Manabu KUROSAWA
(
kurosawa@p.u-tokyo.ac.jp
)
Graduate School of Education, University of Tokyo
7-3-1 Hongo Bunkyo-ku, Tokyo 113-0033 Japan
Tetsuo KAWAHARA
(
tkwhr@p.u-tokyo.ac.jp
)
Graduate School of Education, University of Tokyo
7-3-1 Hongo Bunkyo-ku, Tokyo 113-0033 Japan
This research investigates the process of metaphor comprehension in Japanese. In our experiment, we recorded the subjects' reaction time to understand metaphor sentences primed by either the base term or the target term (or both or neither). To separate the effects of syntactic factor and positional factor, we used inverted metaphor sentences in Japanese (e.g. Kangoku dearu, Sigoto wa.). The comparison of mean reaction times of 4 prime type conditions showed no advantage for base condition over target condition. These results apparently replicated the result of Gentner & Wolff(1997) 's experiment 1, which they insisted are evidence of superiority of their Alignment-first models of metaphor comprehension over Glucksberg et al.(1997)'s Abstraction-first models. However, the pattern of correlations of mean RTs among these conditions suggested that although priming effects by base and those by target are quantitatively equal, the mechanisms underlying these effects are not identical.
How can we understand a metaphor (e.g. ``My job is a jail.'')? Numerous studies have been done to elucidate the process of metaphor comprehension throughout the history of cognitive science. In this paper, we investigate the on-line processing of target (in this example, `my job') and base (in this example, `a jail') in metaphor comprehension.
Gentner & Wolff(1997) 's Alignment-first models claim that process of a metaphor comprehension begins with finding commonalities between base and target, and then aligning of representations between them. However, alignment-first models match the relations between elements as well as the elements themselves; this allows alignment-first models to place non-identical elements in correspondence by virtue of their common role in the representational structure.
On the other hand, Abstraction-first models assume processing begins with the base. In these models, an interpretation is constructed by (1) finding or deriving an abstraction associated with the base; (2) projecting this abstraction from the base to the target; (3) verifying the information in the target by matching the selected abstraction with the target representation.
To examine the validities of these models, experiments with priming methods have been widely used. In these experiments, some parts or all of a metaphor are presented visually as a prime, before entire metaphor is presented on CRT display. The task of participants is to hit the key as soon as they understand the metaphor. The reaction latency is measured as dependent variable. Independent variable is prime type. Metaphors are preceded by one of four types of prime: the target(the metaphor that substituted base with blank line), the base, neither the target nor the base (blank lines), or both target and base (the entire metaphor).
From the viewpoint of alignment-first model, Gentner &Wolff(1997) claimed as follows; In the first stage of processing, all identical predicates in the target and base representations are matched in parallel without regard for structural consistency. Alignment-driven model predicts that people should be faster to interpret a metaphor when it is preceded by the base than when it is preceded by none (but not faster than when it is preceded by the target). In contrast, if people begin comprehension with just the base, as abstraction-driven models claim, they should be faster to interpret a metaphor when it is preceded by the base than when it is preceded by the target.
Gentner &Wolff(1997)'s experiment 1 showed that comprehension of a metaphor when preceded by the target is not faster than when preceded by the base. But they also reported in experiment 4 that there are conditions at which metaphor comprehension preceded by the base is faster than comprehension preceded by the target: relational similarity between base and target is low and the meaning of the base that is used in the metaphor is highly conventional.
Kurosawa &Kawahara(1999) replicated Gentner &Wolff(1997)'s experiment 1 with Japanese metaphor sentences. These were translated from English sentences used in Gentner &Wolff(1997). In Japanese, target of a metaphor is placed at the beginning of the sentence, and base of a metaphor which is followed by dearu that functions as copula is placed at the end of the sentence. Between base and target, particle wa is placed which denote that the beginning word is subject of the sentence. For example, ` `A job is a jail.'' is transcribed in Japanese as ``shigoto wa kangoku dearu,'' in which shigoto means a job and kangoku means a jail.
In Kurosawa &Kawahara's experiment, participants were faster to interpret a metaphor when it is preceded by the base (i.e. -- wa kangoku dearu) as well as when it is preceded the target (i.e. sigoto wa -- dearu) than it is preceded by nothing (i.e. -- wa -- dearu). But they also tested the patterns of correlation among four conditions to examine if the advance is derived from only one source. The result was that correlation between both condition and target condition did not reach significant level, but correlation between both condition and base condition was significant. They argued that the result suggested the possibility that although base condition and target condition are primed equally in latency, the mechanisms that primed each condition are not same.
But we cannot conclude only from this result that target and base play different roles in on-line processing of metaphor, because the syntactic factor (base or target) and positional factor (position that base or target is placed) are always confounded in past experiments. In these experiments, participants might re-read from the beginning of the visually presented metaphor after they have read its prime. If so, the position that the prime is presented will affect the result through perceptual identification process. So, in this paper we examine if the effect of position in visual presentation can affect the result of priming experiment using Japanese inverted sentences of metaphor. In Japanese, inverted metaphor sentence (e.g. Kangoku dearu, Shigoto wa.) holds the original meaning.
The hypotheses are as follows; if the priming effect is attributed to the syntactic factor, latency in base condition (e.g. Kangoku dearu, -- wa.) will correlate with that in both condition, and no significant correlation will be found between both condition and target condition. In contrast, if the priming effect is attributed to the positional factor, latency in both condition will correlate not with that of base condition but that of target condition.
Participants were 20 payed undergraduates of the University of Tokyo.
32 metaphors in Gentner &Wolff(1997) were used. Four graduate school students ( who were not participated in following experiment ) translated the metaphors, and the first author chose the equivalents from the alternatives so as to sound as usual expressions.
Participants were run on on DOS-based computers individually. The session began with ten practice trials. Before the first trial, participants were told that their task was to interpret metaphors and that they might use the prime (s) preceding each metaphor to give them a head start on their interpretations. Then they were told that they should hit space bar as soon as they had it well formulated, and after that described the interpretation orally to the experimenter. The experimenter recorded the speech. For these trials and the subsequent experimental trials, the method was similar to Genter &Wolff(1997)'s experiment 1. First, a line of asterisks that has the same positions as the following metaphor characters appeared on the screen. After 500ms, the line was replaced by the prime. The prime was hold for 1500ms, followed by blank screen for 100ms. Then the entire metaphor was presented just below where prime had been appeared till participants hit space bar. After that, participants described the interpretation orally, and then experimenter resumed the program.
First, mean RT and SD of four conditions for each participants were calculated, and responses that was far 2 SD from mean RT of each condition was removed. Second, mean RT and SD of all participants' mean RT in both condition were calculated, and participants whose mean RT is far 2 SD from mean RT for all 20 participants (2 participants) were also removed. Basic statistics are shown in Table 1. One-way repeated-measures ANOVA indicated an overall effect of prime type across both subjects (Fs(3, 51) = 20.46, p < .001), and items (Fi(3, 93) = 10.40, p < .001). Tukey's HSD procedure was used to test comparisons among the means. Only both condition resulted in faster interpretation than any other conditions, but no other comparisons did not reach significant level.
| Prime type | Both | Base | Target | Blank |
|---|---|---|---|---|
| RT | 2027.5 | 2732.3 | 2829.7 | 3090.8 |
| SD | 823.6 | 1005.1 | 1039.7 | 941.1 |
Then correlation analysis among four conditions were executed. RT in base condition correlated significantly with that in target condition (r = .38, p < .05). Correlation between base condition and both condition was marginally significant (r = .35, p < .10).
But correlation between target condition and both condition did not reach significant level. RT in blank condition correlated only with that in target condition. The correlation matrix is shown in Table 2.
| Both | Base | Target | Blank | |
|---|---|---|---|---|
| Both | ||||
| Base | .35+ | |||
| Target | .24 | .38* | ||
| Blank | .13 | .27 | .43* |
Note: * = p < .05, + = p < .10
Interpretation when primed with entire metaphor was faster than primed with any other. Facilitation when primed with targets did not reach significant level, and also no facilitation was found in base condition. It seems that inversion made the task more difficult to interpret, and facilitation as a whole was reduced in this experiment. So, it is reasonable to say that this result did not show the change in processing of base or target when they were presented in inverted sentences.
In addition, the pattern of correlations in RT data among four conditions was quite similar in Kurosawa &Kawahara(1999)'s previous study. Correlation between base condition and both condition did not reach significant level, but very near ( p < .0523). Like Kurosawa &Kawahara(1999)'s result, significant correlation was found between base condition and target condition, and there was no significant correlation between target condition and both condition. This pattern of correlations reveals that though sentences were presented in inverted order and difficulty of task made no facilitation in RT data for base and target conditions, usual processing of metaphors was executed in this experiment. This result can be explained reasonably to assume that base condition and both condition use a component to process metaphors in common (so a metaphor which takes a long time in processing when primed with base also takes a long time when primed with whole metaphor), but target condition and both condition share no such common component.
Anyway, the pattern of correlations among four conditions was almost consistent even when positions of base and target were switched. So, it is entirely fair to say that difference in correlation pattern between base condition and target condition is derived not from the position they appear but from their syntactic role in prime.
In this experiment and our previous study, correlations between both condition and base condition are quite different from correlations between both condition and target condition. These results seem to suggest that processing of metaphor when primed with base is different from processing when primed with target.
By using inverted metaphor sentences in Japanese as material, we confirmed that priming effects which were observed in previous experimental studies of metaphor were derived from syntactic factor, not from positional factor.
And, correlation analysis among prime type conditions revealed that processing of metaphor when primed with base is different from processing when primed with target.
This paper was presented at The 2nd International Conference on Cognitive Science and The 16th Annual Meeting of the Japanese Cognitive Science Society Joint Conference (27-30 July 1999, International Conference Center, Waseda University, Tokyo, Japan).