Decoding the temporal dynamics of spoken word and nonword processing from EEG

McMurray, Bob; Sarrett, McCall E.; Chiu, Samantha; Black, Alexis K.; Wang, Alice; Canale, Rebecca; Aslin, Richard Ν.

doi:10.1016/j.neuroimage.2022.119457

Cited by 12 publications

(15 citation statements)

References 78 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Although the stimuli were matched for duration, it is worth noting that other aspects of the stimulus sets could explain this observed difference, such as their acoustic properties. However, based on current understanding of the phonological and semantic processes underpinning spoken word recognition during wakefulness (Gaskell & Mirkovic, 2016;McMurray et al, 2022), it is possible that the sleeping brain has better access to meaning when presented with verbal relative to non-verbal memory cues, with an enhanced spindle response reflecting engagement of multi-level decoding pathways.…”

Section: Discussionmentioning

confidence: 99%

Delineating memory reactivation in sleep with verbal and non-verbal retrieval cues

Guttesen

Gaskell

Cairney

2023

Preprint

View full text Add to dashboard Cite

Sleep supports memory consolidation via the reactivation of newly formed memory traces. One way to investigate memory reactivation in sleep is by exposing the sleeping brain to auditory retrieval cues; a paradigm known as targeted memory reactivation (TMR). To what extent to acoustic properties of memory cues influence the effectiveness of TMR, however, has received limited attention. We addressed this question by exploring how verbal and non-verbal memory cues affect oscillatory activity linked to memory reactivation in slow-wave sleep. Fifty-one healthy adult males learned to associate visual stimuli with spoken words (verbal cues) and environmental sounds (non-verbal cues). Subsets of the verbal and non-verbal cues were then replayed during sleep, alongside previously unheard control cues. For a subset of the participants, the voice of the verbal cues was mismatched between sleep and learning. Memory cues (relative to control cues) prompted an increase in theta/alpha and spindle power, which have been heavily implicated in sleep-associated memory processing. Moreover, verbal memory cues were associated with a stronger increase in spindle power than non-verbal memory cues. There were no significant differences between the matched and mismatched conditions when analysing verbal memory cues in isolation. Our findings suggest that verbal memory cues may be more effective than non-verbal memory cues for triggering memory reactivation in sleep, as indicated by an amplified spindle response.

show abstract

Section: Discussionmentioning

confidence: 99%

Delineating memory reactivation in sleep with verbal and non-verbal retrieval cues

Guttesen

Gaskell

Cairney

2023

Preprint

View full text Add to dashboard Cite

show abstract

“…The idea is to train a set of linear classifiers to classify sets of neural data collected under different conditions. This provides insight into how the topographical maps collected with EEG sensors display a pattern of activity can discriminate between stimulus features [36; 44; 45; 46], and also reveals how this discrimination evolves over time [36; 47; 48; 46; 41].…”

Section: Resultsmentioning

confidence: 99%

Neural basis of melodic learning explains cross-cultural regularities in musical scales

Pelofi

Rezaeizadeh

Farbood

et al. 2022

Preprint

View full text Add to dashboard Cite

Seeking exposure to unfamiliar experiences constitutes an essential aspect of the human condition, and the brain must adapt to the constantly changing environment by learning the evolving statistical patterns emerging from it. Cultures are shaped by norms and conventions and therefore novel exposure to an unfamiliar culture induces a type of learning that is often described as implicit: when exposed to a set of stimuli constrained by unspoken rules, cognitive systems must rapidly build a mental representation of the underlying grammar. Music offers a unique opportunity to investigate this implicit statistical learning, as sequences of tones forming melodies exhibit structural properties learned by listeners during short- and long-term exposure. Understanding which specific structural properties of music enhance learning in naturalistic learning conditions reveals hard-wired properties of cognitive systems while elucidating the prevalence of these features across cultural variations. Here we provide behavioral and neural evidence that the prevalence of non-uniform musical scales may be explained by their facilitating effects on melodic learning. In this study, melodies were generated using an artificial grammar with either a uniform (rare) or non-uniform (prevalent) scale. After a short exposure phase, listeners had to detect ungrammatical new melodies while their EEG responses were recorded. Listeners' performance on the task suggested that the extent of statistical learning during music listening depended on the musical scale context: non-uniform scales yielded better syntactic learning. This behavioral effect was mirrored by enhanced encoding of musical syntax in the context of non-uniform scales, which further suggests that their prevalence stems from fundamental properties of learning.

show abstract

“…Since then, follow-up studies have employed similar decoding paradigms with good success (Bae & Luck, 2019). Moreover, other studies have shown that decoding on even a single-trial level is possible (Bayet et al, 2020;Correia, Jansma, Hausfeld, et al, 2015;McMurray et al, 2022;Trammel et al, 2023). Beach et al (2021) applied similar techniques to single-trial MEG data, which share many of the same qualities as EEG data.…”

Section: Scalp-recorded M/eegmentioning

confidence: 99%

“…Traditionally with MVPA, the mean activity of the voxel is used as classifier input, which corresponds to a zero‐order polynomial. More temporally precise measures, however, like iEEG (Rhone et al., submitted) or scalp EEG (McMurray et al., 2022), can take into account fluctuations in the neural signal, which may be informative for stimulus decoding. For example, we can consider the slope of the voltage across a certain time window or a quadratic function that encompasses an ERP component.…”

Section: Introductionmentioning

confidence: 99%

“…For example, we can consider the slope of the voltage across a certain time window or a quadratic function that encompasses an ERP component. The majority of previous work has been developed to use mean voltage (a zero‐order polynomial; Bae & Luck, 2018, 2019; Bayet et al., 2020), but some has used higher‐order polynomial fits (McMurray et al., 2022). In order to examine trade‐offs across all options, we compared results using the mean voltage (zero‐order polynomial), linear fit (first‐order polynomial), and quadratic fit (second‐order polynomial).…”

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Decoding speech sounds from neurophysiological data: Practical considerations and theoretical implications

Sarrett,

Toscano

2023

Psychophysiology

Self Cite

View full text Add to dashboard Cite

Machine learning techniques have proven to be a useful tool in cognitive neuroscience. However, their implementation in scalp‐recorded electroencephalography (EEG) is relatively limited. To address this, we present three analyses using data from a previous study that examined event‐related potential (ERP) responses to a wide range of naturally‐produced speech sounds. First, we explore which features of the EEG signal best maximize machine learning accuracy for a voicing distinction, using a support vector machine (SVM). We manipulate three dimensions of the EEG signal as input to the SVM: number of trials averaged, number of time points averaged, and polynomial fit. We discuss the trade‐offs in using different feature sets and offer some recommendations for researchers using machine learning. Next, we use SVMs to classify specific pairs of phonemes, finding that we can detect differences in the EEG signal that are not otherwise detectable using conventional ERP analyses. Finally, we characterize the timecourse of phonetic feature decoding across three phonological dimensions (voicing, manner of articulation, and place of articulation), and find that voicing and manner are decodable from neural activity, whereas place of articulation is not. This set of analyses addresses both practical considerations in the application of machine learning to EEG, particularly for speech studies, and also sheds light on current issues regarding the nature of perceptual representations of speech.

show abstract

Decoding the temporal dynamics of spoken word and nonword processing from EEG

Cited by 12 publications

References 78 publications

Delineating memory reactivation in sleep with verbal and non-verbal retrieval cues

Delineating memory reactivation in sleep with verbal and non-verbal retrieval cues

Neural basis of melodic learning explains cross-cultural regularities in musical scales

Decoding speech sounds from neurophysiological data: Practical considerations and theoretical implications

Contact Info

Product

Resources

About