Connectives as Processing Signals: How Students Benefit in Processing Narrative and Expository Texts

Abstract

Many young readers fail to construct a proper mental text representation, often due to a lack of higher-order skills such as making integrative and inferential links. In an eye-tracking experiment among 141 Dutch eighth graders, we tested whether coherence markers (moreover, after, because) improve students' online processing and their off-line comprehension of narrative and expository texts. Eye-tracking results show that connectives lead to faster processing of subsequent information as well as shorter rereading times of previous text information. Connectives also trigger readers to make regressions to preceding information. These findings indicate that connectives function as immediate “processing instructions.” Furthermore, all students performed better on local comprehension tasks (i.e., bridging inference questions) after reading texts containing connectives than after reading texts without these markers. These findings apply to both text types and to all students, regardless of reading proficiency. This study highlights the importance of comprehensible texts in which implicit coherence relations are avoided.

Notes

¹ Vmbo is divided into four levels ranging from mainly theoretical (high level, Dutch vmbo-tl) to mainly vocational training (low level, Dutch vmbo-bb). Dutch educational publishers develop textbooks for each level. In this study, the two highest levels are collapsed, because they mainly concern theoretical education (high level, vmbo-tl and vmbo-gl). The two lowest levels are also collapsed, because they mainly concern practical education (low level, vmbo-bk and vmbo-bb).

² We checked our coherence manipulations by using the Coh-Metrix Common Core Text Ease and Readability Assessor (T.E.R.A.) that analyzes texts on five components: narrativity, syntactic simplicity, word concreteness, referential cohesion, and deep cohesion. The component “deep cohesion” shows how well the events, ideas, and information of the whole text are tied together by measuring the different types of words that connect different parts of a text. Analyses show that the deep cohesion scores of the implicit versions ranged from 19% to 56%, whereas the deep cohesion scores of the explicit versions ranged from 96% to 100%. We concluded that our coherence manipulations were effective.

³ Fixations were determined by an algorithm that restricts fixations to data points within 30 pixels. The viewing time in a region was computed as the time from the beginning of the first fixation until the end of a last successive fixation in a region. A minimal fixation length of 50 ms was allowed in the analysis.

⁴ For none of the processing times the distribution of the raw data was comparable with the normal distribution, which is a common phenomenon for processing data (cf. Yan & Tourangeau, 2008). Before the analysis, log-transformations were performed on the data to meet the normality requirements of linear modeling.