Real-time processing in picture naming in adults who stutter: ERP evidence

Abstract

Objective

The aim was to compare real-time language/cognitive processing in picture naming in adults who stutter (AWS) versus typically-fluent adults (TFA).

Methods

Participants named pictures preceded by masked prime words. Primes and target picture labels were Identical or mismatched. Priming effects on naming and picture-elicited ERP activity were analyzed. Vocabulary knowledge correlations with these measures were assessed.

Results

Priming improved naming RTs and accuracy in both groups. RTs were longer for AWS, and correlated positively with receptive vocabulary in TFA. Electrophysiologically, posterior-P1 amplitude negatively correlated with expressive vocabulary in TFA versus receptive vocabulary in AWS. Frontal/temporal-P1 amplitude correlated positively with expressive vocabulary in AWS. Identity priming enhanced frontal/posterior-N2 amplitude in both groups, and attenuated P280 amplitude in AWS. N400 priming was topographically-restricted in AWS.

Conclusions

Results suggest that conceptual knowledge was perceptually-grounded in expressive vocabulary in TFA versus receptive vocabulary in AWS. Poorer expressive vocabulary in AWS was potentially associated with greater suppression of irrelevant conceptual information. Priming enhanced N2-indexed cognitive control and visual attention in both groups. P280-indexed focal attention attenuated with priming in AWS only. Topographically-restricted N400 priming suggests that lemma/word form connections were weaker in AWS.

Significance

Real-time language/cognitive processing in picture naming operates differently in AWS.

Introduction

Persistent developmental stuttering affects slightly less than one percent of the adult population between the ages of 21 and 50, with a 2.2-to-1 male-to-female ratio (Craig et al., 2002). Adulthood stuttering can impact quality-of-life (Craig and Calver, 1991; Craig et al., 2002; Yaruss, 2007; Yaruss, 2010; Blumgart et al., 2010) in domains of social, emotional and mental functioning (Craig et al., 2009). Speech therapy can temporarily alleviate symptoms of stuttering (Bothe et al., 2006). However, relapse after treatment is common in adults who stutter (AWS) (McClure and Yaruss, 2003), and a contributing factor may be that intervention primarily promotes adaptive mechanisms to superficially reduce symptoms of stuttering rather than underlying processing mechanisms in the cognitive-behavioral system that lead to stuttering. Mechanistically, producing speech involves accessing, manipulating and coordinating large amounts of linguistic information while maintaining speech motor control over multiple effectors (Gracco and Löfqvist, 1994). Efficient performance at both levels drives fluent speaking (Levelt, 1989), and there is still much to learn about both levels of performance in AWS. In the present study the focus is on real-time language processing.

In order to produce propositional (i.e., conceptually-novel, spontaneously-generated) speech, humans must draw upon their stored knowledge of words. This happens cognitively, in distinct processing stages that can be isolated experimentally using picture naming. Following a now-common account, the mental lexicon is comprised of different types of linguistic knowledge, including conceptual representations, syntactic words (lemmas), and phonological word forms (lexemes). All of this information is connected in a network-like system (Dell, 1986; Levelt et al., 1999). During picture naming, visual feature processing of a pictured object activates conceptual representations which, in turn, activate lemmas (Collins and Loftus, 1975; Bierwisch and Schreuder, 1992; Roelofs, 1992, 1997). Lemmas compete for activation until one emerges having the greatest activation strength and is selected (Levelt et al., 1999). The whole-word form (lexeme) associated with the selected lemma is activated, and the phonological information associated with the word is retrieved (phonological encoding) (Bock and Griffin, 2000; Berg and Schade, 1992). Finally, the word is prepared for articulation (see van der Merwe, 1997). These processes unfold in just hundreds of milliseconds (van Turennout et al., 2003).

Psycholinguistic research to date with AWS, reviewed in (Maxfield et al., 2010, 2012) and more broadly elsewhere (Bloodstein and Ratner, 2008; Ratner et al., 2009), has been driven by tests of naming speed and accuracy, lexical decision speed and accuracy, depth and breadth of word associations, level of vocabulary knowledge, and changes in stuttering frequency as a function of linguistic context - among others. These types of behavioral assessments have provided rich insights into how the processes, outlined previously, may operate in AWS. However, even psycholinguists investigating language processes in typically-fluent speech production recognized, long ago, the limitations of using behavioral measures for this purpose (e.g., van Turennout et al., 1997; Schmitt et al., 2000; Jescheniak et al., 2002). Similarly, psychologists came to acknowledge that traditionally-used behavioral cognitive measures (e.g., reaction times) have limits in their ability to index specific stages of processing in non-linguistic tasks too (Meyer et al., 1988). Perhaps the most important limitation is that behavioral psycholinguistic measures may not be synched closely enough to specific language processing mechanisms to detect deficits in them (Hagoort and Kutas, 1995). That is, we may be missing important language processing differences in people who stutter when relying on behavioral measures, alone. Therefore, additional work is needed to describe real-time language processing in AWS, with increasing precision.

Brain event-related potentials (ERPs) can be used toward this aim. Scalp-recorded ERPs reflect at least some of the electrophysiological activity generated by the brain as people process stimuli, make decisions and regulate behavior. As described in Hagoort and Kutas (1995), “…in contrast to RTs which are punctate, ERPs are coextensive with the linguistic stimulation and beyond. It is thereby possible to monitor the immediate consequences of a particular experimental manipulation (e.g., a syntactic or semantic violation) as well as its downstream effects, if any” (p.109). Crucially, averaged ERP activity can be decomposed into several different components, many of which reliably index specific language or cognitive processes (see Otten and Rugg, 2005 for a review). Since the late 1990's, ERPs have been used to investigate hypotheses about mechanisms of language production in typically-fluent adults (TFA). In some of the earliest work of this type, two ERP components (lateralized readiness potential and NoGo N200) were used to study the relative timing of semantic, grammatical and phonological encoding processes in TFA (van Turennout et al., 1997, 1998; Schmitt et al., 2000, 2001 a,b; Abdel Rahman et al., 2003; Schiller et al., 2003), as well as the interaction of these different processing levels (Schiller et al., 2003). At around the same time, N400-like components were utilized to study the direction and extent of activation spreading through the mental lexicon during speech production in TFA (Jescheniak et al., 2002, 2003). Error-related ERP components have also been employed to investigate mechanisms of self-monitoring during language production in TFA (Ganushchak and Schiller, 2006, 2008a,b, 2009; Schiller et al., 2009). More recent work, reviewed in Ganushchack et al. (2011), has continued using paradigms combining ERPs and language production to investigate the time-course of lexical retrieval stages from lemma to lexeme (e.g., Eulitz et al., 2000; Koester and Schiller, 2008; Costa et al., 2009; Dell'Acqua et al., 2010), the locus of picture-word interference effects in lexical retrieval (Dell'Acqua et al., 2007; Hirschfield et al., 2008; Aristei et al., 2011), and language production in bilingualism (Christoffels et al., 2007; Chauncey et al., 2009; Verhoef et al., 2009). This same approach has also been extended to investigate language production after stroke (Laganaro et al., 2009; 2011) and, as discussed next, in stuttering.

Brain electrophysiological measures are not new in research on stuttering. For example, a number of studies, some of them dating back decades, have compared known ERP components in AWS versus TFA. Among others, these studies have examined contingent negative variation (e.g., Zimmerman and Knott, 1974; Pinsky and McAdam, 1980; Prescott and Andrews, 1984; Prescott, 1988), P300 activity (e.g., Ferrand, Gilbert and Blood, 1991; Morgan et al., 1997; Hampton and Weber-Fox, 2008; Sassi et al., 2011), error-related components (Arnstein et al., 2011) and auditory evoked potentials (Hampton and Weber-Fox, 2008; Liotti et al., 2010; Maxfield et al., 2010). Of particular relevance is the work of Christine Weber-Fox and colleagues, who have used ERPs to investigate language processing in AWS in receptive mode (i.e., during word recognition and sentence processing). For example, Weber-Fox (2001) reported that AWS versus TFA evidenced attenuated ERP effects to both grammatical and semantic word classes during a sentence reading task. In a later study, Weber-Fox et al. (2004) reported that ERP correlates of phonological processing, elicited during a rhyme judgment task for pairs of printed words, were similar in AWS and TFA. The former findings were taken to indicate that neural functions related to lexical retrieval may be altered in AWS, while the latter findings were taken to indicate that adulthood stuttering may not stem from phonological processing deficits. This line of work has also been extended to investigate syntactic processing in AWS (e.g., Cuadrado and Weber-Fox, 2003; Weber-Fox and Hampton, 2008). As discussed in Maxfield et al. (2012), it remains an open question whether differences observed between AWS and TFA in receptive language processing generalize to language production (although see Pickering and Garrod, 2007, 2013).

In two experiments, Maxfield et al. (2010, 2012) used ERPs to investigate lexical-semantic and phonological processing in AWS in speech production using picture naming. Picture-word priming was used, a paradigm adopted from (Jescheniak et al., 2002) in which a picture on each trial elicits a self-generated label (the prime), followed by an auditory word (the probe, which may relate to the picture label in form or meaning, or share no relationship). ERPs were measured to auditory probe words, and the focus was on probe-elicited N400 activity. N400 is an ERP component that is elicited by lexical-semantic processing and is sensitive to priming, i.e., its amplitude varies inversely with the degree of activation from the prime (see Fishler, 1990; Van Petten and Kutas, 1991; Rosler and Hahne, 1992; Kutas and Federmeier, 2011). In (Maxfield et al., 2010; 2012), TFA evidenced typical semantic and phonological picture-word N400 priming effects. In contrast, AWS evidenced reverse or absent N400 priming in both experiments, pointing to atypical lexical-semantic (Maxfield et al., 2010) and phonological (Maxfield et al., 2012) processing of target picture labels. One limitation of those studies, however, is that picture-word priming is still a fairly off-line approach, i.e., probe-elicited N400 activity is used to draw inferences about upstream processing of self-generated picture labels. Additionally, picture-word priming imposes fairly artificial task demands (e.g., each picture is named at a delay, after the auditory probe has been presented, followed in some designs by probe word verification). Thus, it is possible that atypical results seen for AWS were, at least in part, task artifacts (see Maxfield et al., 2012). The present study investigates language processing during, rather than immediately after, picture naming in AWS - and without the artificial task demands imposed by picture-word priming.

For this purpose, we adopted a modified version of a masked picture priming paradigm from Chauncey et al. (2009). In that experiment, TFA named color photographs of common objects preceded by masked printed prime words. Naming RTs and ERPs were time-locked to picture onset. Pictures in an Identity priming condition were named faster, and more accurately, than pictures preceded by Control (unrelated) primes. Identity priming (versus Control) also modulated ERP activity in three time intervals: 1) at anterior sites peaking at ∼250 ms after picture onset (N300), which was associated with processing of object-specific representations; 2) centrally peaking at ∼400 ms (N400), which was associated with semantic processing; and 3) at left-temporal sites peaking at ∼550 ms, which was attributed to phonological and/or articulatory processing.

For the current experiment, the paradigm used in Chauncey et al. (2009) was modified in three ways. First, we eliminated stimulus familiarization. A recent report has shown that familiarizing participants with stimuli in language production research can prime the activation of target words and lead to qualitatively different results than testing without familiarization (Collina et al., 2013). This modification ensured that naming was as spontaneous as possible. Second, standardized black-line drawings (Szekely et al., 2004) with high name agreement were used, consistent with our previous studies (Maxfield et al., 2010, 2012) and many of the ERP/language production studies cited previously. Finally, the timeline of stimulus presentation was modified such that each picture remained on-screen until naming triggered a voice key to remove the picture and pause the experiment. Participants were instructed to name each picture fully and then cue-up the next trial with a button press. This modification allowed AWS sufficient time to name each picture completely even if stuttering was encountered. In light of these modifications, a direct replication of Chauncey et al. (2009) was not anticipated, and an exploratory analytic approach was used to detect ERP priming effects. Analysis aimed to determine whether picture-elicited ERP activity was modulated by Identity priming versus Control, similarly between AWS and TFA.

Method

Participants

Participants gave written informed consent to participate, completed a medical and language history questionnaire, and were paid 20 dollars upon completion of the study. Participants were a convenience sample of 19 AWS (6 female, mean age = 26 years, 1 month) and 19 TFA (5 female, mean age = 24 years, 10 months). The difference in age between groups (mean difference = 14.79 months) was not statistically significant (t(36) = .72, p = .47). All participants were monolingual speakers of English. All had normal or corrected-to-normal vision. Two TFA, and one AWS, were left-handed. No participants took medications that affect cognitive function, and all were neurologically healthy.

All participants minimally had a high school education or GED-equivalent. Eleven AWS had also completed some form of post-secondary education at time of testing, including trade school (n = 2), undergraduate college degree (n = 6), graduate college degree (n = 2) or doctorate (n = 1). Similarly, 10 TFA had at least some post-secondary education at time of testing, including a completed undergraduate college degree (n = 7) or graduate college degree (n = 3).

Participants in the AWS group self-reported a history of stuttering. Videotaped samples of both read and spontaneous speech were analyzed to confirm the presence of stuttering. Additionally, the impact of stuttering was assessed using the Overall Assessment of the Speaker's Experience with Stuttering (OASES) (Yaruss and Quesal, 2006). OASES scores averaged 46.96 (SD = 10.19). Overall impact ratings were distributed as follows: Mild (n = 1), Mild-to-Moderate (n = 7), Moderate (n = 10), and Severe (n = 1). Six of the 19 AWS reported a family history of stuttering. One AWS reported a family history of dyslexia, and another AWS reported a family history of learning difficulty (the reporting participants were, themselves, not affected by these conditions). A subset of AWS did have a history of co-existing conditions, including mild articulation deficit (n = 2), mild attention deficit disorder (n = 1, untreated using medication), mild learning disability (n = 1), delayed speech (n = 2), and vocal nodules (n = 1).

All of the TFA, and all but one AWS, scored within normal limits on both the Peabody Picture Vocabulary Test-Fourth Edition (PPVT-4) and on the Expressive Vocabulary Test-Second Edition (EVT-2). One AWS had a standard score of 84 on the PPVT-4, indicative of slightly below-normal receptive vocabulary knowledge. As a group, the AWS scored lower than the TFA on both the PPVT-4 (AWS mean = 103.58, SD = 10.59; TFA mean = 111.32, SD = 11.81) (t(36) = 2.13, p = .04), and EVT-2 (AWS mean = 103.53, SD = 12.19; TFA mean = 114, SD = 13.78) (t(36) = 2.48, p = .02). It is important to emphasize that both the PPVT and EVT have construct validity as measures of vocabulary knowledge and not language performance (i.e., they are not language processing measures). Because we used picture stimuli with high name agreement in the main experiment, and included only correctly-named trials in the final analysis, with no differences in accuracy observed between groups (see Results), we were not concerned about average low-normal vocabulary scores in AWS. However, we did take vocabulary knowledge into account when interpreting the ERP results.

Stimuli

Stimuli were 300 black-line drawings of common objects, selected from the International Picture Naming Project (IPNP) (Szekely et al., 2004). The target picture label (i.e., the most frequently-used label) for each line drawing, according to IPNP norms, was a noun. Percent naming agreement for each picture in English, also normed as part of the IPN Project, was 84% or better (mean agreement = 95.40%, SD = 5.26). Target picture labels had no more than three syllables (mean = 1.52, SD = .62) and no more than eight letters (mean = 5.12, SD = 1.45).

Target labels of the 300 line drawings served as prime words. Following the procedure in Chauncey et al. (2009), 50 of the line drawings were randomly assigned to the Identity priming condition, for which they were paired with a printed probe word identical to the target label. The remaining 250 line drawings were randomly assigned an unrelated prime word (i.e., each picture was randomly paired with a label from another remaining picture). Of those, 50 were randomly selected as Control items, and the remaining 200 items were treated as Fillers. This procedure was repeated for each participant. Each picture and prime word appeared just once during the experiment.

Procedure

Participants were informed that on each trial they would see a picture preceded by a rapid letter scramble. Instructions were to pay attention to the letter scramble, and then name aloud the picture, emphasizing accuracy of naming over speed. As shown in Figure 1, each trial comprised a crosshair fixation point displayed for 500 ms, followed by a pattern mask displayed for 200 ms, after which a printed prime word was displayed for 70 ms, and then a backward mask comprised of eight different capitalized consonants for 50 ms, and finally a picture. The picture remained on-screen until naming triggered a voice key, at which time the picture was replaced by a blank screen for 900 ms, followed by instructions to “Press any button for the next trial” which remained on-screen until a push-button response was made. AWS were instructed to say the picture label on each trial completely if they encountered a moment of stuttering, before cueing-up the next trial.

Figure 1.

Events comprising each trial and their durations.

Each participant received a total of 300 trials, presented in a single block lasting ∼15 minutes in duration. Participants were encouraged to take breaks between trials as needed. The order of item presentation was completely randomized. Following Chauncey et al. (2009), ten different eight-letter backward masks were used (RKMVDGJH, CZXNHGFV, BPHMNKRZ, DKXVTRWQ, TRFZGSQD, BZJPFCLM, MBGXSHQT, VNGQSFJK, LDSCNGQR, QTRMNPBK), each appearing 30 times with random selection.

Apparatus and Recording

Each participant sat in a sound-attenuating booth facing a 19-inch monitor. Maximum onscreen height and width of pictures measured 10.7 centimeters. Viewing distance was ∼90 cm. The visual angle of the pictures subtended ∼6.8 degrees. Eprime (Psychological Software Tools, Version 1.1) controlled the experiment and logged naming reaction times registered using a voice key (Psychological Software Tools).

Each participant wore a nylon QuikCap (Neuroscan) fitted with 32 active recording electrodes positioned according to the International 10-20 system (Klem et al., 1999). Electrodes were referenced to a midline vertex electrode. A ground electrode was positioned on the midline, anterior to Fz. Two bipolar-referenced vertical electro-oculograph (VEOG) electrodes, and two bipolar-referenced horizontal electro-oculograph (HEOG) electrodes, recorded electro-ocular activity. Electrodes were constructed of Ag/AgCl. EEG was recorded continuously at a sampling rate of 500 Hz, controlled using SCAN software, Version 4.3 (Neuroscan). Electrode impedance was 5 kOhm or less. Continuous EEG data were low-pass filtered online at a corner frequency of 100 Hz (time constant: DC).

EEG-to-Average ERP Data Reduction

The continuous EEG record of each participant was segmented into epochs. Each epoch comprised EEG data recorded from each electrode during presentation of the picture on each trial, beginning 300 ms before and terminating 1000 ms after picture onset. Trials eliciting incorrect picture names were excluded. Filler trials were epoched and included in the processing sequence until averaging, primarily to ensure that an eye-blink correction algorithm would function as accurately as possible. In order to retain as many trials as possible (Picton et al., 2000), an Independent Component Analysis (ICA)-based (Bell and Sejnowski, 1995) ocular artifact correction procedure (Glass et al., 2004) was implemented in Matlab. After ICA blink correction, channels with fast-average amplitude exceeding 200 microvolts (large drift) were marked bad, as were channels with differential amplitude exceeding 100 microvolts (high-frequency noise). Any EEG trial with more than three bad channels was rejected. For any accepted trial with channels marked bad, the EEG activity at those channels was replaced using a three-dimensional spline interpolation procedure implemented in Matlab (Nunez and Srinivasan, 2006, Appendices J1-J3). Accepted EEG trials were then averaged together, separately for each condition. As a result, each participant had two sets of ERP averages (one for Identity, one for Control). For each participant, no fewer than 40 trials went into the set of ERP averages for each condition (TFA averaged 47.68 trials for Identity (SD = 1.45) and 46.53 trials for Control (SD = 2.57); AWS averaged 47.05 trials for Identity (SD = 1.99) and 45.89 trials for Control (SD = 1.79). The averaged ERP data were low-pass filtered at a corner frequency of 40 Hz, truncated to the critical time window (-100 to 800 ms), re-referenced to left mastoid, and baseline-corrected (-100 to 0 ms).

Analysis

Behavioral data

Naming accuracy and reaction times (RTs) were both analyzed. Naming on each trial was correct if the participant used the target label. Naming was incorrect for trials eliciting no response, a whole-word substitution, a phonological error or a multi-word response. Accuracy data were analyzed by repeated-measures analysis of variance (ANOVA) with Condition entered as a within-subjects factor with two levels (Identity versus Control) and Group entered as a between-subjects factor with two levels (TFA versus AWS). Naming RTs were analyzed using this same approach. Statistical tests of naming accuracy and RT were two-sided and had an alpha-level of 0.05.

ERP data

ERPs were submitted to a covariance-based temporal principal component analysis (PCA) (see Dien, 2010) followed by unrestricted Varimax rotation of the covariance loadings (following Kayser and Tenke, 2003, and using their published Matlab code). The aim was to identify distinct windows of time (hereafter, temporal factors) during which similar voltage variance was registered across consecutive sampling points in the average ERP waveforms. As in our previous work (Maxfield et al., 2010, 2012), PCA was used here because it can facilitate objective component identification and help to address component overlap. This particular adaptation of temporal PCA (unrestricted Varimax rotation of covariance loadings) was adopted for its additional potential to provide more stable statistical results than may be provided by other types of PCA solutions (Kayser and Tenke, 2003). All of these properties seemed advantageous in the present context, as masked priming can produce ERP amplitude modulations that are smaller in magnitude than when priming is more conscious (e.g., Ruz et al., 2003), thus requiring a particularly cautious approach to detecting effects in the data set. To illustrate this point, we note that a truncated PCA solution with 12 Promax-rotated temporal factors (k=2) yielded results essentially similar to those reported herein. However, even relatively minor changes to the truncated PCA settings (e.g., retaining a few more or less temporal factors based on different truncation rules, and using a kappa setting of 3 versus 2 to generate the Promax loadings) changed some outcomes. This led to uncertainty about which truncated solution represented a stable solution, again prompting us to adopt the unrestricted PCA approach here.

Subject ERP averages were combined into a matrix comprised of 451 columns (one column per time point in the -100 to 800 ms epoch) and 2,356 rows (averaged ERP voltages for 38 participants, in each of two conditions, at each of 31 electrodes excluding the left mastoid reference electrode). Each rotated temporal factor is defined by a set of loadings that describe the time-course of each temporal factor, including its peak latency. Each temporal factor is also associated with a set of scores. As described in Kayser et al. (2003), “The factor scores can be interpreted as weighted time window amplitudes… The correspondence between the time course and topography of PCA factors and ERP components (e.g., N1, N2, P3) allows identification of physiologically relevant factors for further analysis, which is analogous to the identification of relevant deflections or time intervals from visually inspecting grand mean ERPs (i.e., this approach also uses common ERP knowledge and reasoning)…” (p. 4). These same authors and others (Dien and Frishkoff, 2005; Dien et al., 2007) also point out that while PCA has several known limitations (e.g., misallocation of variance can result from latency jitter), peak-based or windowed amplitude measures of ERP data are subject to these same limitations, sometimes more severely and/or less explicitly.

To limit the focus of analysis, only temporal factors accounting for at least 1% of the variance were targeted (see Kayser et al., 2003; Foti et al., 2009). As reported below, six temporal factors explained at least 1% of the variance, one of them an artifact associated with baseline drift. Scores associated with each of the five remaining temporal factors were free to vary in amplitude as a function of condition, group and electrode location. These scores were analyzed in two steps. In a first pass, scores at three midline electrodes (Fz, Cz, Pz) were analyzed, separately for each temporal factor, using multivariate analyses of variance (MANOVA) with Electrode entered as a within-subjects factor with three levels, Condition entered as a within-subjects factor with two levels (Identity, Control), and Group entered as a between-subjects factor with two levels (TFA, AWS). Second, for each temporal factor, a topographic analysis was carried out using scores at 20 electrodes covering the left and right hemispheres, at two levels of verticality and five levels of anteriority: Left Lateral (FP1, F7, T7, P7, O1), Right Lateral (FP2, F8, T8, P8, O2), Left Medial (F3, FC3, C3, CP3, P3), and Right Medial (F4, FC4, C4, CP4, P4). These scores were analyzed by MANOVA with Laterality entered as a within-subjects factor with two levels (Left, Right), Verticality entered as a within-subjects factor with two levels (Lateral, Medial), Anteriority entered as a within-subjects factor with five levels, Condition entered as a within-subjects factor with two levels (Identity, Control), and Group entered as a between-subjects factor with two levels (TFA, AWS). MANOVAs were two-sided with an alpha level of .05. F-statistics were exact. Statistically significant effects were followed-up with Bonferroni-corrected pairwise comparisons when appropriate.

Results

Behavioral Data

Naming accuracy

Naming accuracy was high for both groups (Table 1). Accuracy was affected by Condition (F(1,36) = 11.31, p = .002, partial eta-squared = .24), with Identity-primed trials more accurate than Controls (mean difference = 1.18). This suggests that processing of target picture labels was more stable with Identity priming than without. Naming accuracy was not affected by Group (F(1,36) = .89, p = .35), or by the interaction of Group and Condition (F(1,36) = .45, p = .51).

Table 1.

Mean naming accuracy and RTs (and standard deviations) for each group in each condition.

Measure	TFA	AWS
Naming Accuracy (n=50 items per condition)
Identity:	48.05 (1.22)	47.89 (1.66)
Control:	47.11 (2.18)	46.47 (1.54)
Untrimmed Naming RTs (in ms)
Identity:	1047.35 (261.52)	1491.56 (582.30)
Control:	1125.11 (244.55)	1589.44 (715.39)

Naming reaction time

Naming RTs (Table 1) were affected by Condition (F(1,36) = 7.74, p = .009, partial eta-squared = .18), with Identity priming eliciting faster naming than Controls (mean difference = 87.83 ms). This, again, points to more stable processing of target picture labels with versus without Identity priming. Naming RTs were also affected by Group (F(1,36) = 8.33, p = .007, partial eta-squared = .19), with longer naming times in AWS than in TFS (mean difference = 454.27 ms). Group and Condition did not interact to affect naming RTs (F(1,36) = .10, p = .75).

Trimmed naming RTs (not summarized in Table 1) were also assessed to enhance power for detecting effects that might have been more subtle. Trimming involved removing data points greater than one standard deviation from the mean, separately for each participant. This particular trimming approach was chosen because it was shown to increase power for detecting statistically significant effects when subject variability is high in RT data (see Ratcliffe, 1993), as was seen in our AWS group (see Table 1). Comparable to the untrimmed data, trimmed naming RTs were affected by Condition (F(1,36) = 12.64, p = .001, partial eta-squared = .26) with Identity faster than Controls (mean difference = 50.96 ms), and by Group (F(1,36) = 6.57, p = .02, partial eta-squared = .15) with longer naming RTs in AWS versus TFA (mean difference = 318.12 ms), but not by the interaction of Condition and Group (F(1,36) = .06, p = .81).

ERP Data

Grand average waveforms for each condition are shown at 23 electrodes for AWS in Figure 2, and for TFA in Figure 3. Visual inspection reveals a grossly similar pattern of ERP activity between groups from picture onset to the end of the epoch. Unrestricted temporal PCA produced six temporal factors accounting for at least 1% of the variance in the data set. As shown in Figure 4, their peak latencies ranged from 54 ms to 800 ms after picture onset. Each temporal factor will, hereafter, be labeled by its peak latency (e.g., T54). T800 was discarded as an artifact of baseline drift (Van Boxtel, 1998; Kayser and Tenke, 2003). In addition, statistically significant effects were not detected for T54 or T132. Based on visual inspection of the grand averages, we suspect that the backward maskers elicited a P1 component (evident as the earliest positive-going peak at anterior electrodes) which was likely indexed by T54; this component would have peaked at ∼124 ms after backward mask onset consistent with P1 component latency (Heinze et al., 1994; Mangun, 1995). We further suspect that picture onset elicited a P1 component (evident as a topographically-widespread early positivity) that was likely indexed by T132. The three remaining temporal factors had peak latencies between 200 and ∼520 ms after picture onset and, as reported next, all three were associated with statistically significant effects.

Figure 2.

Grand average ERP waveforms for AWS in each condition at each of 23 electrodes.

Figure 3.

Grand average ERP waveforms for TFA in each condition at each of 23 electrodes.

Figure 4.

Variance-scaled factor loadings for each of six temporal factors accounting for at least 1% of the variance in the ERP data set, plotted individually in order of their peak latencies. Factor loadings for temporal factors accounting for less than 1% of the variance are also shown in a single plot (bottom).

T200 effects

T200 variance at midline electrodes was not affected by Electrode, Condition, Group or their interaction in any combination. Topographically, T200 variance was affected by an interaction of Verticality, Anteriority and Condition (F(4,33) = 3.89, p = .01, partial eta-squared = .32). Bonferroni-corrected t-tests detected Condition effects at homologous electrode pairs FP1/FP2 (p = .009), F7/F8 (p = .04), O1/O2 (p = .003) and (marginally) at P7/P8 (p = .05), all lateral on the Verticality factor but dividing Anteriority such that frontal and posterior but no temporal effects were detected. For all four homologous electrode pairs, Identity had larger negative-going amplitude than Control. This pattern can be seen in the grand averages for both groups, particularly at frontal electrodes (see Figures 2 and 3).

T280 effects

T280 variance at midline electrodes was not affected by Electrode, Condition, Group or their interaction in any combination. Topographically, T280 variance was affected by an interaction of Group, Verticality, Anteriority and Condition (F(4,33) = 3.48, p = .018, partial eta-squared = .297). Bonferroni-corrected t-tests detected Condition effects, for the AWS group only, at homologous electrode pairs CP3/CP4 (p = .039), P7/P8 (p = .03), O1/O2 (p = .02), and (marginally) at P3/P4 (p = .05). For all four homologous electrode pairs, Control had a larger positive-going amplitude than Identity. This effect can be seen in the grand averages for the AWS, particularly at posterior electrodes (see Figure 2).

T518 effects

T518 variance at midline electrodes was affected by an interaction of Electrode and Condition (F(2,35) = 11.59, p < .001, partial eta-squared = .4). Bonferroni-corrected t-tests detected a Condition effect at Cz (p = .002) and at Pz (p = .04). At both sites, Identity scores had a larger positive-going amplitude versus Control. This effect can be seen in the grand average waveforms for both groups, particularly at electrode Cz (see Figures 2 and 3).

Topographically, T518 variance was affected by an interaction of Group, Verticality, Anteriority, Laterality and Condition (F(4,33) = 2.75, p = .045, partial eta-squared = .25). Bonferroni-corrected t-tests detected Condition effects, for the TFA group only, at electrodes FC4 (p = .04), C4 (p = .01), CP3 (p = .03), CP4 (p = .03), P3 (p = .03), P4 (p = .01) and P8 (p = .02). At each of these seven electrodes, Identity scores had larger positive-going amplitude than Control for the TFA. These effects can be seen in the TFA grand average waveforms (Figure 3).

Correlations of Vocabulary Scores and Dependent Measures

Since between-groups differences were detected in vocabulary scores (both EVT and PPVT), an important consideration was whether vocabulary knowledge influenced any of the observed behavioral or ERP effects. As noted in Picton et al. (2000), including vocabulary scores as covariates (in the analyses reported previously) would not necessarily remove the effects of vocabulary, particularly if the groups differed systematically on both vocabulary and each dependent measure (naming RTs, naming accuracy and temporal PCA factor scores, respectively). As an alternative, we computed correlations between standard scores from each vocabulary measure and each of our dependent measures, separately for each group in each condition.

For the behavioral data, PPVT and EVT scores were not found to correlate at a statistically significant level (or even marginally) with naming RT or with naming accuracy in either condition for the AWS. In contrast, for the TFA, a moderate positive correlation was detected between PPVT scores and naming RTs in the Identity condition (r = .51, p = .03) and in the Control condition (r = .61, p= .006). Additionally for the TFA, a modest positive correlation of marginal statistical significance was detected between EVT scores and naming accuracy in the Control condition (r = .42, p = .07).

For the ERP data, correlations were computed between scores from each vocabulary measure and temporal factor scores at each of the 23 targeted electrodes, separately for each group, condition and temporal factor combination (for T132, T200, T280 and T518, respectively, each of which was presumed to have captured picture-elicited ERP variance).

Vocabulary scores were shown to correlate with T132 scores in each group as follows. For the TFA, moderate negative correlations were found between EVT scores and T132 scores: a) in the Identity condition at electrodes P4 (r = -.5, p = .03) and O2 (r = -.49, r = .03); and b) in the Control condition at electrodes Pz (r = -.5, p = .03), P4 (r = -.5, p = .03), and O1 (r = -.47, p = .04). For the AWS, moderate positive correlations were found between EVT scores and T132 scores: a) in the Identity condition at electrodes FP1 (r = .46, p = .048), FP2 (r = .47, p = .04), F7 (r = .52, p = .02), and F8 (r = .47, p = .045); and b) in the Control condition at electrodes FP1 (r = .48, p = .04), F8 (r = .51, p = .03) and T8 (r = .48, p = .04). Additionally for the AWS, moderate negative correlations were found between PPVT scores and T132 scores: a) in the Identity condition at electrodes O1 (r = -.51, p = .03) and O2 (r = -.53, p = .02); and b) in the Control condition at electrodes P3 (r = -.47, p = .04), O1 (r = -.56, p = .01) and O2 (r = -.53, p = .02).

Significant correlations were also detected between vocabulary scores and T200 scores in each group as follows. For the TFA, a moderate positive correlation was found between PPVT scores and T200 scores in the Control condition at electrode P7 (r = .46, p = .047). For the AWS, moderate negative correlations were found between EVT scores and T200 scores in the Control condition at electrodes P7 (r = -.46, p = .048) and P8 (r = -.5, p = .03).

Finally, for the TFA group only, a moderate positive correlation was detected between PPVT scores and T280 scores in the Control condition at electrode T7 (r = .53, p = .02).

No correlations were detected between vocabulary scores and T518 scores in any condition for either group.

Discussion

Adults who stutter (AWS) and typically-fluent adults (TFA) spontaneously named pictures preceded by masked printed prime words. Prime words were Identical to, or mismatched (Control), target picture labels. Our main question was whether Identity priming modulated picture-elicited ERPs relative to Control, similarly between groups as picture naming unfolded in real-time. A number of behavioral and electrophysiological effects were observed, including differences between groups.

At a behavioral level, participants in each group named pictures faster and more accurately in the Identity condition versus Control, consistent with effects seen in Chauncey et al. (2009). Additionally, naming latencies were faster for TFA than for AWS regardless of priming condition, consistent with previous naming RT studies with AWS (Van Lieshout et al., 1991, 1993; Prins et al., 1997) and with children who stutter (CWS) (Pellowski and Conture, 2005; Hartfield and Conture, 2006). Group and Condition did not interact to affect naming RTs or accuracy, indicating that gains in performance due to priming were not disproportionate for one group versus the other. This resembles other evidence of similar naming times in various priming conditions in AWS versus TFA (Burger and Wijnen, 1999; Hennessey et al., 2008), but contrasts disproportionate effects of priming on naming times that were seen in CWS versus typically-fluent children (Pellowski and Conture, 2005; Hartfield and Conture, 2006; Anderson, 2008). Finally, naming RTs were shown to correlate positively with receptive vocabulary scores for the TFA (in both priming conditions) but not for the AWS. This same pattern has also been seen in studies involving children who do versus do not stutter (Pellowski and Conture, 2005; Anderson, 2008). One possibility is that factors affecting PPVT performance (namely, frequency of word usage, date of word entry into the mental lexicon and polysemy, see Miller and Lee, 1993) are closely coupled with word retrieval economy in TFA but not in AWS. Alternatively, it may be that naming RTs have more limited construct validity as measures of word retrieval economy in AWS, who have been shown to differ from typically-fluent speakers on speech/vocal RT even in the absence of task demands on word retrieval (see Smits-Bandstra, 2010).

At an electrophysiological level, both groups evidenced a topographically widespread positivity peaking at ∼130 ms after picture onset. The time-course of this effect is consistent with a visual P1 component (Heinze et al., 1994; Mangun, 1995). P1 amplitude was not modulated by Identity priming in either group. However, it was shown to correlate with vocabulary scores in both groups albeit it in different respects. P1 amplitude at posterior electrodes was negatively correlated with expressive vocabulary scores in TFA, and with receptive vocabulary scores in AWS. It is well-established that posterior P1 amplitude is larger to attended versus unattended object locations (Mangun, 1995; Di Russo et al., 2001), although picture naming as used here required attention to a relatively fixed location. Posterior P1 amplitude was also shown to decrease in amplitude with better versus poorer depth-of-knowledge about pictured objects (Abdel-Rahman and Sommer, 2008). This was interpreted as indicating that conceptual knowledge influenced visual perception, possibly by “…reducing the need to draw attention to perceptual analysis” (p. 1061). A similar account of vocabulary/posterior P1 correlations, in the current task, would suggest that in TFA better expressive vocabulary was associated with reduced attention in perceptual analysis, while in AWS better receptive vocabulary had this same effect. Why should different types of vocabulary knowledge influence attention in visual perception differently in TFA versus AWS? Abdel-Rahman and Sommer (2008) proposed that conceptual/semantic knowledge may influence P1 -indexed perceptual processing either in a top-down direction (via connections between semantic and sensory cortical areas) or because conceptual knowledge is perceptually-grounded (with brain areas involved in visual analysis storing semantic features of words). Although not much is known about this level of processing in AWS, naming speed in children who stutter (ages 3 - 5 years) was shown to be more heavily influenced by functional relationships between prime words and self-generated picture labels (e.g., prime word SHINE, target label SUN) than by physical relationships (e.g., prime word BALL, target label SUN) (Hartfield and Conture, 2006). One possibility is that perceptual information in word meaning continues to hold less prominence in expressive vocabularies of AWS but perhaps takes on greater prominence in receptive vocabularies of at least some AWS, driving the correlation seen here in AWS between receptive vocabulary and posterior P1.

Additionally, P1 amplitude at frontal/temporal electrodes was positively correlated with expressive vocabulary scores in AWS. This correlation may be understood through the results of another picture naming study, in which low versus high naming agreement was shown to decrease P1 amplitude at left frontal and parietal regions (Cheng et al., 2010). The authors suggested that a greater number of concept networks having a top-down influence on object recognition became activated for pictures with low versus high name agreement, contributing to ambiguity in naming and reduced frontal P1 amplitude. Cheng et al. (2010) further suggested that greater attention was deployed to processing of low name agreement pictures. Another interpretation is that anterior P1 activity in their study reflected suppression of unattended/irrelevant conceptual information (Luck, 1995; also see Hilimire et al., 2009), with low name agreement pictures (associated with a greater number of potentially competing concepts) requiring greater suppression than high name agreement pictures. Borrowing on this interpretation, poorer expressive vocabulary in our AWS may have been associated with increased suppression of irrelevant conceptual information in early picture processing. One speculation is that in AWS with poorer expressive vocabulary, perceptual processing of pictures involved greater competition in activated concept networks. It seems worth noting that attenuated P1 at the right temporal scalp area almost differentiated AWS from TFA (electrode T8, p = .07, following a marginally significant interaction of Group, Laterality, Anteriority and Verticality in the T132 scores (F(3,44) = 2.62, p = .053), not previously reported). Further investigation is necessary to understand the root of both of the vocabulary/P1 correlations seen here for AWS. For now, they reveal that processes in early perceptual / conceptual processing may operate differently in at least some AWS.

Beyond P1, both groups evidenced Identity priming of ERP activity peaking at ∼200 ms after picture onset. At frontal electrodes, and at posterior electrodes, ERPs had greater negative-going amplitude in the Identity condition than in the Control condition. Chauncey et al. (2009) did not observe Identity priming effects at this latency. However, Eddy et al. (2006) - in an investigation of masked priming effects on processing of black-and-white photographs viewed passively with food items requiring a button press - did observe anterior and posterior Identity priming effects peaking at ∼190 ms after stimulus onset, with Control more negative in amplitude than Identity anteriorly and vice-versa posteriorly (N/P190). Based on the topographic distribution of this effect including its polarity inversion, the authors attributed it to a single neural source in visual cortex involved in early feature processing. In contrast, we relate the early effects in the present study to the N2 family of ERP components. Folstein and Van Petten (2008), in a review of N2 effects elicited in visual modality, proposed that posterior N2 indexes orienting of visual attention while frontal N2 indexes cognitive control. Cognitive control has been operationalized as involving processes in planning and problem-solving (Monsell, 2003) and in initiating and monitoring action (Chan et al., 2008). Based on these interpretations, increased posterior and frontal N2 amplitudes with Identity priming would suggest that supplying target labels prior to picture naming via masked priming set-up participants to process the pictures, and to monitor and control their naming responses, with greater flexibility. Moderate correlations between receptive vocabulary and Control condition T200 activity at posterior sites in TFA (electrode P7) and AWS (electrodes P7, P8) may have impacted the posterior N2 effect, although this effect was most prominent at occipital electrodes.

Later into processing, the AWS and TFA groups differed in ERP activity at ∼280 ms after picture onset. The AWS but not the TFA evidenced Identity priming. In AWS ERPs had greater positive-going amplitude at posterior electrodes in the Control condition than in the Identity condition. The morphology of this effect (topography, time-course and polarity), which was not observed in Chauncey et al. (2009) or in Eddy et al. (2006), is consistent with a P280 component. Rudell and Hua (1996) reported a similar component in an experiment wherein bilingual Chinese-English speakers selectively attended to superimposed English and Chinese word images. Attended words were recognized with high accuracy and evoked an occipital positivity peaking just before 300 ms, while non-attended words were recognized with poor accuracy and elicited no such ERP effect. The authors concluded that the observed ERP effect was highly sensitive to selective attention. Later, Mangels et al. (2001) elicited a parietal-occipital P280 (among several other components) in an experiment requiring participants to remember word lists for later recall. Lists were studied under focused versus divided attention. ERPs were recorded to words during the study phase in each attention condition and averaged on the basis of recall accuracy. Memory and attention both modulated P280 amplitude, but memory did not impact P280 amplitude when separated as a function of attention. On the other hand, studying word lists under focused attention versus easy divided attention versus difficult divided attention elicited P280 activations that were graded in amplitude (from larger to smaller to virtually non-existent, respectively). Like Rudell and Hua (1996), Mangels et al. (2001) proposed that P280 indexes modulations of selective attention. They interpreted P280 against a model of visual selective attention (Posner and Abdullaev, 1994; Posner and Dehaene, 1994) in which medial frontal brain regions drive where and how attention should be directed to meet task demands, after which posterior parietal regions focus attention to facilitate retrieval of specific information. Mangels et al. (2001) proposed that P280 indexed this latter process of enhancing focal attention to facilitate selection of words for further processing. Based on this interpretation, the P280 effect seen here would suggest that AWS enhanced focal attention on the path to naming, with this process attenuated by Identity priming. Vocabulary was not shown to correlate with this effect in the AWS.

Maxfield et al. (2010, 2012) reported other ERP evidence that AWS may enhance focal attention on the path to picture naming. In Maxfield et al. (2010), AWS produced an atypical, reverse semantic N400 priming effect during probe word processing. That is, when the labels of preceding pictures were semantically-related to subsequent probe words, probe-elicited N400 activity was larger in amplitude than N400 to probes unrelated in meaning to their picture primes (rather than smaller in amplitude, the expected outcome demonstrated by a TFA group). In Maxfield et al. (2012), AWS evidenced an atypical, reverse phonological N400 priming effect. That is, when probe words were phonologically-related to target labels of their picture primes, probe-elicited N400 activity was larger in amplitude than N400 activity elicited by probe words unrelated to the labels of their preceding pictures (rather than smaller in amplitude, the expected outcome demonstrated by a TFA group). One interpretation was that, at picture presentation, AWS engaged a center-surround inhibition mechanism aimed at ensuring stable activation of target picture labels by limiting activation spreading to semantically- or phonologically-related neighbors. Center-surround inhibition is a compensatory attentional mechanism proposed by Dagenbach et al. (1990) for retrieving words that are poorly-represented in the mental lexicon (e.g., newly-learned vocabulary words). “…when activation from the sought-for code is in danger of being swamped or hidden by activation in other related codes, activation in the sought-for code is enhanced, and activation in related codes is dampened by the operation of the center-surround retrieval mechanism” (p. 343). In Maxfield et al. (2010, 2012), the reverse N400 priming effects produced by AWS were seen as possible signs that this group had to reactivate (or disinhibit) semantic or phonological neighbors of target picture labels (when those neighbors appeared as probe words) after they had just (presumably) been inhibited by a center-surround mechanism, with reactivation eliciting enhanced N400 activity relative to a control condition (for similar results in TFA see Mari-Beffa et al., 2005; Bertmeitinger et al., 2008). The P280 effect observed here provides more direct evidence that, in AWS, focal attention is heightened on the path to picture naming.

We can imagine at least three different scenarios for why AWS might need to enhance focal attention on the path to picture naming. First, it is possible that target words activate unstably on the path to naming due to impoverished or atypical connections in the mental lexicons of AWS. Consistent with this possibility, CWS evidence immature vocabulary development as noted previously (Hartfield and Conture, 2006), and decrements in vocabulary knowledge have been seen in AWS too (Wingate, 1988; Prins et al., 1997) including in our own AWS group. A second possibility is that there are insufficient attentional resources to support word production in AWS, resulting in unstable activation of target words on the path to naming. Lexical-semantic as well as phonological processes in word production have been shown to draw upon domain-general attentional resources (Ferreira and Pashler, 2002; Roelofs, 2008). There is evidence that attentional resources are allocated away from lexical processing in AWS, which may reflect a strategy for managing fluency (Bosshardt, 2006) or aberrant attentional resource allocation (Arends et al., 1988; Heitmann et al., 2004; also see Bajaj, 2007). A third possibility is that, instead of target words activating unstably in AWS, their semantic or phonological neighbors become too strongly activated. AWS often use linguistic devices (e.g., word substitutions, circumlocutions) to limit stuttering (see Vanryckeghem et al., 2004). It seems possible that such strategies could become automatized over time, perhaps resulting in automatic over-activation of related competitors as a response to frequent disfluency. In any of the scenarios suggested here, there may be less differential activation of the target in comparison to its competitors. A reasonable compensatory strategy would be for AWS to enhance focal attention in order to ensure stable activation of target words. This seems like a plausible explanation for why AWS evidenced enhanced P280 activation without Identity priming. Controlled lexical processing of this sort is also suspected in people with Broca's aphasia (see Bushell, 1996; Blumstein et al., 2000). With Identity priming, P280 was attenuated in our AWS group, presumably because priming the target label immediately preceding picture onset helped it achieve a stable level of activation without heightened focal attention.

Finally, both groups evidenced Identity priming of ERP activity peaking at ∼518 ms after picture onset. At central and parietal midline electrodes, ERPs had greater positive-going amplitude in the Identity condition than in the Control condition. Despite the overall positive-going direction of T518 activity in both conditions, that Control was negative in amplitude relative to Identity is consistent in topography and time-course with an N400-like effect. Both Chauncey et al. (2009) and Eddy et al. (2006) interpreted an ERP pattern having a similar morphology as N400-like too. N400-like activity can be elicited by a range of tasks that require meaningful processing of stimuli (Kutas and Federmeier, 2011). N400 has been interpreted as reflecting lexical-semantic processing occurring at the interface between conceptual representations and amodal lexical representations (Blackford et al., 2012). N400 amplitude varies inversely with the strength of activation that emerges from a priming context (Van Petten and Kutas, 1991). In tasks using lexical (as opposed to sentence) priming contexts, as in the current study, N400 is proposed to reflect relatively automatic lexical-semantic processing (Van Petten, 1995). Based on these interpretations, the N400-like effect seen here would suggest that automatic activation spreading to target lemmas was facilitated by Identity priming in both groups. Additionally, Barry et al. (2001) proposed that when prime and target are identical lexically (as in our Identity condition), priming has the added benefit of strengthening lemma/word form connections. Based on this account, the N400-like effect seen here on Identity trials would reflect not only that automatic activation spreading to target lemmas was more efficient; so too was automatic activation spreading from target lemmas to their associated word forms.

While both AWS and TFA showed an N400-like effect, the scalp topography of this effect was much more widespread for the TFA than for the AWS. In TFA this effect spanned from right frontal-central to right lateral parietal electrodes, but was limited to midline electrodes in the AWS. Vocabulary was not shown to correlate with this effect at any electrode in either group. Similarly, Weber-Fox (2001) reported that N400-like activity elicited by semantically incongruous words in sentence contexts was not as topographically widespread in AWS versus TFA. One possibility is that lemma selection in our AWS was accomplished primarily during the time interval associated with T280, reducing N400-indexed lexical processing activity. Alternatively, or in addition, it may be that lemma/word form connections are less well-established in the neural architecture of the mental lexicon in AWS. Consistent with this possibility, Anderson (2008) reported that repetition priming decreased naming times in both CWS (ages 3-5 years) and age-matched peers, with this effect disproportionately larger in the stuttering group for later-acquired labels. This points to decrements in lemma/word form connections for at least some types of words beginning in early childhood in people who stutter. Whatever the specific cause, weaker activation of neural sources in N400 priming could limit the scalp topography of this effect (Alain et al., 1999).

To summarize, vocabulary knowledge correlated with P1 amplitude differently between groups, possibly related to differences in how perceptual and conceptual information is coded and processed in at least some AWS. A second, common effect was that masked Identity priming modulated anterior and posterior N2 amplitude, suggesting that visual attention and cognitive control in picture naming were enhanced by priming in both groups. Third, a P280 component was sensitive to Identity priming in AWS but not in TFA, suggesting that AWS enhanced focal attention in lexical selection on the path to word production without priming (possibly related to controlling competitive lexical activation spreading). Finally, an N400 priming effect was topographically less widespread in AWS versus TFA, suggesting that AWS recruited a smaller assortment of neural resources during a stage of word production associated with activation of target lemmas and associated word forms.

These findings and their suggested interpretations raise several questions. One has to do with the nature of receptive and expressive vocabulary knowledge in AWS. Crow (1986) outlines several distinctions between these two domains of vocabulary including what type of nuanced knowledge is required to produce versus understand words successfully. A systematic investigation of vocabulary across the lifespan in people who stutter may prove useful in understanding how the acquisition, refinement and maintenance of word knowledge may contribute to persistence of / recovery from stuttering. A related question raised by the current results is why lemma/word form connections may be weakened in AWS? Anderson (2008) reviews this issue as it relates to children who stutter, including a theory that the ability to establish network connections in the mental lexicon (i.e., between conceptual, lemma and word form representations) declines with age as responsiveness to learning (brain plasticity) declines. This issue could also be addressed in an investigation of vocabulary acquisition, refinement and maintenance in people who stutter across the lifespan. A final question raised by the current results is why enhanced focal attention in lexical selection on the path to picture naming may be needed in AWS? Three possibilities were suggested previously, and two additional possibilities emerge based on current results, namely, that lexical selection might be made unstable by differences in upstream processing of conceptual knowledge in at least some AWS, or by diminished lemma/word form connections which may reduce activation levels of target words via diminished feedback activation from word form properties to target lemmas (the latter would assume that lexical selection is influenced not only by top-down conceptual processing but also by bottom-up word form processing, see Dell, 1986). Understanding what may contribute to enhanced focal attention in lexical selection in AWS would raise further questions. For example, if atypical conceptual processing contributes to unstable lexical selection, would then enrichment of conceptual knowledge in expressive vocabulary (e.g., strengthening perceptually-grounded aspects of expressive vocabulary) be an appropriate target of intervention for stuttering? Might strengthening lemma/word form connections have a role in intervention for adulthood stuttering, and how might plasticity in vocabulary refinement mediate this process in AWS? If future research shows that there are insufficient attentional resources to support word production, is there a role for attentional process training (e.g., Sohlberg et al., 1994) in interventions for stuttering? Does reducing/eliminating the use of linguistic avoidance behaviors - an already common target of stuttering modification therapy (Rochford, 1983) - stabilize lexical selection? For now, it seems clinically informative that Identity priming affects processing in both AWS and CWS (Anderson, 2008). Primed word production is used to treat word retrieval deficits post-stroke (Howard, 2000). Analogous approaches may have a place in interventions for stuttering too. In addition to these questions, a methodological concern is why an anterior N300 priming effect, reported in Chauncey et al. (2009) and Eddy et al. (2006), was not observed here? One possibility is that visually-rich photographic stimuli used in those studies drove N300 priming. In general, this study further demonstrates the utility of using a neurophysiological approach to investigate language and cognitive processes in speech production in AWS. Results suggest that lexical processing differs in AWS versus TFA and may warrant attention in interventions for adulthood stuttering.

Highlights.

• Psycholinguistic research with adults who stutter may benefit from increased precision afforded by brain event-related potentials (ERPs).

• ERPs were elicited and compared in adults who stutter (AWS) versus typically fluent adults (TFA) in a masked picture priming task.

• Results suggest that language/cognitive processing differs in AWS versus TFA and may therefore warrant attention in interventions for adulthood stuttering.