University of Glasgow Department of Psychology	Johannes Gutenberg - Universitaet, Mainz Psychologisches Institut

Several models concerned with sentence comprehension of human beings exist. They assume either a serial or a parallel processing of the information and one prominent experimental method to test hypotheses is the measurement of reaction times. In this work the constituent comparison model (CCM; Carpenter & Just, 1975) will be examined for English and German native speakers. This is a serial model whereby subjects have to decide as quickly as possible whether or not a sentence is true for a given picture. (All subject do the task in their maternal tongue.) According to this model sentences and pictures are coded in an abstract, propositional format each consisting of a number of constituents. After the encoding these constituents are compared one after the other and the total number of constituent comparisons determines the reaction time. This model was assumed to be independent from language. However, response pattern of German native speakers seem to differ from pattern of English native speakers. This is suggested by previous experiments, albeit, without providing strict evidence. It took German speakers longer to verify affirmative sentences than English speakers (see Fig. 1, below). The difference for negative sentences (German speakers were slower than English speakers) was marginally significant and response latencies were virtually identical for embedded sentences. The CCM was expected to be valid, which was confirmed for either language but no identical model can be assumed as reaction times differed for some conditions (affirmative sentences). For each language a linear model fitted the data well and the CCM was supported by analyses of error rates and variances of response latencies. The differences in response latency between languages resulted from the German language system being morphologically richer than the English one, i.e. there are more grammatical cues (e.g. word order, gender, number etc) that need to be processed thus increasing response latency for German speaking subjects. The number of grammatical cues that must be processed additionally in German stays the same across all experimental conditions but the total number of grammatical cues increases with sentence type (affirmative, negative, embedded). Therefore, the impact of this difference on response latency diminishes with increasing redundance, i.e. more information in total but a constant amount of additional information in German. This phenomenon is interpreted as a different encoding process in English and German hence leaving the comparison process according to the CCM the same and valid for both languages. Some problems concerning the propositional coding remain. One condition was operationalised in two ways but different response latencies were observed. This suggests that redundant phrases were marked and henceforth not processed in order to save effort. It keeps unclear how this is achieved. Furthermore, it is questionable whether the CCM can be applied equally to affirmative and negative sentences on one side, and embedded sentences on the other side for the latter one include a relation that is not considered in the CCM.

Figure 1: Mean response latencies for both language groups. The difference is significant (p < .05) for the TA condition and marginally significant (p < .1) for the FA condition. Note that conditions FE and TEN result in the same number of constituent comparisons.

For further details you may check out some of these excerpts or email me:    m a i l <at> d - k o e s t e r <dot> d e

1. About the model
2. Language dependency
3. Some results
4. Discussion extract

(c) Dirk Koester; first posted January 06/01                                                                                                                 last modified June 04/03