Naming Light Verb Constructions in Persian Speakers with Aphasia: A Linear Mixed Effects Model

Abstract

Introduction It is an open question whether highly frequent linguistic constructions such as compounds are processed as single units in the brain or are decomposed into smaller representations (Macgregor & Shtyrov, 2013; Wynne, Wheeldon, & Lahiri, 2018). One distinctive feature of Persian is that the majority of verbs (actions) are realized as light verb constructions (LVCs). A light verb may combine with a preverbal element such as noun, adjective, adverb, preposition or prepositional phrase to create an LVC (Mahootian, 2010). To give an example, the light verb /dadan/ meaning “to give” can be combined with a preverbal element like /hadar/ meaning “waste”. When /hadar/ and /dadan/ are used together, a new construction is derived that means “to waste”. Hypothesis It is hypothesized that a greater impairment in naming LVCs compared to the simple verbs in people with aphasia (PWA) might be observed due to the higher combinatorial demands in LVCs compared to simpler counterparts where such demand is not necessary. Methods Fifty-seven PWA were presented with 80 pictured actions (LVC= 63, simple= 17). All pictures were normed first with non-impaired Persian speakers in terms of rated visual complexity, imageability, AoA, familiarity, and name agreement (Nilipour, Bakhtiar, Momenian, & Weekes, 2016). Results and Conclusion Given the heterogeneity of naming performance in PWA, a Mixed-effect model (LME) including random intercept of subjects and items was used (see Bakhtiar, Jafari, & Weekes, 2017). The results showed that there was a significant difference between LVCs and simple verbs in naming accuracy (p<.01) when only number of syllables was input into the model as a covariate. Results showed greater impairment with LVC production compared with simpler verbs as predicted. However, when other variables such as level of education, age, aphasia type, frequency, number of syllables and AoA were entered into the model, the effect was no longer significant. The findings are more in the favor of the hypothesis that LVCs may be processed in a non-decompositional manner but caution should be taken against testing the hypothesis with evidence from fixed effects of interest only (see Crepaldi et al., 2011). Given that all LVCs used in our sample were semantically transparent, future studies could test whether LVCs with different degrees of semantic transparency reveal significant effects.