Sublexical, lexical and phonological short-term memory processes: Evaluating models of speech perception and short-term memory

The current research examined the processes involved in speech perception and the relation of speech perception to phonological short-term memory (pSTM). Contrary to traditional models which assume that sublexical processing necessarily precedes lexical processing (e.g., McClelland & Elman, 1986; Norris, 1994), several studies have demonstrated that patients may perform better on lexical than sublexical speech perception tasks (e.g., Miceli, Gainotti, Caltagirone, & Masullo, 1980). These findings have led to proposals of separate sublexical and lexical routes in speech perception (dual route models; e.g., Hickok & Poeppel, 2000). However, studies demonstrating these dissociations did not closely match the phonological similarity of targets and distractors or task demands between sublexical and lexical tasks. Experiment 1, which rectified these design problems, demonstrated a close correspondence between sublexical and lexical processing across several speech discrimination tasks for brain damaged patients and supported obligatory sublexical processing models of speech perception. Experiment 2 sought to provide converging evidence by tracking patients’ eye movements while they performed sublexical and lexical identification tasks. As in Experiment 1, there was evidence a) of a close correspondence between sublexical and lexical processing and b) for obligatory sublexical processing. However, there were some dependent measures that failed to support the obligatory sublexical processing framework. Dual route models additionally claim that pSTM depends on the brain regions involved in perception, including a sensory-motor interface region involved in rehearsal (e.g., Hickock & Poeppel, 2000). In contrast, traditional models of pSTM assume a phonological maintenance buffer separate from perceptual processes (Martin & Breedin, 1992). In Experiment 3 we found behavioral and structural neuroimaging (lesion-symptom mapping) support for the buffer approach, though no significant results were found in attempts to localize sublexical and lexical processing