Word structure and decomposition effects in reading

Abstract

Theories on the processing of compound words differ on the role attributed to access to individual constituents. These theories are mostly based on empirical evidence obtained in experimental settings that could induce artificial effects normally not occurring in natural processing. In this study we investigated the processing of compounds as compared to noncompound complex words in Italian through a reading task with eye movement recording. We included both head-initial and head-final compounds, in order to test whether the position of the head may influence the reading process. After ruling out the effects of length and frequency, we observed that pseudocompounds (i.e., words with a segment homograph to a real word in the leftmost part) elicited longer total reading times than all other types of complex words, including compounds. Furthermore, head-final compounds elicited longer total reading times than head-initial compounds. The results suggest that a word structure resembling a compound may induce longer processing, presumably related to unexpected morphological structures. The results also converge with previous evidence that in some cases there is a higher processing costs for head-final as opposed to head-initial compounds, possibly indexing a reanalysis of the stimulus in order to correctly assign the constituent properties. However, a deeper analysis restricted to compounds revealed a more complex scenario where several variables interact with headedness (namely, first and second constituent frequency, compound frequency, and compound length), and future studies are needed to discriminate among possible interpretations. Overall, our findings suggest that longer reading times are related to solving incongruities due to noncanonical structures, rather than to morphologically complexity per se.

Keywords:

• Morphology
• Morphological decomposition
• Compound words
• Compound headedness
• Eye movements.

Preliminary results of this study were presented at the 51st Annual Meeting of the Academy of Aphasia (Lucerne, 2013). The study was ideated and elaborated jointly by the three authors. Data preparation and collection were conducted by the first and third authors. Data analysis was conducted by the first author. We thank Maria Concetta Morrone (CNR, Pisa) and Pier Marco Bertinetto (Laboratorio di Linguistica, SNS, Pisa) for technical facilities, Chiara Bertini for technical support, and Petar Milin for the very useful comments. We also thank an anonymous reviewer for the help in the interpretation of statistical analyses.

This study has been supported by the University of Padua [grant number STPD11B8HM_004 to Carlo Semenza].

Notes

¹ As part of the syntactic object, the determiner was included in the region of interest. Fixations landed on the determiner index the processing of the target word as well, and removing them would lead to loss of important information.

² Given that the study in Marelli and Luzzatti (2012) included also high-transparency items, one may ask why we do not replicate exactly their results here with a different distribution of transparency. We believe that there are many statistical reasons that can explain the discrepancies. For example, the mixed-effect model used by Marelli and Luzzatti (as the one used here) assumes linearity: If nonlinear effects are present for highly transparent compounds (more numerous in our data), some nuances may fail to be evidenced.

³ Another interpretation of whole-word frequency effect has been suggested in the literature (Baayen, Wurm, & Aycock, 2007), but a discussion of this alternative explanation is not relevant to the aims of the current study.