |
|---|
| Research Field |
Linguistics
|
|---|
| Research Institution | University of Tsukuba |
|---|
Principal Investigator |
MIYAMOTO Edson T. University of Tsukuba, Graduate School of Humanities and Social Sciences, Lecturer, 大学院・人文社会科学研究科, 講師 (60335479)
|
|---|
| Project Period (FY) |
2004 – 2005
|
| Keywords | Sentence processing / Frequency / Word order |
|---|
| Research Abstract |
This research investigates how frequency of occurrence affects the way how Japanese readers comprehend sentences. Results can be summarized as follows.
1.First the low frequency of scrambled transitive constructions was established by counting the number of sentence-initial accusative NPs and determining that in the majority of the cases (over 98% or 4,537 instances) the subject was missing rather than coming after the direct object. Similarly, 32 native Japanese speakers preferred to leave out the subject when asked to write sentences starting with an accusative marked NP (84.3% or 210 completions), and only included a subject in a scrambled configuration in the remaining 15.7% of the time (similar numbers were obtained for dative marked NPs). According to these results, a sentence-initial object NP should rarely be expected to be followed by an overt subject. Thus, when such a subject does appear, a slowdown in reading times should be expected. Future research should investigate to wh
… More
at extent frequency can account for the behavioral results reported in previous literature.
2.Based on the number of hits returned by the internet search engine Google, 24 transitive verbs were selected so that 12 had low-frequency (2,140 to 350,000 hits, M=129,998) and 12 had high-frequency (1.22 million to 13.3 million hits, M=3.185 million). A sentence was created for each verb and 32 native Japanese speakers' reading times were measured for the canonical and the scrambled order of each sentence. Furthermore, the frequency with which each verb was preceded by either order was determined in terms of the number of hits on Google and on newspaper corpora (from the Mainichi Shinbun and the Nihon Keizai Shinbun). The results from Google and the corpus were well correlated (R=0.63,P<0.0001) ; moreover, both sets of counts yielded negative correlations with the reading times of each verb (R=-0.4,P<0.001). Hence, the higher the cooccurrence of a verb with NPs in a given order, the faster its corresponding reading times when preceded by NPs in such an order. More studies are needed, but at this point the results suggest that readers keep track of how often a given verb has been seen with NPs in a certain order and are able to use such information in order to facilitate processing. Less
|
