Electrophysiological signatures of phonological and semantic maintenance in sentence repetition.

Verbal short-term memory comprises resources for phonological rehearsal, which have been characterized anatomically, and for maintenance of semantic information, which are less understood. Sentence repetition tasks tap both processes interactively. To distinguish brain activity involved in phonological vs. semantic maintenance, we recorded magnetoencephalography during a sentence repetition task, incorporating three manipulations emphasizing one mechanism over the other. Participants heard sentences or word lists and attempted to repeat them verbatim after a 5-second delay. After MEG, participants completed a cued recall task testing how much they remembered of each sentence. Greater semantic engagement relative to phonological rehearsal was hypothesized for 1) sentences vs. word lists, 2) concrete vs. abstract sentences, and 3) well recalled vs. poorly recalled sentences. During auditory perception and the memory delay period, we found highly left-lateralized activation in the form of 8-30 Hz event-related desynchronization. Compared to abstract sentences, concrete sentences recruited posterior temporal cortex bilaterally, demonstrating a neural signature for the engagement of visual imagery in sentence maintenance. Maintenance of arbitrary word lists recruited right hemisphere dorsal regions, reflecting increased demands on phonological rehearsal. Sentences that were ultimately poorly recalled in the post-test also elicited extra right hemisphere activation when they were held in short-term memory, suggesting increased demands on phonological resources. Frontal midline theta oscillations also reflected phonological rather than semantic demand, being increased for word lists and poorly recalled sentences. These findings highlight distinct neural resources for phonological and semantic maintenance, with phonological maintenance associated with stronger oscillatory modulations.

Semantic and phonological contributions to short-term repetition and long-term cued sentence recall.

The function of verbal short-term memory is supported not only by the phonological loop, but also by semantic resources that may operate on both short and long time scales. Elucidation of the neural underpinnings of these mechanisms requires effective behavioral manipulations that can selectively engage them. We developed a novel cued sentence recall paradigm to assess the effects of two factors on sentence recall accuracy at short-term and long-term stages. Participants initially repeated auditory sentences immediately following a 14-s retention period. After this task was complete, long-term memory for each sentence was probed by a two-word recall cue. The sentences were either concrete (high imageability) or abstract (low imageability), and the initial 14-s retention period was filled with either an undemanding finger-tapping task or a more engaging articulatory suppression task (Exp. 1, counting backward by threes; Exp. 2, repeating a four-syllable nonword). Recall was always better for the concrete sentences. Articulatory suppression reduced accuracy in short-term recall, especially for abstract sentences, but the sentences initially recalled following articulatory suppression were retained better at the subsequent cued-recall test, suggesting that the engagement of semantic mechanisms for short-term retention promoted encoding of the sentence meaning into long-term memory. These results provide a basis for using sentence imageability and subsequent memory performance as probes of semantic engagement in short-term memory for sentences.

Localization of electrophysiological responses to semantic and syntactic anomalies in language comprehension with MEG.

Syntactically and semantically anomalous words encountered during sentence comprehension are known to elicit dissociable electrophysiological responses, which are thought to reflect distinct aspects of language processing. However, the sources of these responses have not been well characterized. We used beamforming analysis of magnetoencephalography (MEG) data to map generators of electrophysiological responses to linguistic anomalies. Anomalous words occurred in the context of a sentence acceptability judgement task conducted in both visual and auditory modalities. Time-frequency analysis revealed that both kinds of violations elicited event-related synchronization (ERS) in the delta-theta frequency range (1-5 Hz), and desynchronization (ERD) in the alpha-beta range (8-30 Hz). In addition, these responses were differentially modulated by violation type and presentation modality. 1-5 Hz responses were consistently localized within medial prefrontal cortex and did not vary significantly across violation types, but were stronger for visual presentation. In contrast, 8-30 Hz ERD occurred in different regions for different violation types. For semantic violations the distribution was predominantly in the bilateral occipital cortex and left temporal and inferior frontal regions, and these effects did not differ for visual and auditory presentation. In contrast, syntactic responses were strongly affected by presentation modality. Under visual presentation, syntactic violations elicited bilateral 8-30 Hz ERD extending into dorsal parietal and frontal regions, whereas effects were much weaker and mostly statistically insignificant in the auditory modality. These results suggest that delta-theta ERS reflects generalized increases in working memory demands related to linguistic anomaly detection, while alpha-beta ERD reflects specific activation of cortical regions involved in distinct aspects of linguistic processing, such as semantic vs. phonological short-term memory. Beamforming analysis of time-domain average signals (ERFs) revealed an N400m effect for semantic anomalies in both modalities, localized to left superior temporal and posterior frontal regions, and a later P600-like effect for syntactic anomalies in both modalities, widespread over bilateral frontal, posterior temporal, and parietal regions. These results indicate that time-domain averaged responses and induced oscillatory responses have distinct properties, including localization and modality dependence, and likely reflect dissociable and complementary aspects of neural activity related to language comprehension and additional task-related processes.