The effect of similarity-based interference on bottom-up and top-down processing in verb-final languages: Evidence from Hindi

Abstract

Sentence comprehension is known to be driven by both bottom-up integrative and top-down predictive processes. While integrative processes are known to be subject to working memory constraints, the impact of such constraints on top-down processing is less clear. Previous work has argued that verb-final languages provide rather weak and equivocal support for working memory constraints during bottom-up integrative processes. For these languages, top-down prediction has been shown to be more dominant. Here, we report a series of cloze completion and self-paced reading studies on a verb-final language, Hindi, to investigate if preverbal nouns with similar case marking lead to increased processing difficulty at the clause-final verb. Results show no effect of case similarity on reading times at the verb, implying that a solely bottom-up dependency completion process driven by memory constraints cannot explain these data. Another key finding is that verb prediction failures increase in configurations where preverbal nouns have similar case markings. Model evaluation suggests an explanation based on representation distortion – when the pre-verbal input is stored in memory, it probabilistically distorts to a non-veridical (or less accessible) memory representation, and this degraded representation of the context generates potentially faulty predictions of the upcoming verb. Together, the current work reveals two new insights: (i) Both top-down prediction and bottom-up integration assumptions are necessary to explain the reading data from verb-final languages, and (ii) top-down prediction is subject to working memory constraints due to representation distortion of prior sentence input stored in memory.