Spectro-temporal correlates of lexical access during auditory lexical decision

Highlights

• Alpha de-synchronization has been associated with lexical processing.

• We test sensitivity to lexical activation, with matched bottom-up input.

• Auditory semantic priming protocol in MEG.

• Left auditory alpha de-synchronization attenuated for primed words.

Abstract

Lexical access during speech comprehension comprises numerous computations, including activation, competition, and selection. The spatio-temporal profile of these processes involves neural activity in peri-auditory cortices at least as early as 200 ms after stimulation. Their oscillatory dynamics are less well understood, although reports link alpha band de-synchronization with lexical processing. We used magnetoencephalography (MEG) to examine whether these alpha-related oscillations reflect the speed of lexical access, as would be predicted if they index lexical activation. In an auditory semantic priming protocol, monosyllabic nouns were presented while participants performed a lexical decision task. Spatially-localizing beamforming was used to examine spectro-temporal effects in left and right auditory cortex time-locked to target word onset. Alpha and beta de-synchronization (10–20 Hz ERD) was attenuated for words following a related prime compared to an unrelated prime beginning about 270 ms after stimulus onset. This timing is consistent with how information about word identity unfolds incrementally in speech, quantified in information-theoretic terms. These findings suggest that alpha de-synchronization during auditory word processing is associated with early stages of lexical access.

Introduction

Lexical access during speech comprehension comprises numerous computations, including lexical activation, competition, and selection (e.g. Marslen-Wilson, 1987, McClelland and Elman, 1986, Norris, 1994). Activation describes the stochastic retrieval from memory of lexical representations cued by a spoken or written stimulus; competition and selection describe down-stream stages whereby one representation is chosen from a set of activated possibilities for subsequent processing. While a substantial body of literature has focused on the spatial and temporal profile of the neural substrates of these computations (see e.g. Friederici, 2012, Hickok and Poeppel, 2007 for reviews), there is growing interest in the oscillatory dynamics, i.e. spectro-temporal properties, of the underlying neural generators (e.g. Bastiaansen & Hagoort, 2006). One reason for this shift is the advent of neurophysiological models of speech perception processes that posit a central role for oscillatory mechanisms (e.g. Giraud & Poeppel, 2012). Another is that pathological oscillatory patterns in disorders such as autism (Coben et al., 2008, Cornew et al., 2012, Edgar et al., 2013, Gandal et al., 2010, Uhlhaas and Singer, 2007) and schizophrenia (Edgar et al., 2008, Gandal, Edgar, Klook, & Siegel, 2011) have raised interest in characterizing the role of such activity in both non-pathological and pathological language processing.

Recent work studying time-locked spectral changes during auditory speech processing with magnetoencephalography (MEG) has found that decreases in power relative to baseline, or event related de-synchronization (ERD; Pfurtscheller & Lopes da Silva, 1999), in left auditory cortex between roughly 6 and 14 Hz (alpha band, extending into theta and beta bands) are sensitive to various lexical factors, including lexicality, word frequency, and word repetition within 200–600 ms of word onset (Tavabi et al., 2011a, Tavabi et al., 2011b). Words that are semantically incongruent in a sentential context also show a left-lateralized decrease in alpha- and beta-power (i.e. increased ERD) relative to congruent words (Wang et al., 2012). These results accord well with electroencephalography (EEG) findings showing left-lateralized alpha- and beta-band ERD effects of word-class (Bastiaansen, van der Linden, ter Keurs, Dijkstra, & Hagoort, 2005), a finding also observed in a population of older adults (Mellem, Bastiaansen, Pilgrim, Medvedev, & Friedman, 2012). Thus, converging evidence from MEG and EEG implicates ERD spanning theta, alpha, and beta frequency bands in lexical processing. However, it remains to be seen how this ERD relates to the different subcomponents of lexical access identified in cognitive models of that process.

The majority of prior studies have manipulated lexical processing by presenting different classes of words (e.g. high or low frequency, open or closed class, congruent vs. incongruent). Such manipulations alter numerous factors simultaneously: for example, word frequency effects co-vary with word neighborhood effects, leading to confounding influences on lexical activation and competition (Vitevitch, Luce, Pisoni, & Auer, 1999). Tavabi et al. (2011b) partially address this concern by holding target words constant while varying whether or not words are repeated, but repetition may facilitate multiple stages of speech perception, from phoneme decoding through lexical activation, selection, or task-specific decision processing. Thus, it is difficult to draw strong conclusions about the precise stage(s) of processing indexed by associated neural activity from the finding that theta-alpha ERD is affected by repetition priming alone.

If ERD centered in the alpha-band is associated with lexical activation, then it should be attenuated when lexical activation is facilitated. Semantic priming is a familiar mechanism for facilitating lexical activation (Meyer & Schvaneveldt, 1976), whether via automatic spreading activation or controlled executive processes (Neely, 1991). Changes at the activation stage, however, can also have down-stream consequences by reducing competition and speeding selection and these effects can be challenging to tease apart (but cf. Pylkkänen et al., 2002, Vitevitch et al., 1999). Thus, is important to consider carefully the temporal characteristics of any responses in order to distinguish early activation from later competition and selection effects.

No studies to date have examined local synchrony via spectro-temporal power in an auditory semantic priming protocol, though at least two have examined power or coherence during priming with visual stimulation. Mellem, Friedman, and Medvedev (2013) report that priming during visual word recognition with a letter recognition task leads to decreased gamma ERS in right-posterior electrodes for related targets beginning around 150 ms after stimulus onset. This priming effect is complemented by a later (300–800 ms) increase in gamma ERS in mid-line posterior electrodes as well as a late (600–1000 ms) reduction in alpha ERD in left frontal sites. Kujala, Vartiainen, Laaksonen, and Salmelin (2012) report results from an MEG study in which participants read a list of words that were either semantically or phonologically related. They find an increase in long-range coherence in the theta band between left and right temporal sites associated with semantic relatedness. While both results point towards a role for low frequency (theta/alpha) activity, Mellem et al. also find evidence for a relatively early role of gamma oscillations in lexical processing.

Earlier studies using event-related potentials demonstrated that semantic priming attenuates the evoked N400 response component beginning approximately 200–300 ms after stimulus onset for both visual (Kutas & Hillyard, 1984) and auditory (Holcomb & Neville, 1990) presentation. Converging evidence from MEG has found that semantic priming leads to a sustained reduction in left superior temporal activation during visual and auditory word processing (Vartiainen, Parviainen, & Salmelin, 2009). Left posterior-temporal activation around 300–400 ms after word onset (i.e. the visual M350) has been found to correlate with lexical activation, but not competition or selection (Pylkkänen et al., 2002). Imaging studies using fMRI localize auditory semantic priming effects to left hemisphere superior temporal gyrus near Heschl’s gyrus, middle frontal gyrus bilaterally, and precentral gyrus (Rissman, Eliassen, & Blumstein, 2003). While the latter effects are consistent with response differences during lexical decision for target words for related and unrelated word pairs, the observed superior temporal activation is consistent with effects of facilitated lexical activation.

These data, in combination with the spectro-temporal lexical effects above, offer constrained hypotheses concerning the temporal (200–400 ms) and spatial (superior temporal gyrus) properties of lexical activation during auditory speech perception. They also implicate both low-frequency ERD spanning theta, alpha, and low-beta bands and high-frequency gamma ERS (e.g. Mellem et al., 2013, Tavabi et al., 2011a, Tavabi et al., 2011b). These studies further suggest that lexical facilitation manifests as an attenuation of event-related power (ERD or ERS; see also Wang et al., 2012). Notably, while Tavabi et al. do not report high frequency gamma activity in their auditory studies, both Tavabi et al. and Mellem et al. report theta–alpha ERD. Given the differences in task, modality, and methodology, these results need not be at odds, but they leave open the question of whether we expect an early reduction in low-frequency ERD and/or an early reduction of gamma ERS associated with auditory semantic priming.

In the present study we tested whether both alpha-band ERD and gamma-band ERS signals in left and right auditory cortex are sensitive to semantic priming, as would be expected if the oscillatory pattern in this region were associated with lexical activation. We used an auditory semantic priming protocol in MEG with 83 target words that were related (REL) or unrelated (UNREL) to a preceding prime word; pronounceable non-words (NON) could also appear as targets, and subjects performed a lexical decision on the target word. Target words used in UNREL and REL conditions were matched in bottom-up characteristics, which included word frequency and cohort entropy, a measure that quantifies the uncertainty surrounding the recognition of a word based on the existence of other words that begin with the same phonemes. We also explored whether cohort entropy, which reflects the amount of competition during lexical activation, provided insight regarding how incremental information about lexical identity modulated the target neural signals. MEG data were analyzed using Synthetic Aperture Magnetometry (SAM) to identify the spectro-temporal profile of lexical priming effects in the auditory cortex bilaterally.