Differences in Predictive Processing in Online Sentence Processing in Three-Year-Old Children

The ability to interpret speech as it unfolds in sentences is a complex skill that is essential to successful spoken communication. However, variability in sentence processing skills, such as predictive processing, can impair not only concurrent communication success but also future language development. Children with developmental language disorder (DLD) have known deficits in morphosyntax (Leonard, 2014), lexical representation (Sheng & McGregor, 2010b), and speed of processing (Leonard et al., 2007), but less is known about the impact of these impairments on processing sentences in real time during early stages of language development.  The present study examines individual and group differences in online predictive processing skills in 36-month-old children using eye-tracked simple transitive sentences of the structure Article-Agent-Action-Article-Target (e.g., The pirate chases the ship). Participants listened to the sentences while viewing pictures that corresponded with the sentential input in four different ways (i.e., Target, Agent-related, Action-related, Unrelated). Core Language Index (CLI) scores from the Clinical Evaluation of Language Fundamentals-Preschool 2  (CELF-P2) were used to from two sets of groups: 1) “high language” (n=33) and “low language” (n=22) groups using a median split of CLI scores (median = 102) and 2) “not at risk for DLD” (n=50) and “at risk for DLD” (n=6) group based on a CLI cutoff score of 85 used in clinical practice, which falls one standard deviation below the mean and suggests risk for DLD.

 Using eye movements as an index of online sentence processing, no individual or group differences were found in terms of prediction of the Target or locally-coherent activations of the Action-related item. These results indicate that three-year-old children at risk for language impairment are predicting highly expected items as well as entertaining alternative sentence representations simultaneously, indicating graded activations. These results contradict previous findings that adolescents with DLD do not make graded predictions (Borovsky et al., 2013). However, we found that children of higher language ability (as quantified by scores on the CELF-P2) completed significantly more fixations to the Agent-related picture. This finding suggests one way (i.e., Agent-related prediction) in which 36-month-olds’ online processing of sentences differs based on overall language skill that is inconsistent with accounts of an over-reliance on global interpretations in DLD later in development.