xml version 1.0 encoding UTF-8 standalone no
record xmlns http:www.loc.govMARC21slim xmlns:xsi http:www.w3.org2001XMLSchema-instance xsi:schemaLocation http:www.loc.govstandardsmarcxmlschemaMARC21slim.xsd
leader nam 2200373Ka 4500
controlfield tag 001 002028847
007 cr mnu|||uuuuu
008 090914s2009 flu s 000 0 eng d
datafield ind1 8 ind2 024
subfield code a E14-SFE0002820
Huffman, Jessica Lauren.
Semantic and phonological priming effects on N400 activation in people who stutter
h [electronic resource] /
by Jessica Lauren Huffman.
[Tampa, Fla] :
b University of South Florida,
Title from PDF of title page.
Document formatted into pages; contains 81 pages.
Thesis (M.S.)--University of South Florida, 2009.
Includes bibliographical references.
Text (Electronic thesis) in PDF format.
ABSTRACT: To date, research on mechanistic aspects of fluency disorders has focused heavily on motor contributions to stuttering. Only recently have researchers begun to explore psycholinguistic contributions to stuttering. Psycholinguistic planning for speech heavily involves the activation and processing of lexical information. We used a neuroscience approach to compare word activation in mental lexicon while completing a picture naming task in people who stutter (PWS) versus fluent individuals (PWNS). Twenty-eight individuals ranging in age from 19 52 years old participated in a picture-word priming task adopted from Jescheniak et al. (2002). Electroencephalogram (EEG) was recorded while participants saw black and white line drawings, followed immediately by an auditory probe word that was either Semantically-Related, Phonologically-Related, or Unrelated to the label of the preceding picture. EEG was also recorded to Filler (naming-only) trials.Averaged ERPs were generated for each condition. Two principal component analyses (PCA) were conducted in order to summarize patterns in the ERP data and test for differences in ERPs elicited by different conditions. One PCA compared Semantically-Related probe word trials, Semantically-Unrelated probe word trials, and Filler trials. The second PCA compared Phonologically-Related probe word trials, Phonologically-Unrelated probe word trials, and Filler trials. The primary goal of each analysis was to determine whether each probe word condition elicited ERP activity that was different from Filler (naming-only) trials. Relative to Filler trials, all four types of probe words elicited a series of ERP components, some related to sensory processing of the probe words, and some related to linguistic processing of the probe words including N400-type ERP activity. Crucially, N400 priming was observed for PWNS on Semantically-Related trials, but not for PWS.This result indicates that the activation of semantic word networks on the path to picture naming may operate differently in PWS versus PWNS. In contrast, no differences were found between groups for Phonological N400 priming. Discussion relates these effects to the larger body of existing literature on psycholinguistic ability in PWS. Discussion also focuses on how the activation of semantic word networks may differ in PWS versus PWNS, and how therapy for stuttering might address such differences.
Mode of access: World Wide Web.
System requirements: World Wide Web browser and PDF reader.
Advisor: Nathan D. Maxfield, Ph.D.
x Speech-Language Pathology
t USF Electronic Theses and Dissertations.
Semantic and Phonological Pr iming Effects on N400 Activa tion in People Who Stutter by Jessica Lauren Huffman A thesis submitted in partial fulfillment of the requirements for the degree of Master of Arts Department of Speech-Language Pathology College of Arts and Sciences University of South Florida Major Professor: Nathan D. Maxfield, Ph.D. Stefan A. Frisch, Ph.D. Jacqueline J. Hinckley, Ph.D. Date of Approval: March 23, 2009 Keywords: Event-related potentials, Psycholinguistic, Lexical, Adults Copyright 2009, Jessica Lauren Huffman
i Table of Contents List of Tables iii List of Figures iv Abstract vi Introduction 1 A Mechanistic Model of Speech Production 3 Linguistic Utterance Planning 3 Speech Motor Processes 6 A Heavy Focus on Motor Performance in Pe ople Who Stutter 6 Loss of Control, and Why Premot or Aspects May Perpetuate Stuttering 8 Linguistic Utterance Planning in People who Stutter 9 Semantic Network Activation in PWS 11 Phonological Network Activation in AWS 13 A Cognitive Neuroscience Approach to the Study of Lexical Network Activation 16 Summary and Research Questions 18 Methods 20 Participants 20 Stimuli 21 Preparation of Auditory Probe Words 22 Procedures 23 Apparatus and Recording 25 EEG-to-Average-ERP Data Reduction 27 EEG ocular artifact correction 27 EEG trial rejection 28 Final EEG processing 28 Analysis 29 Behavioral Analysis 29 ERP Analysis 29 Temporal-spatial PCA 30 Results 32 Behavioral Data 32
ii ERP Data 34 Analysis of Semantic Conditions 34 Summary of ERP Findings Rela ted to Semantic Picture-Word Priming 42 Analysis of Phonological Conditions 43 Summary of ERP Findings Relate d to Phonological Picture-Word Priming 49 Discussion 51 Summary of Experiment and Findings 51 Discussion of Semantically Related Findings 52 Semantically Driven Findings at ~126 ms 52 Semantically Driven Findings at ~476 ms 54 Discussion of Phonologically Relate d Findings 59 Phonologically Driven Findings at ~268ms 59 Summary, Conclusions and Di rections for Future Research 62 List of References 66 Appendices 75 Appendix A: Research Design 76 Appendix B: Picture Labels and Pr iming Words 79 Appendix C: Behavioral Data 81
iii List of Tables Table 1 Mean & Standard Deviation fo r Naming Accuracy in Each of the Five Conditions 32
iv List of Figures Figure 1. Model of the Normal Sp eech Production Process (adopted from Postma, 2000) 4 Figure 2. Illustration of a picture stimulus and its Semanticallyand Phonologically-Related auditory probe words. 22 Figure 3. Illustration of pr ocedure for Experimental trials. 24 Figure 4. Illustration of procedure for Filler trials. 25 Figure 5. Topographic plot of electrode s on cap (left) a nd picture of cap (right). 26 Figure 6. Grand average ERP waveform s for the Fluent Group (left) and Stuttering Group (right) at Fz (top) Cz (middle), Pz (bottom) for Filler Items, Semantically-Unrelat ed Probe Words, and Semantically-Related Probe Words. 35 Figure 7. Factor loadings of four rele vant temporal factors, each capturing a time window during which distinct ERP activity wa s detected. 37 Figure 8. Spatial factors associated with statistically significant experimental effects within each of four cr itical time windows. 38 Figure 9. Mean factor scores and 95 % confidence intervals summarizing the ERP variance registered at T126, frontal region. 40 Figure 10. Mean factor scores and 95 % confidence intervals summarizing the ERP variance registered at T 476, frontal region. 41 Figure 11. Mean factor scores and 95 % confidence intervals summarizing the ERP variance registered at T476, parietal region, separately for each group. 42
v Figure 12. Grand average ERP waveforms for the Fluent Group (left) and Stuttering Group (right) at Fz (t op), Cz (middle), Pz (bottom) for Filler Items, Phonologically-U nrelated Probe Words, and Phonologically-Related Probe Words. 44 Figure 13. Factor loadings of six rele vant temporal factors, each capturing a time window during which distinct ERP activity was detected. 45 Figure 14. Spatial factors associated with statistically significant experimental effects within each of six critical time windows. 46 Figure 15. Mean factor scores and 95 % confidence intervals summarizing the ERP variance registered at T268, right parietal region. 49
vi Semantic and Phonological Priming Eff ects on N400 Activation in People Who Stutter Jessica L. Huffman ABSTRACT To date, research on mechanistic aspects of fluency disorders has focused heavily on motor contributions to stuttering. Only r ecently have researchers begun to explore psycholinguistic contributions to stuttering. Psycholinguistic planning for speech heavily involves the activation and pro cessing of lexical informa tion. We used a neuroscience approach to compare word activation in mental lexicon while completing a picture naming task in people who stutter (PWS) versus fluent individuals (PWNS). Twenty-eight individuals ranging in age fr om 19 52 years old participated in a picture-word priming task adopted from Jesc heniak et al. (2002). Electroencephalogram (EEG) was recorded while participants sa w black and white line drawings, followed immediately by an auditory probe word that was either Semantically-Related, Phonologically-Related, or Unrelated to the la bel of the preceding picture. EEG was also recorded to Filler (naming-only) trials. Averaged ERPs were generated for each condition. Two principal component analys es (PCA) were conducted in order to summarize patterns in the ERP data and test fo r differences in ERPs elicited by different conditions. One PCA compared Semantically-R elated probe word trials, SemanticallyUnrelated probe word trials, and Filler tr ials. The second PCA compared PhonologicallyRelated probe word trials, Phonologically-Unrel ated probe word trials, and Filler trials.
vii The primary goal of each analysis was to determine whether each probe word condition elicited ERP activity that was different from Filler (naming-only) trials. Relative to Filler trials, all four types of probe words elicited a series of ERP components, some related to sensory processi ng of the probe words, and some related to linguistic processing of the probe words in cluding N400-type ERP activity. Crucially, N400 priming was observed for PWNS on Seman tically-Related trials but not for PWS. This result indicates that the activation of se mantic word networks on the path to picture naming may operate differently in PWS versus PWNS. In contrast, no differences were found between groups for Phonological N400 prim ing. Discussion relates these effects to the larger body of existing liter ature on psycholinguistic ability in PWS. Discussion also focuses on how the activation of semantic word networks may differ in PWS versus PWNS, and how therapy for stuttering might address such differences.
1 INTRODUCTION Stuttering is a disorder of fluency th at emerges during childhood. The prevalence1 of stuttering in children is estimated at 5% (Guitar, 2006; Bloodste in, 1995; Andrews et al, 1983), while prevalence in adults is estima ted at 1% (Bloodstein, 1995; Andrews et al, 1983; Yairi & Ambrose, 2005). Prevalence rate declines because 20 Â– 80% of children diagnosed with stuttering spontaneously recover2 (Bloodstein, 1995; A ndrews et al, 1983; Guitar, 2006). Still, at least three million peop le in the United States stutter persistently into adulthood. Of those people, at least th ree men stutter for every woman (Bloodstein, 1995). These individuals come from all walks of life (Guitar, 2006). Historically, stuttering has been conceptualized by speech and language researchers in a number of different ways. C onture (1996) surmised that the best way to define the phenomenon is to describe what happens during moments of stuttering, both in terms of what can be observed behaviorally and what the speaker reports experiencing emotionally. In this spirit, Marcel Wingate (1964) defined stuttering as follows: 1. (a) Disruption in the fluency of verbal expression, which is (b) characterized by involuntary, audible or si lent, repetitions or prol ongations in the utterance of short speech segments, namely: sound, syllables, and words of one syllable. These disruptions (c) usually occur freque ntly or are marked in character and (d) are not readily controllable. 1 Prevalence is a term used to describe the percentage of people w ho stutter at any given time period. 2 Spontaneous recovery is a term used fo r individuals who recover from stuttering without receivi ng treatment.
2 2. Sometimes the disruptions are (e) accompanied by accessory activities involving the speech apparatus, relate d or unrelated body structures, or stereotyped speech utterances. These activ ities give the appearance of being a speech-related struggle. 3. Also, there are not infrequen tly (f) indications or repor t of the presence of an emotional state, ranging from a general condition of Â“excitementÂ” or Â“tensionÂ” to more specific emotions of a negative nature such as fear, embarrassment, irritation, or the like. (p.488) Though broad in scope, WingateÂ’s definition is still limited in the following way: Much of what happens during speech produc tion is not observable behaviorally. Speech production is driven, in large part, by a number of cognitive processes that unfold covertly, in the mind, within the fraction of a second that separates one Â’s initial intention to speak from actual articulation. Some of th ese processes are psycholinguistic in nature, i.e., involve retrieving and processing linguistic inform ation, while others involve computing speech motor movements. When WingateÂ’s definition was published, little was known about processing linguistic info rmation or computing speech motor movements prior to actual speech production. The purpose of this study was to examine how premotor, psycholinguistic processes operate in adults who stu tter (hereafter, PWS), specifically, the activation of word networks on the path to picture naming. In the sections that follow, we begin by outlining a model of the psycholinguistic and motor mechanisms involved in speech pr oduction. The subsequent section discusses speech motor performance in PWS, includi ng consideration of why a purely speech
3 motor approach fails to account for all sy mptoms of stuttering. Finally, we review psycholinguistic research in the area of fluency disorders, and outline an innovative neuroscience approach for studying word network activation in PWS. A Mechanistic Model of Speech Production Planning Two primary premotor processes driv e planning for speech production. One involves generating a linguistic utterance plan. The second involves generating a speech motor plan and program. Each process is outlined below. Linguistic Utterance Planning Linguistic utterance planning is the process by which an individual selects words for expressing an intended message, arranges those words into phrase and sentence structures, and retrieves the phonological form of those words (Levelt, 1983). In order to generate a linguistic utterance plan, a speak er must possess two types of linguistic knowledge. At one level, a speaker must have a robust lexicon. Individuals can know as many as 50,000 to 100,000 words by adulthood (Miller 1991), all of which are stored in a mental lexicon. A speaker must also possess knowledge about the syntactic, phonological, and discourse rule s that govern language use. Assuming the speaker has linguistic competence at each level, she must also be able to put to use (i.e., retrieve and process) this knowledge efficiently via a set of psycholinguistic pr ocessing activities, outlined next. As shown in Figure 1, the path to sp eech production begins with concept formation. Here, the speaker conceptualizes what she wants to say. Next, she
4 activates words in her mental lexicon that convey the meaning of her intended message. This process is known as lexical selection. As words are activated in mental lexicon, the speaker grammatically encodes her message, i.e., assigns each word a grammatical function and a position in the utterance. Function words are also retrieved during grammatical encoding. The final process of linguistic utterance planning is phonological encoding, during which the speaker retrieves the segments (phonemes) of each word, and assigns them to specific positions within the syllable structure of the emerging utterance. Resulting from this process is a linguistic utterance plan, i.e. a set of words, grammatically arranged, and phonologically sp ecified. Before the linguistic utterance plan is sent forward for articulation, an in ternal monitor checks th at semanticallyand pragmatically-appropriate words have been selected, grammatical structure properly specified, and phonemes retrieve d and correctly assi gned positions with in the utterance (see Figure 1, lexical monitor, syntax mon itor and inner loop, re spectively). In rare instances, psycholinguistic sp eech planning errors go undete cted, resulting in overt speech errors (Fromkin, 1973; Cutler, 1982). More often, planning errors are detected internally. In those cases, the internal mon itor interrupts speech production and initiates a repair (Levelt, 1983), disrupting fluency. Cu rrently, the frequency of which PWS make speech planning errors, perhaps resulting in disrupted fluency, is unknown. As noted above, of the many psycholinguistic activitie s outlined here, our aim was to begin to examine how the lexical selecti on process operates in PWS.
5 Figure 1. Model of the Normal Speech Produc tion Process (adopted from Postma, 2000).
6 Speech Motor Processes As linguistic utterance plans are generate d, the brain translates them into speech motor movements. This involves speech motor planning and programming (van der Meurwe, 1997). Speech motor planning involves establishing a set of movement goals for speech production. These goals dictate where and when the speaker will move the articulators in order to produce the de sired speech sounds. Speech motor programming involves generating a set of instructions that specify how the speech muscles will move in order to realize the goals set forth in the speech motor plan. The amount of force, range, and velocity to be used during specific m ovements is specified in the speech motor program, as are trajectories along wh ich articulators should be moved. The speaker executes the speech motor program, resulting in a series of controlled, sequenced speech motor movements. As an utterance is articulated, proprioceptive and auditory feedback loops ar e used to monitor speech motor control (see bottom right of Figure 1). Feedback allows the speaker to determine whether she has reached intended motor targets. If a mistake is detected, the speaker has the ability to adjust the motor program Â“on-lineÂ”. A Heavy Focus on Motor Performance in PWS The aforementioned processes must operate effi ciently in order for speech to be produced fluently. To date, a preponderance of resear ch has been aimed at investigating motor aspects of stuttering, i.e., to de termine whether deficient motor skill is what sets the stage for moments of stuttering. There is, at leas t, face validity for focusing on motor aspects
7 because PWS sometimes exhibit observable struggle behavior during moments of stuttering. Many PWS report knowi ng exactly what they want to say, but report having difficulty initiating articulation or transitio ning between articulatory targets. A huge body of evidence generated over the last four decades, reviewed by Peters, Hulstijn and Van Lieshout (2000), confirms that impaired mo tor coordination and timing are persistent factors in the speech motor performance of PWS. For example, researchers have consistently shown that lar yngeal reaction times of PWS ar e longer than those of people who do not stutter (hereafter, PWNS), an effect that becomes more apparent with increased utterance complexity (see Bloodstei n, 1995). Such findings are taken to suggest that deficient motor skill has a role in se tting the stage for mo ments of stuttering. As a result of the heavy focus on moto r contributions to stuttering, many of the techniques available today for the treatme nt of adulthood stuttering are motor-based. Explicit planning of oral motor movements, coupled with prolonged speech techniques and the use of self-imposed contingencies (i.e., rewarding oneself when using these techniques), do appear to alleviate stutte ring in adults (Bothe, Davidow, Bramlett, Ingham, 2006). However, stuttering relapse is still very common in adults. Often, treatment helps initially but its effectiveness diminishes over time. One survey by McClure and Yaruss (2003) confirmed that treatment for stuttering is not a one-time solution: 85% of adults who undergo speech th erapy report having two or more different treatment experiences, while 31 % have five or more different treatment experiences. Such findings lead to the hypot hesis that in addition to sp eech motor difficulty, PWS may have difficulty managing other typically automatic aspects of speech production.
8 Linguistic Utterance Planning in PWS In the past 20 years, there has been a shift in focus from motor contributions to stuttering to an examination of whether the covert linguistic processes underlying speech production operate inefficiently in PWS. This new line of research is primarily concerned with how PWS process lexical knowledge. This is because linguistic planning for speech production is driven in large part by the ra pid, cognitive processing of lexical knowledge (Dell, 1986; Garrett, 1988; Butterworth, 1989; Dell & OÂ’Seaghdha, 1992; Levelt, Roelofs, & Meyer, 1999); processing that can be elicited experi mentally via picture naming. In the fraction of a second that separa tes picture presentation from articulation of a picture label, words whose meanings relate to the pictured object are activated in mental lexicon. This is called semantic network activation (Levelt et al., 1999). Soon after (in just tens or hundreds of milliseconds), the phoneme constituents of each word become available and, due to the network orga nization of mental le xicon, activate still other words sharing the same phonemes. This is called phonological network activation (Levelt et al., 1999). The set of potential pi cture labels and phonologically associated words competes for activation. Some words ga in activation strength, and their semantic and phonological properties become available to the speaker, while other words lose activation strength. Efficiency in this process is subs erved by 1) the appropriate development and maintenance of network connections between semanticallyand phonologically-related words in mental lexicon; as well as 2) limits placed on the degree of activation spreading allowed between word s, i.e., too many words should not be able to enter into competition on the path to picture naming (Dell & OÂ’Seaghdha, 1991).
9 Several modern-day theories attribute mo ments of stuttering to breakdowns in these processes, specifically at the level of phonological encoding (Wingate, 1988; Perkins, Kent, & Curlee, 1991; Postma & Kolk, 1993; Karn iol, 1995). The general premise is that phonological encoding, during which a target wordÂ’s phonemes are retrieved (Dell, 1986), is delayed in PWS. According to at l east one theory, the delay occurs because a clear Â“winnerÂ” in the compe tition among words for activation does not always emerge in PWS (Postma & Kolk, 1993). The hypothesized re sult is the undesirably strong activation and subsequent retrieval of phonemes from a semantic or phonological associate of the speakerÂ’s intended word. As noted above, when the internal monitor detects an incorrect phoneme, it signals the speaker to initiate a re pair. If PWS were to frequently generate phonological planning errors due to inefficien t resolution in the competition among word entries, then their internal monitors may fre quently trigger the repair process, setting the stage for moments of stuttering (Postma & Kolk, 1993). Unfortunately, some psycholinguistic resear ch in the area of fluency disorders loses the forest for the trees, focusing exclusiv ely on the time-course of phonological processes (Wijnen & Boers, 1994; Postma & Kolk, 1993). As outlined above, activating the phonological codes of words is inextricably tied to the efficiency with which activation spreads through both semantic and phonologica l word networks. There does exist a limited body of evidence about how lexical ne twork activation operates in PWS at both semantic and phonological levels. However, as ou tlined below, much of this evidence has emerged from research using primarily beha vioral means, which may not be optimally suited for investigating psycho linguistic processes in PWS.
10 Semantic Network Activation in PWS On tests of word association, PWS res pond equally fast (Crowe & Kroll, 1991; also see Taylor, Lore, & Wa ldman, 1970) or faster (Jense n, Markel, & Beverung, 1986) than PWNS. At first glance, this suggests that PWS are equall y skilled at accessing semantic word networks. However, on a task requiring participants to monitor sentences for category-specific words (Bosshardt & Fr ansen, 1996), PWS were slower than PWNS, suggesting difficulty accessing words from sp ecific semantic networks. In addition, PWS have been shown to use fewer synonyms to generate definitions from those produced by PWNS (Wingate, 1988), and word associati ons vary widely between PWS (Crowe & Kroll, 1991). This may suggest that PWS maxi mize speed on word association tasks by strategically using less common responses. A di fferent possibility is that less desirable words automatically gain activation strength on par with more desirable words in PWS. Evidence for this latter effect was repor ted by Newman and Ratner (2007), who reported that PWS made more errors associat ed with lower frequency words than PWNS on a confrontation naming task. The PWS, as a group, were shown to produce naming errors that were lower in frequency than errors produced by PWNS, e.g., Â“androgenyÂ” for Â“boyÂ”. This might be expected if PWS were using word substitutions. A greater number of errors of this type might also point to less restraint on activation spreading in the mental lexicons of PWS. However, th is conclusion was challenged recently by Hennessey, Nang, & Beilby (2008), who used a pi cture naming reaction time (hereafter, RT) task to assess linguistic encoding deficits in PWS. When a probe word appears just before a picture-to-be-named, the picture is named more slowly when the probe word is
11 semantically related to the picture label (s emantic interference) than when the two are unrelated. Semantic interference was of the same magnitude for PWS and PWNS, suggesting that semantic network activation is no less restrained for PWS than it is for PWNS. The studies may not be entirely compar able, however, as Hennessey et al. (2008) used probe words that were highly related to the picture labels (e.g., baby-child), which can attenuate semantic interference and ev en induce priming, versus more distantly related pairs (e.g., horse-whale both of which are animate) (Mahon, Costa, Paterson, Vargas, & Caramazza, 2007). Therefore, the stim uli used may not have been sensitive to subtle differences in semantic netw ork activation between PWS and PWNS. The findings reviewed so far tentatively suggest that semantic network activation is less restrained in PWS. Other findings s uggest that semantic ne twork activation is too restrained in PWS. Wingate (1988) reported that PWS scored lower than PWNS on the Verbal Scale of the Wechsler Adult Intellig ence Scale (hereafter, WAIS), which requires individuals to define words. PWS used a higher average number of words than PWNS, but provided poorer definitions, determined in part by the smaller number of synonyms used. Fewer synonyms indicate, somewhat tent atively, that networ k connections among related words are less well-developed in PWS. Results of two other studies help to substantiate this claim. Prins, Main, and Wampler (1997) reported significantly lower scores on the Peabody Picture Vocabulary Te st (hereafter, PPVT) for PWS than for PWNS, though it is important to note that the PWS still scored within normal limits. The PPVT has construct validity as a measure of receptive vocabulary. Scores on this test are influenced by word frequency and polysemy (Miller & Lee, 1993), the latter of which
12 reflects an ability to adapt words in oneÂ’s vocabulary in order to accommodate new meaning. Low-normal PPVT scores may refl ect sub-clinical difficulty organizing networks of semantically related words in order to accommodate complex meanings. Another sign of poorer semantic network or ganization is that PW S have significantly more difficulty than PWNS disambiguating wo rds in confusing sentences (Watson et al., 1994). It is also interesting to note that PWS stutter more on words that are semantically less-predictable from context than on predictabl e words. One interpretation of this finding posits that PWS have difficulty making lexica l decisions at points of uncertainty in sentence planning (Bloodstein, 1995). Though a tentative hypothesis, inefficient access to, or competition among, semantically re lated words could account for this effect. Phonological Network Activation in PWS In addition to semantically-related wo rds, networks of phonologically-related words become activated on the path to speech production. Burger and Wijnen (1999) examined spoken RTs as PWS and PWNS r ecited lists of phonologically-related words, as well as lists of unrelated words. Facilitation from priming, i.e., the reduction in RTs observed with phonologically-related words ve rsus unrelated word list priming, was equivalent between groups. Hennessey et al (2008) reported similar results for the phonological manipulation in their picture-wo rd task. When a probe word appears directly after a picture-to-be-named, the pict ure is named more quickly when the probe word is phonologically-related to the picture label (phonologi cal priming) than when the two are unrelated. RT facilitation from phonol ogical priming was numerically longer for
13 PWS than for PWNS, but statis tically there was no group diffe rence. The results of both studies suggest that differences in RTs between groups without priming cannot be attributed to disproportionately high co mpetition among phonologically-related words in PWS. Results from other studies run counter to this conclusion. Weber-Fox, Spencer, Spruill, and Smith (2004) asked PWS and PWNS to judge whether pairs of printed words rhymed. Th e words were similar orthographically and rhymed; were dissimilar orthographically and did not rhyme; rhymed but were orthographically dissimilar; or were orthographically sim ilar but did not rhyme. PWS were significantly slower than PWNS when j udging the latter type of stimulus pairs. Weber-Fox et al. (2004) interpreted this effect as suggesting that PWS are particularly sensitive to increased cognitive load, which here, was elicited by phonologic / orthographic incongruency. However, slower phonological monitoring times have been observed in PWS even without incongruency. In thei r study, Sasisekaran, De Nil, Smyth, and Johnson (2006), PWS and PWNS monitore d internal speech for target phonemes during tacit picture naming. Thos e participants also comple ted other tasks designed to assess RTs for simple motor movements, aud itory monitoring of tone sequences, and overt naming. The PWS performed on-par w ith PWNS for all but phoneme monitoring, during which they were signifi cantly slower. Having ruled out motor slowness, auditory monitoring slowness, and naming slowness, Sa sisekaran et al. (2006) concluded that PWS are slower in some aspect of the phonologi cal encoding process. One possibility is that activation spreading to, and compe tition among, phonologically-related words takes longer to resolve in PWS, sl owing phoneme monitoring times.
14 As noted above, one reason for ineffici ent lexical activation may be that activation spreading is less restrained. Unrestrained activation spreading at a phonological level may be evident in the occurrence of phoneme errors. In PWNS, phoneme errors occur more often for lower-frequency words than for higher-frequency words (Stemberger & MacWhinney, 1986; Dell, 1990). According to Dell (1990), as the phonemes of a lower-frequency word (e.g., guy) gain activation strength, bottom-up activation from those phonemes to other, higher-frequency words can occur (e.g., activation of /g/ for Â“guyÂ” could spread bo ttom-up to the lexical entry Â“goÂ” which, in turn, spreads activation to /o/ before /ai/ in Â“guyÂ” can be retrieve d, resulting in a phoneme error). This same phenomenon might help to explain why lower-frequency words attract higher rates of stuttering (Bloodstein, 1995). As noted above, phoneme errors (e.g., elicited by lower-frequency words) might be de tected internally, di srupting fluency in order to initiate a repair. When access to word form information is artificially sped-up, leaving little or no time fo r error correction Â– as with tongue twister tasks Â– PWS generate more speech sound errors than PWNS (Postma & Kolk, 1990; Eldridge & Felsenfed, 1998). These findings lead us to speculate that activat ion spreading at a phonological level may be less rest rained in PWS versus PWNS.
15 A Cognitive Neuroscience Approach to the Study of Lexical Network Activation A limitation of the work reviewed above is that lexical network activation3 was assessed offline, using behavioral measures po tentially influenced by factors such as the motor abilities, metalinguistic skills, and preferences-for-respondi ng brought about by participants. A dependent variab le somewhat immune to these factors is the event-related potential (ERP), generated by the brain auto matically (i.e., not under conscious control) as people process information, make decisi ons, and regulate behavior. Specific ERP components mark the activation of specific c ognitive and linguistic processes. Most relevant for our purposes is N400, an ER P component elicited by lexical stimuli (Fischler, 1990), peaking in amplitude at ~ 400 550 milliseconds after word onset. Crucially, N400 amplitude is inversely relate d to a wordÂ’s activation level in memory (Van Petten & Kutas, 1990). A word whose activation has been primed by a preceding stimulus elicits a relatively small N400, while an unprimed word elicits a relatively large N400. Weber-Fox and colleagues have used this pr operty to assess lexical activation in PWS. In one study (Weber-Fox, 2001), particip ants read sentences silently, some of which contained word violations (e.g., "S he looked at her watch to check the rain ."). N400, while expectedly large in response to words semantically incongruous with their sentence contexts in PWNS, was reduced in amplitude in PWS. Weber-Fox and Hampton (2008) reported similar results from an aud itory task. Both studies assessed lexical activation as PWS processed sentences, i.e., as comprehenders, and it is unclear to what 3 Lexical network activation is used here to describe both the semantic and phonological psycholinguistic aspects of speech production.
16 extent sentence processing mirrors psycho linguistic planning for speech production. At least indirectly, attenuated N400 for word comprehension corroborates behavioral evidence, cited above, that lexical ac tivation operates differently in PWS. N400 priming can also be used to assess lexical network activation on the path to picture naming, using a method called pictureword priming (see Jescheniak, Schriefers, Garrett, & Friederici, 2002). Picture-word priming involves presenting a picture on each trial, followed by an auditory probe word. Partic ipants are instructed to label the picture, but not to name it aloud until several hundred milliseconds after the probe word has been presented (i.e., when prompted by a cue to na me the picture). ERPs are measured at the onset of each auditory probe word. Each probe word elicits ERP activity, including activation of the N400 component. The genera l aim of this research design is to manipulate the amplitude of the N400 by ma nipulating the relationship between the picture labels and auditory probe words. If preceding picture label and subsequent probe word are unrelated, then the N400 activated in response to the pr obe word should be relatively large in amplitude. If, on the other hand, preceding picture label and subsequent probe word are related in some way, N400 am plitude should be attenuated. Jescheniak et al. (2002) found that both seman tic relatedness (e.g., picture of grass followed by probe word mower ) and phonological relate dness (e.g., picture of grass followed by probe word grab ) attenuated N400 amplitude in typically fluent speakers. These results were interpreted as reflecting that, when speakers search for picture labels on the path to picture naming, a set of related words becomes activated in mental lexicon, i.e., via the spreading activation process de scribed above. Some of those words will be semantically-
17 related to the target picture label, while ot hers will be phonologically -related to it. When one of those words is presented auditori ly, the N400 component is attenuated in amplitude, presumably because it was preactivated during the search for the target picture label. If activation spreading through sema ntic or phonological word networks operates inefficiently in PWS, then picture-word priming effects on N400 amplitude seen for PWNS should be less robust or absent for the PWS. We adopted this design to investigate the activation of lexical netw orks in PWS, with some ke y modifications to the design used by Jescheniak et al. (2002) as described in Appendix A. Summary and Research Questions Stuttering is a serious speech disorder that can be desc ribed in terms of observable characteristics. Moments of st uttering may be a re sult of dyscoordination in at least some of the processes that drive normal speech production. To date, research has shown differences in overt motor aspects of speech pr oduction for PWS. However, little research exists to determine if other, covert pro cesses involved in speech production differ in PWS. The purpose of this study was to answer two specific research questions. 1) Does picture-naming activate a network of Semantica lly-Related words in adults who stutter in the same manner as that seen for adults w ho do not stutter, as evidenced by semantic N400 priming effects in a picture-word prim ing task? 2) Does picture-naming activate a network of Phonologically-Related words in ad ults who stutter in the same manner as that seen for adults who do not stutte r, as evidenced by phonological N400 priming
18 effects in a picture-word priming task? In or der to answer these tw o questions, we used the picture-word priming paradigm created by Jescheniak et al. (2002), outlined above. At least three possible outcomes were fore seen. First, if the activation of word networks operates normally in PWS then we would expect to see a typical N400 priming effect wherein the N400 would have a d ecreased amplitude for Semanticallyand Phonologically-Related trials ve rsus Filler and Unrelated tria ls (Jescheniak et al., 2002). If, on the other hand, semantic or phonological co mpetitors undesirably gain activation in PWS on the path to naming, N400 will be larger in amplitude rather than smaller when probe words are related to the picture labels an indication of uncontrolled competition in mental lexicon. Finally, if ac tivation of semantic or phonolog ical word networks operates typically on a gross scale but is sub-clinically inefficien t in PWS (e.g., due to limited network connections), N400 priming should appear but may be reduced in amplitude relative to PWNS. The method used to determ ine whether or not t ypical picture-word N400 priming effects are evidenced in PWS is described in further detail below.
19 METHOD Participants In total, 35 individuals were tested : 17 PWS, and 18 PWNS. Of the 17 PWS participating in the study, 14 were included fo r data analysis (12 men and 2 women, with a mean age of 29.9 years, ranging from 19 to 52 years). Of the 18 PWNS, 14 were included for data analysis (7 men and 7 wo men, with a mean age of 30.14 years, ranging from 19 to 45 years). All included participan ts were monolingual English speakers with normal or corrected-to-normal vision, no h earing deficit, and normal language function. None of the participants we re taking medications that ca n affect cognitive function, and none had a history of neurological injury. A speech sample was collected from each of the 14 PWS in order to confirm their diagnos is of stuttering. All participants gave informed consent to participate in the st udy, completed a medical history questionnaire, and were paid 10 dollars per hour for their participation. A total of seven individuals were ex cluded for the following reasons. One PWS was excluded due to nonnative English-speak ing status; a second PWS had unilateral hearing loss; and a third PWS took prescrip tion medication that can alter cognitive function. One PWNS was excluded due to susp ected head injury; a second PWNS for taking prescription medication that can alter cognitive function; a third PWNS for selfreported Attention Deficit Disorder; and a f ourth PWNS whose recorded EEG data were found to be atypically noisy.
20 Stimuli The study was conducted using thirty-eight simple line drawings of common objects, selected from the IPNP Mini Data base Query, a database of normed pictures (Szekely et al., 2004). All objects were depict ed as black and white line drawings measuring 2 inches in height by 2 inches in width with similar st yle and quality (see Figure 2 for an example). The most freque ntly-used label for each drawing, as determined using norms (gathered as part of the International Pi cture Naming Project), was a noun with no more than two syllables. The average phoneme length for the labels was 3.9, and the average frequency of the labels was 3.2 tokens per million words. For each picture, two probe words were select ed: One being the strongest (semantic) free associate of the picture label but phonol ogically unrelated to it; a second word semantically unrelated to the picture la bel but sharing the word-initial phoneme. Semantic associates were found using the Univ ersity of South Florida Free Association Norms website (Nelson, McEvoy, & Schreiber, 1 998). It is important to note that each of the two probe words was also reassigned to a different picture to which it was completely unrelated. Using the example shown in Figur e 2, Â“waterÂ” would have been assigned to the picture Â“fishÂ”, as well as to a different picture to whic h it was completely unrelated. Therefore, Â“waterÂ” would have appeared twi ce: Once as a word Semantically-Related to its picture, and once as a word Semantically -Unrelated to its picture. Using the same example, Â“frostÂ” would have been assigned to the picture Â“fishÂ”, as well as to a different picture to which it was completely unrelate d. Therefore, Â“frostÂ” would have appeared twice: Once as a word Phonologically-Relate d to its picture, and once as a word
21 Phonologically-Unrelated to its picture. Appendix B lists each picture label; the Semantically-Related probe word for each pict ure; the reassignment of each word in the Semantically-Related list to a Semantica lly-Unrelated picture; the PhonologicallyRelated probe word for each picture; and the reassignment of each word in the Phonologically-Related list to a Phonologically-Unrelated pi cture. Word frequency and number-of-phoneme statistics are shown for each word. Preparation of Auditory Probe Words The probe words were transformed into a set of auditory stimuli as follows. A female, native speaker of English read aloud each word, several times consecutively. All readings were recorded to digital audiotape, digitized at a sampling rate of 44.1kHz, and then processed using Sony Sound Forge 8.0 edit ing software. The best-spoken exemplar of each word was selected; its waveform spliced from the original recording and saved as a separate sound file (.WAV format). The loudness of each word was normalized to an RMS amplitude of 15 dB, and a noise gate us ed to reduce high-frequency noise (hiss). Strongest semantic fr ee associate = water Word-initial phonological probe word = frost Figure 2. Illustration of a picture stimulus and its Sema nticallyand PhonologicallyRelated auditor y p robe words.
22 Procedures Prior to testing, each partic ipant was asked to familiar ize themselves with the black and white line drawings they would be seeing on the computer monitor during the experiment. Instructions were that participan ts would see a picture appear on the screen, and that they were to name the picture once a naming cue (!!!) appeared on the screen. Participants were told they could proceed to the next trial by pushing any button on a response box. To ensure each participant had a good understanding of the task, they were asked to verbally summarize the task require ments using their own words before testing began. In addition to the main task instruc tions, participants were asked to minimize movements while participating in the experiment. Each participant was tested in a single session, during which th ey received a total of 228 trials (38 Phonologically-Related tr ials, 38 Phonologically-U nrelated trials, 38 Semantically-Related trials, 38 Semantically -Unrelated trials, and 76 naming-only Filler trials). The experiment consisted of two diffe rent trial types: Experimental trials, and Filler trials. As shown in Figure 3, each Expe rimental trial consisted of a fixation cross (Â“+Â”) that stayed on the m onitor for 550 milliseconds, followe d by a black and white line drawing that remained on the monitor for 450 milliseconds, followed by a spoken word (either the Semantically-Related Probe, Phonologically-Related Probe, or Unrelated probe), followed by an articulation cue (Â“!!!Â” ) that remained on the monitor until the participant spoke the label fully and presse d a button for the next trial. Eight hundred milliseconds separated the onset of the spoken word from the visual naming signal.
23 The second type of item was a filler item. As shown in Figure 4, for Filler trials the participant saw a fixation sign (Â“+Â” ) that remained on the monitor for 550 milliseconds, followed by a black and white line drawing that remained on the monitor for 450 milliseconds, followed by an articu lation cue (Â“!!!Â”). 1450 milliseconds separated the onset of the picture from the onset of the articulation cue, which stayed on the monitor until the participant pressed the button to begin the next trial. The 228 items were presented in a single, la rge block of trials. Trials for each of the five different conditions we re presented in random order. Each trial was separated by an intertrial interval of 2 100 milliseconds. Each of the 38 pict ures appeared a total of six different times during testing: Twice in Fille r trials, and once in each of the four probe word conditions. This procedure is closely related to the experimental design used by Jescheniak et al. (2002) (see Appendix A). Figure 3.Illustration of procedur e for Experimental trials.
24 Figure 4. Illustration of pro cedure for Filler trials. Apparatus and Recording Each participant was seated in a dimly lit, sound-attenuating booth, facing a 19inch LCD computer monitor. The auditory probe words were pres ented auditorily via high-quality, insert earphones (Etymotic Resear ch, Model E-2). Participants signaled the experimental software (Eprime) to progress from one trial to the next by using a pushbutton response box (Psychological Software Tools). In addition to behavioral data, continuous EEG was recorded from each par ticipant as follows. During testing, each participant wore a nylon QuikCap (Neuroscan ) (see Figure 5). The cap was fitted with a set of 62 active recording electrodes, positioned in a geodesic pattern covering the forehead, top, sides, and back of the head, as well as one reference (midline Cz reference)
25 and one ground electrode. Four additional elec trodes recorded elec tro-ocular activity. A recording electrode was also affixed to each mastoid process. The electrodes were constructed of silver / silver chloride (Ag / AgCl). Conductive electrolyte QuikGel (Neuroscan) was used as the medium between each electrode and the scalp. Placement of the cap took between 10 and 30 minutes. Continuous EEG was recorded from each pa rticipant during testing at a sampling rate of 500 Hz (1 recording every 2 millis econds from each electrode). SCAN software, Version 4.3 (Neuroscan), was used to cont rol EEG recording. Electrode impedance was kept below 30 kOhm (Ferree, Luu, Russell, & Tucker, 2001). The continuous EEG data were low-pass filtered online, at a corner frequency of 100 Hz. E-Prime experimental control software (Psychological Software Tool s, version 1.1), run on a PC computer, was used to present the picture stimuli. Figure 5. Topographic plot of electrodes on cap (left) and picture of cap (right).
26 EEG-to-Average-ERP Data Reduction The continuous EEG record of each part icipant was segmented into individual epochs. Each epoch was comprised of EEG data that had been recorded, from each of the 66 active recording electrodes, during presentati on of the target auditory word in each trial, beginning 300 milliseconds before the onset of the word, and terminating at 1000 milliseconds following the onset of the word. Epochs of the same duration were also created for each Filler trial, beginnin g 300 milliseconds before a word would have appeared (in non-Filler trials) and termina ting 1000 milliseconds later. The epoch length was eventually truncated to a critical in terval of ERP activity (-100 to 800 milliseconds relative to stimulus onset) following averag ing. However, we began with an extended epoch to ensure that the procedures, describe d next, would adequately correct or reject artifacts on the leading a nd trailing edges of this critical time interval. EEG ocular artifact correction Inspection of the EEG data recorded reveal ed that most participantsÂ’ recordings were contaminated by eye blink ar tifact. In order to salvage as many trials as possible, we used an ocular artifact corre ction procedure modified from Dien (2005). The segmented EEG data for each participant were submitted to an Independent Component Analysis (ICA) (Bell & Sejnowski, 1994) After the ICA decomposition of each EEG record into 66 components, the inverse weights (scalp map) of each component were correlated with a blink template generated by averaging at each channel the peak activity of two blink exemplars sampled from each participant. Any component whose inverse weights
27 matched the blink template (r = .9 or bette r) was identified as a blink component. The activity related to each blink component was removed from each tria l if it reduced the overall EEG variance for that trial. At leas t one blink component was identified for each participant. On average 195 trials (SD = 23.16) were corrected for blink activity. EEG trial rejection After ICA blink correction, channels whose fast-avera ge amplitude exceeded 200 microvolts (large drift) were marked bad; as were channels whose differential amplitude exceeded 100 microvolts (high-frequency noise). Any EEG trial with more than three bad channels (5% of the total number of channels ) was rejected from further analysis. No participant lost more than 20% of their trials for any conditio n, and most participants lost well under 10% of their trials per co ndition, due to bad channel artifact. Final EEG processing For any accepted trial with channels mark ed bad (<=3), the EEG activity at those channels was replaced using spherical spline interpolation (Ferree, 2000). The EEG trials were averaged together, separately for each condition. As a result, each participant had five sets of ERP averages: Semantic ally-Related, Semantically-Unrelated, Phonologically-Related, Phonologically-Unrelate d, and Filler. For each participant, no fewer than 30 artifact-free trials went into the set of ERP averages for each Related or Unrelated condition, while no fewer than 69 artif act-free trials went into the set of ERP averages for the Filler condition. The averaged ERP data were truncated to include only
28 the critical time window (-100 to 800 milliseconds), rereferenced to left mastoid, and baseline-corrected (-100 to 0 milliseconds).
29 ANALYSIS Behavioral analysis ParticipantsÂ’ naming responses were scor ed as correct, incorrect, or as no response given. Performance on each trial was sc ored as Â“correctÂ” only if the label used was the precisely-spoken one-word label indicat ed for each picture prior to beginning the experiment (see Procedures above). All other re sponses were scored as incorrect or as no response given. Incorrect responses included two-word answers (e.g., Â“match stickÂ” for match), phonological errors (e.g., Â“cambullÂ” for camel), semantic errors (e.g., Â“deskÂ” for bed), and unrelated word errors (e.g., Â“spiderÂ” for door). All trials sc ored as incorrect or no response given were rem oved from final analysis. ERP Analysis Dominant patterns of variance in the ERP data set were identified using Principal Component Analysis (PCA). PCA is a data reduction technique that can be used to summarize large data sets with great effi ciency. PCA was used here as an ERP preprocessing step, the results of which we re used to describe specific patterns of variance in the ERP data set and to test for experimental effects associated with those patterns of ERP variance.
30 Temporal-spatial PCA The ERP data related to the Semantic asp ect of the task and, separately, the ERP data related to the Phonological aspect of the task, were submitted to a two-step, covariance-based, temporal-spatial PCA (Spe ncer, Dien, & Donchin, 2001). For step one of each analysis, the averaged ERP data we re combined into a single data matrix comprised of 451 columns (one column for each of the sampling points in the critical time window) and 5,208 rows (the averaged ERP voltages for 28 participants, at each of 62 electrodes, in each of the three conditions). This matrix was used as input to a temporal PCA. The aim of this initial, tem poral PCA was to identify distinct windows of time in the ERP averages (hereafter, tempor al factors) during which similar voltage variance was registered across consecutive sa mpling points. As reported below, for the Semantic portion of the tas k, a total of 11 dominant-varia nce temporal factors were retained. For the Phonological portion of th e task, 13 temporal factors were retained. For each analysis, a subset of temporal factors was singled-out because their timecourse (i.e., peak latency) was consistent with that of the standard N400 effect or sensory-evoked ERPs (e.g., N1, P2, which were targeted to assess whether the auditory probe words were processed at a sensory level in addition to lexical-semantic processing). In step two, a spatial PCA wa s performed on the factor scores of each selected temporal factor. That is, the scores for each temporal factor (representing the voltage variance within a speci fic time window) were reconfi gured into a matrix with 62 columns (one column per electrode) and 84 rows (scores for the temporal factor, for each of the 28 participants, in each of the three different conditions). This matrix was then
31 submitted to a spatial PCA, in order to identify topographically coherent regions of voltage activity (hereafter, sp atial factors) with in the time window represented by each temporal factor. The following specific procedures were us ed to conduct each principal component analysis. First, in order to determine how many dominant-variance components were extracted by each PCA, we used Rule M (Preisendorfer & Mobley, 1988). Rule M estimates how many components extracted from a real data set account for more variance than corresponding components extracted fr om a data set of normally-distributed, randomly-sampled noise having the same dimensi ons as the real data set. All components meeting this criterion for each PCA were reta ined and rotated to simple structure using Promax (Hendrickson & White, 1964) with Kaiser normalization and k=2 (Richman, 1986; Tataryn, Wood, & Gorsuch, 1999). All PC analyses and Promax rotations were completed using the Matlab-based PCA Toolbox (Dien, 2005). In order to test for experimental effect s, factor scores summarizing the voltage variance associated with specific pairs of temporal and spatial factors were submitted to a repeated-measures ANOVA with Condition as a w ithin-subjects factor with three levels (Unrelated, Related, Filler) and Group as a between-subjects factor with two levels (Stuttering, Fluent). When the sphericity assumption was viol ated, the degrees of freedom were corrected (Greenhouse & Geiser, 1959). This correction is reflected in the reported p-values. As noted above, we were particularly interested in identif ying temporal-spatial factor combinations whose time-course and scalp topographic distribution, respectively, were consistent with N4 00 activation or with aud itory sensory potentials.
32 RESULTS Behavioral Data Each subjectÂ’s responses were scored fo r naming accuracy. All data were assessed quantitatively and qualitatively for simila rities and differences among the PWS and PWNS groups. Table 1 below depicts the mean number correct and the standard deviation per trial type for each group in each condition (see Appendix B for individual scores on all trial types). Table 1. Mean & Standard Deviation for Na ming Accuracy in Each of the Five Conditions.4 Group Data Type Filler Trials SemanticallyUnrelated Trials SemanticallyRelated Trials PhonologicallyUnrelated Trials PhonologicallyRelated Trials PWS Mean 75.36 37.71 37.79 37.64 37.79 SD 1.15 0.47 0.43 0.63 0.43 PWNS Mean 75.43 37.86 37.71 37.64 37.86 SD 1.09 0.36 0.61 0.84 0.36 Quantitatively there were few differences in the mean number of incorrect trials per trial type between groups. As seen in Ta ble 1, PWS had a slightly higher mean for Semantically-Related trials than PWNS. In contrast, PWNS had somewhat higher means for Filler, Semantically-Unrel ated, and Phonologically-Related trials than the PWS. Both groups had the same mean on th e Phonologically-Unrelated tria ls. The standard deviation 4 A total of 76 items were possible for the Fi ller condition, while a total of 38 items were possible for each of the four Experimental conditions.
33 for each trial type also closely resemble d one another between groups (see Table 1 above). The PWS had a slightly higher standa rd deviation on the Filler, SemanticallyUnrelated, and Phonologically-Related trials; while, the PWNS had a minimally higher standard deviation on the Semantically-Rel ated and Phonologically -Unrelated trials. A repeated-measures ANOVA was run, in or der to determine whether the subtle differences between groups, noted above, were statistically significant. Trial type was entered as a within-subjects fact or with four levels, while gr oup (Fluent versus Stuttering) served as a between-subjects factor. The ANOVA revealed th at there was no main effect of Trial Type (F[3,78]=1.15, p=.34), no main effect of Group (F[1,26]=.049, p=.83), and no two-way interaction of Group and Trial Type (F[3,78]=.41, p=.75). Overall, these findings confirm that the pictures were eas ily recognized and named. Some of the minor naming difficulty encountered (i .e., trials scored as incorr ect) seemed to be caused by momentary lapses of attention on trials, and by fatigue toward the end of the task. At least one PWS also occasionally produced exaggerate d labels (e.g., Â“cheese and sausage pizzaÂ” for Â“pizzaÂ”), perhaps as a secondary st rategy for naming those items fluently. Naming accuracy was also assessed qualita tively, in order to determine whether the nature of the errors was different for the PWNS versus PWS. Our assessment revealed that some differences, though sub tle, could be discerned between groups in terms of different error type s and number of subjects who made errors. Of the PWS who made errors, five different error types we re observed: No response given, two-word answers instead of one-word answers (e.g., Â“m atch stickÂ” for match), phonological errors (e.g., Â“cambullÂ” for camel), semantic errors (e.g., Â“deskÂ” for bed), and unrelated word
34 errors (e.g., Â“spiderÂ” for door). Of the PWNS who made errors, three different error types were observed: No response given, semantic errors (e.g., Â“toadÂ” for frog), and two-word answers instead of one-word answers (e.g., Â“swiss cheeseÂ” for cheese). It is interesting to note that in contrast to the PWS, no PWNS made phonological errors As a whole, the PWS had seven subjects who made an error of any type, while the PWNS only had five subjects make errors. The greatest number of errors made by any participant was nine, with the least number of errors being one. The results of our qualitative analysis reveal that although both groups made errors the number of PWS who generated errors was greater than the number of PWNS, and the error patterns were slightly different. ERP Data Analysis of Semantic Conditions Grand average waveforms for both Groups are shown in Figure 6 at midline electrodes (Fz, Cz, Pz) for Semantically-Unrelat ed trials, Semantically-Related trials, and Filler trials. Relative to Filler trials, Semantica lly-Related and -Unrelated trials elicited a sequence of ERP activity, beginning with an early negativity, followed by positive-going activity, and then later negativ e-going activity. As describe d above, the data for these conditions were submitted to a Temporal-Spati al PCA. A total of 11 temporal factors were identified. That is, 11 different time windows contained distin ct, large-variance
35 -10 -5 0 5 10 amplitude (microvolts) 800 600 400 200 0 time (milliseconds) Unrelated Related Filler -10 -5 0 5 10 amplitude (microvolts) 800 600 400 200 0 time (milliseconds) Unrelated Related Filler -10 -5 0 5 10 800 600 400 200 0 -10 -5 0 5 10 800 600 400 200 0 -10 -5 0 5 10 800 600 400 200 0 -10 -5 0 5 10 800 600 400 200 0 Figure 6. Grand average ERP waveforms for th e Fluent Group (left) and Stuttering Group (right) at Fz (top), Cz (middl e), Pz (bottom) for Filler Items, Semantically-Unrelated Probe Words, and Semantically-Related Probe Words.
36 ERP activity. Those 11 temporal factors accoun ted for 81.23% of the variance in the data set. A spatial PCA was then conducted for the time periods associated with each of the 11 temporal factors, in order to identify topographically cohe rent regions of ERP activity (spatial factors) within each time window.5 Factor scores for each temporal-spatial combination were submitted to ANOVA to test for condition effects, group effects, and group-by-condition interactions in the volta ge variance within the time window represented by the temporal factor, at the s calp region represented by the spatial factor. Just four of the 11 time windows yielde d spatial factors w hose scores were associated with statistically significant effect s. Figure 7 shows the factor loadings for each of the four time windows. The largest consecutive factor loadings indicate the sampling points during which a di stinct pattern of ERP activity was registered. The peak latency (in milliseconds), given by the highest factor loading, is labeled for each time window. Each time window will, hereafter, be labeled using its peak latency (T126, T204, T306, and T476, respectively). Figure 8 show s the spatial factors, within each time window, at which statistically significant ERP effects were detected. The largest factor loadings indicate the electrode sites primarily defining each spatial factor. 5 Eleven separate spatial PCAÂ’s were conducte d, one for each of the 11 temporal factors.
37 4 2 0 factor loadings 800 600 400 200 0 time (milliseconds) 126 ms 4 2 0 800 600 400 200 0 204 ms 4 2 0 800 600 400 200 0 306 ms 4 2 0 800 600 400 200 0 476 ms Figure 7. Factor loadings of f our relevant temporal factors, each capturing a time window during which distinct ER P activity was detected.
38 Frontal Posterio r Time Window T126 T204 T306 T476 Figure 8. Spatial factors associated with statistically significant experimental effects within ea ch of four critical time win dows.
39 Several of the temporal-spatial combinations captured ERP activity that differentiated both the Semantically-Related and Semantically-Unrelated probe word trials from the Filler trials. For exampl e, both conditions elic ited ERP activity more negative in amplitude than Fillers at the posterior region of the scalp during T126 (see Figure 8) (main effect of condition, F(2,52)=9.57, p=.000; Sema ntically-Related versus Filler, p=.000; Semantically-Unrelated versus F iller, p=.02). This effect is consistent with posterior N1 activation, an ER P index that presentation of either type of probe word aroused the central auditory system. Later in time, Semantically-Related and -Unrelated conditions also elicited ERP activity more posi tive in amplitude than Fillers at the frontal region of the scalp during T204 (see Figure 8) (main effect of condition, F(2,52)=12.07, p=.000; Semantically-Related versus Filler, p= .001; Semantically-Unrel ated versus Filler, p=.01), and during T306 (see Figure 8) (m ain effect of condition, F(2,52)=15.06, p=.000; Semantically-Related versus Filler, p=.002; Semantically-Unrelat ed versus Filler, p=.000). The former is consistent with P2 activation, another indicator that word presentations aroused the central auditory system, while the la tter is consistent with P300 activation, an ERP index of c ontext-updating. These effects reveal differences in how probe word trials versus Filler trials were processed. Three additional effects, reported next, were sensitive to the semantic rela tionship between probe word and preceding picture. T126, frontal component. In addition to a posterior N1 component, reported above, T126 also generated frontal ERP activit y (see Figure 8) associated with a main effect of condition (F(2,52)=70.56, p=.000). As shown in Figure 9, both Semantically-
40 Filler Related UnrelatedFactor Score Amplitude, 95% CI1.0 .5 0.0 -.5 -1.0 -1.5 -2.0 Figure 9. Mean factor scores and 95% c onfidence intervals summarizing the ERP variance registered at T126, frontal region. Unrelated and Semantically-Related probe words elicited ERP activity more negative in amplitude than Fillers, differences conf irmed by Bonferroni-corrected pairwise comparisons (Unrelated versus Filler, p= .000; Related versus Filler, p=.000). These effects are consistent with frontal N1 activ ation, an ERP index of capture of auditory attention. Critically, Related was even more negative in amplitude than Unrelated (p=.02), suggesting that a semantic relati onship between preceding picture and probe word captured greater attention than when the two events were unrelated. T476 components. T476 generated a frontal compone nt (see Figure 8) associated
41 with a main effect of condition (F( 2,52)=4.00, p=.03). As shown in Figure 10, Semantically-Unrelated was more negative in amplitude than Filler, a difference confirmed via pairwise-comparison with B onferonni correction (p=.003). As discussed below, this effect is consistent wi th a frontal N400-type ERP component. Filler Related UnrelatedFactor Score Amplitude, 95% CI.4 .2 0.0 -.2 -.4 -.6 -.8 -1.0 -1.2 Figure 10. Mean factor scores and 95% c onfidence intervals summarizing the ERP variance registered at T476, frontal region. T476 also generated a posterior compone nt (see Figure 8) associated with a statistically significant interaction of Group and Condition (F(2,52)=3.59, p=.04). As shown in Figure 11, for the PWS, Semantically -Related was more negative in amplitude than Filler (p=.001), as was Semantically -Unrelated (p=.000). For the PWNS only
42 Semantically-Unrelated was more negative in amplitude than Filler (p=.003). The former is consistent with activati on of a standard N400 ERP for the PWS in response to probe words both Semantically-Related and -Unrelated to preceding pictures; versus standard N400 activation for the PWNS only in respons e to probe words Semantically-Unrelated to the pictures. Fluent StutteringFactor Score Amplitude, 95% CI1.0 .5 0.0 -.5 -1.0 -1.5 -2.0 -2.5 Unrelated Related Filler Figure 11. Mean factor scores and 95% c onfidence intervals summarizing the ERP variance registered at T476, parietal region, separately for each group. Summary of ERP Findings Related to Semantic Picture-Word Priming Presenting an auditory probe word after a picture was shown here to elicit several ERP components, including a posterior N1, anterior P2 and P300. Probe words also
43 elicited an anterior N1 that was larger in amplitude than when the probes were related to the labels of preceding pictures. Finally, probes that were Semantically-Unrelated to preceding pictures elicited two N400-like eff ects: One frontal, and one posterior. For the PWS, probes that were Sema ntically-Related to preceding pictures also elicited a posterior N400 effect. As disc ussed below, this latter eff ect suggests that semantic picture-word priming was not as robust for PWS as for PWNS. Analysis of Phonological Conditions Grand average waveforms for both Groups are shown in Figure 12 at midline electrodes (Fz, Cz, Pz) for P honologically-Unrelated trials, Phonologically-Related trials, and Filler trials. Relative to Filler trials, Phonologically-Rel ated and -Unrelated trials elicited several ERP activations, specificall y, early negative-going activity, later positivegoing activity, and then a negative-going wa ve. The data for these conditions were submitted to a Temporal-Spatial PCA. A total of 13 temporal factors were identified, i.e., 13 different time windows contained distinct, large-variance ERP activity, accounting for 80.16% of the variance in the data set. A spat ial PCA was then conducted for each of the 13 time windows in order to identify topograp hically coherent regions of ERP activity (spatial factors) within each time window.6 Factor scores for each temporal-spatial factor combination were submitted to ANOVA to test for condition effects, group effects, and group-by-condition interactions in the volta ge variance within the time window represented by the temporal factor, at the s calp region represented by the spatial factor. 6 Thirteen separate spatial PCAÂ’s were conducted, one for each of the 13 temporal factors.
44 -10 -5 0 5 10 amplitude (microvolts) 800 600 400 200 0 time (milliseconds) Unrelated Related Filler -10 -5 0 5 10 amplitude (microvolts) 800 600 400 200 0 time (milliseconds) Unrelated Related Filler -10 -5 0 5 10 800 600 400 200 0 -10 -5 0 5 10 800 600 400 200 0 -10 -5 0 5 10 800 600 400 200 0 -10 -5 0 5 10 800 600 400 200 0 Figure 12. Grand average ERP waveforms for th e PWNS (left) and PWS (right) at Fz (top), Cz (middle), Pz (bottom) for Filler Items, Phonologically-Unrelated Probe Words, and Phonologically-Related Probe Words.
45 4 2 0 factor loadings 800 600 400 200 0 time (ms) 76 ms 4 2 0 800 600 400 200 0 268 ms 4 2 0 800 600 400 200 0 130 ms 4 2 0 800 600 400 200 0 324 ms 4 2 0 800 600 400 200 0 214 ms 4 2 0 800 600 400 200 0 446 ms Figure 13. Factor loadings of six relevant te mporal factors, each capturing a time window during which distinct ER P activity was detected.
46 Frontal Posterior Time Window T76 T130 T214 T268 T324 T446 Figure 14. Spatial factors associat ed with statistically significant experimental effects with in each of six critical time win dows.
47 Six of the 13 time windows yielded spatial factors associated with statistically significant effects. Figure 13 shows the factor loadings and peak late ncies for each of the six time windows (hereafter, T76, T130, T214, T268, T324, and T446, respectively). Figure 14 shows the spatial f actors, within each time wi ndow, at which statistically significant ERP effects were detected. As for the Semantic manipulations, se veral of the tempor al-spatial factor combinations observed here captured ERP activity that differentiated both the Phonologically-Related and Phonologically-Unrelat ed probe word trials from the Filler trials. For example, both conditions elicited ER P activity more positive in amplitude than Fillers at the frontal region of the scalp during T76 (see Figure 14) (main effect of condition, F(2,52)=6.73, p=.003; Phonologically -Related versus Filler, p=.02; Phonologically-Unrelated versus Filler, p=.03) This effect is consistent with P1 activation, an ERP measure of auditory inhibi tion and sensory gati ng. It is unclear why this effect was not also observed for the Semantic conditions. Later in time, Phonologically-Related and -Unrelated bot h generated frontal ERP activity more negative in amplitude than Fillers during T130 (see Figure 14) (main effect of condition, F(2,52)=43,77, p=.000; Phonologically-Related versus Filler, p=.000; PhonologicallyUnrelated versus Filler, p=.000). A nearly iden tical effect was obser ved within the same time window (T130) at posterior electrodes (s ee Figure 14) (main ef fect of condition, F(2,52)=10.28, p=.000; Related versus Filler, p=.001; Unrelated versus Filler, p=.01). The posterior negativity is consistent with an N1 effect indexing arousal of the central auditory system, while the frontal negativity is consistent with an N1 effect indexing
48 capture of auditory attention. Even later in time, Related and Unrelated both generated frontal ERP activity more positive in amplitude than Fillers during T214 (main effect of condition, F(2,52)=14.06, p=.000; Related versus F iller, p=.001; Unrelated versus Filler, p=.003), and during T324 (main effect of condition, F(2,52)=10.72, p=.000; Related versus Filler, p=.001; Unrelated versus Filler, p=.003). The former is consistent with P2 activation, an additional indicator that word presentations aroused the central auditory system, while the latter is consistent w ith P300 activation, an ERP index of contextupdating. Finally, during T446, both Related a nd Unrelated generated ERP activity more negative in amplitude than Filler at posterior electrodes (see Figure 14) (main effect of condition, F(2,52)=5.42, p=.007; Related versus Filler, p=.03; Unrela ted versus Filler, p=.03). This activity is consistent with ac tivation of a standard N400 effect for both Phonologically-Related and unrel ated probe words. Therefore, a shared word-initial phoneme between probe word and preceding pict ure label did not modulate the amplitude of the standard N400. However, an additiona l effect was sensitive to word-initial phoneme overlap. T268, right posterior component. T268 generated a right posterior component (see Figure 14) associated with a main effect of Condition (F(2,52)=3.74, p=.03). As shown in Figure 15, Related and Unrelated elicited ERP activity more negative in amplitude than Filler. However, Bonferroni-corrected pairwi se comparisons revealed that only Unrelated and Filler were statistically different (p=.04). This effect is consis tent with a Phonological Mismatch Negativity, observed when the phonological make-up of a probe word is dissimilar to that of a preced ing word, and attenuated when probe and preceding word are
49 phonologically similar. Filler Related UnrelatedFactor Score Amplitude, 95% CI.8 .6 .4 .2 .0 -.2 -.4 -.6 -.8 -1.0 Figure 15. Mean factor scores and 95% c onfidence intervals summarizing the ERP variance registered at T 268, right parietal region. Summary of ERP Findings Related to Phonological Picture-Word Priming Presenting an auditory probe word after a picture was shown he re to elicit several ERP components, including an anterior P 1, anterior and posterior N1 components, anterior P2 and P300. The probe words examin ed here, which were either Unrelated to the label of the preceding picture or shar ed a word-initial phoneme, also elicited a posterior standard N400 effect. Unlike the pa ttern reported above for semantic picture-
50 word similarity, anterior N1 amplitude wa s not modulated by initial phoneme similarity between picture label and probe word. Initial phoneme similarity also failed to modulate the amplitude of the standard N400 effect. Interestingly, a frontal N400 effect was not observed here for either Unrelated or Relate d picture-word combinations. However, an earlier negativity (peaking at ~268 ms) wa s elicited by Phonologically-Unrelated pictureword pairs but not by Phonologically-Related pa irs. None of the effects summarized here differentiated the PWNS from PWS groups.
51 DISCUSSION Summary of Experiment and Findings The aim of this study was to examine th e activation of sema ntic and phonological word networks in PWS and PWNS using a ne uroscience approach. ERPs were recorded at the presentation of auditory probe words in a picture-word priming task. Auditory probe words were either Semantically-R elated to their corresponding pictures, Phonologically-Related (shared the initial phoneme) to their corresponding pictures, or Semanticallyand Phonologically-Unrelated to their corresponding pi ctures. ERPs were also recorded on trials that required only naming of pictures wit hout presentation of an auditory probe word (Filler trials). The ta sk was designed to answ er two questions: 1) Does picture-naming activate a network of se mantically-related words in adults who stutter in the same manner as that seen for adults who do not stutter, as evidenced by semantic N400 priming effects in a pictureword priming task? 2) Does picture-naming activate a network of phonologica lly-related words in adults who stutter in the same manner as that seen for adults who do not stutter, as eviden ced by phonological N400 priming effects in a pict ure-word priming task? Semantically-Related and Semantically-U nrelated probe words each elicited ERP components not observed on Filler trials. Sema ntically-Unrelated pr obe words elicited N1, P2, P3, and two N400-like components. Semantically-Related probe words elicited an even larger N1 activation than that seen for Semantically-Unrelated words. Semantically-Related probe words also elicited P2 and P300 activations. While
52 Semantically-Related probe words did not e licit any N400-like activ ations for the PWNS, a robust bilateral parietal N400 activation was observed in response to SemanticallyRelated probe words for the PWS. Similar to the ERP results obtained for the Semantic probe word conditions, both Phonologically-Related and Phonologically-Unrel ated probe words elicited ERP activity not observed on Filler trials. P honologically-Unrelated probe words elicited N1, P2, P3, and N400 ERP components, as well as an ERP component resembling the Phonological Mismatch Negativity, described in further detail below. Phonologically-Related probe words elicited all of these same compone nts except for the Phonological Mismatch Negativity. This pattern of results was seen for the PWNS and PWS groups. Effects that were most central to our research question are discussed in the sections that follow. Discussion of Semantic ally Related Findings Semantically Driven Findings at ~126ms ERPs elicited by auditory probe words in a picture-word priming study included two early sensory-evoked potentials typically seen in response to auditory stimuli (N1 and P2). The most significant of these was the N1 ERP component which, as noted above, had a frontal scalp distribution and a p eak latency at ~126 ms after the onset of auditory probe words. The functional significance of auditory N1 has been investigated a number of times in the past. Naatanen a nd Picton (1987), who reviewed this body of research, concluded that there are at least three different N1-type components generated
53 by the brain. All three N1 responses can be el icited by the onset or offset of an auditory stimulus. However, it is possible to differe ntiate the N1 components by their latency, location on the surface of the scalp, and Â– mo st importantly Â– by their sensitivity to different task and subject factors. One N1 co mponent is a frontocentral negativity that is sensitive to auditory selectiv e attention. For example, if pa rticipants are instructed to listen to tones of different frequencies, a nd respond by pressing a button only when a tone of low frequency is presented to one ear, N1 is larger in amplitude to those target stimuli than to other tones (i.e., high tones presented to the same ear and any tone presented to the opposite ear). The N1 component seen in response to auditory probe words in our study closely resembles this N1 component, functionally indicating that participantsÂ’ attention was captured by the auditory stimuli. As reported above, N1 had a larger amp litude when auditory probe words were Semantically-Related to their corresponding pictures than when the stimuli were Semantically-Unrelated. This suggests that a conceptual-semantic relationship between preceding picture and probe words captured gr eater attention than when the two events were unrelated. A similar finding has been reported in at least one previous study. Novick, Lovrich, and Vaughan (1984) conducte d a study wherein participants were randomly presented with both real words and nonsense words under four different conditions. Depending on the condition, participants were asked to respond to all spoken words; to a specific real word; to a specific nonsense word; or to a spoken word belonging to a specific semantic category. For tr ials where participants had to monitor for specific semantic categories, Novick et al. (1984) reported a slightly late r negative going
54 waveform, as compared to other conditions, initiated at ~150ms and lasting to ~ 250 ms post stimulus onset. Similarly to the N1 s een in our study, these results suggest that semantic (categorical) processing of lexical stimuli can differentially impact auditory selective attention, as evidenced by modula tions in frontocentral N1 activation. Semantically Driven Findings at ~ 476 Milliseconds Two N400-like components were observed for the semantic task. One was a negative-going component with a frontal scalp distribution, whose amplitude was larger for Semantically-Unrelated trials than for Filler trials. A frontal N400-like component has been observed by other researchers in respons e to Semantically-Unrelated word pairs (Franklin, Dien, Neely, Waterson, & Huber, 2007) When the words in each pair were both Semantically-Related and highly associated with one another (e.g., dog-cat), the frontal N400-like component was attenuated in amplitude. In contrast, word pairs that were Semantically-Related but not strong se mantic associates (e.g., dog-lizard) were not shown to modulate the amplitude of frontal N 400. This pattern of results seen by Franklin et al. (2007) suggests that fr ontal N400 is sensitive to c oncept formation; a level of psycholinguistic processing that precedes lexical selection. As noted in the Method, we selected a uditory probe words that were strong conceptual associates of their corresp onding picture labels (determined via free association norms published by Nelson, McEvoy, & Schreiber, 1998). When the auditory probe words were reassigned to semantically and conceptually unrelat ed pictures, frontal N400 activation was observed for both the PW NS and PWS. In other words, presenting
55 auditory probe words that were not strongl y conceptually relate d to their preceding pictures elicited a robust fr ontal N400 wave. One interpretation is that this wave represents the activation of c oncepts represented by the auditory probe words. This is based on our finding that, when auditory probe words followed pictures to which they were strongly conceptually associated, frontal N400 was not detected. Absence of frontal N400 activation for this condition suggests that preceding pictures conceptually primed auditory probe words, an effect that was seen for both the PWNS and PWS groups. The other N400-like component observed in our Semantic task had a posterior scalp distribution. This compone nt is consistent with the standard (or Â“classicÂ”) N400 effect. As noted in the Introduction, the amplitu de of standard N400 is inversely related to a wordÂ’s activation level in memory (Van Petten & Kutas, 1991). When a target word is primed semantically (i.e., preceded by a sema ntically-related word or words), it elicits a relatively small posterior N400 component ve rsus when the target word is unprimed. This effect is known to occur as long as pa rticipants attend to th e prime word (Deacon & Shelley-Tremblay, 2000); which, in the case of our experiment, was the picture label on each trial. While the PWNS exhibited this prim ing effect for Semantically-Related words, the PWS did not exhibit this effect. The posterior N400 priming effect seen for the PWNS group is consistent with N400 priming effects reported by Jescheniak et al. (2002), who also tested PWNS. This priming effect indicates that retrieving a picture label activates Semantically-Related words in the mental lexicon. When one of t hose words is presented auditorily directly after the picture-to-be-named, the N400 ERP elicited by the auditory probe word is
56 relatively smaller in amplitude than when the label of the preceding picture is Unrelated to the probe word. Absence of N400 attenua tion (i.e., N400 priming) for SemanticallyRelated probe words in the PWS group may be in terpreted in at least two different ways. One interpretation is that semantic network activation operates inefficiently for PWS. That is, labeling a picture may not automatically activ ate a network of Semantically-Related words, as appears to happen for PWNS. This result coincides to some extent with previous research review ed in the Introduction. Most notably, Wingate (1988) reported that PWS performed more poorly on the WAIS. Qualitative analysis of performance on the WAIS indicated that PWS used fewer synonyms to generate definitions, an indirect indicator of poor se mantic network connections. In a different study, Bosshardt and Fransen (1996) reported that PWS had slower reaction times identifying category-specific words than PWNS This, too, indirectly points to weakness in the activation of semantic networks in PWS. Finally, Prins, Ma in, and Wampler (1997) found that lower-frequency word s had a large effect on lexi calization time despite the vocabulary levels of PWS. Participants in th is study heard a word on each trial and were then shown pictures of various items. Inst ructions were to press the space bar on a keyboard when the picture corresponding to the word was shown. PWS were found to be slower than PWNS when selecting word-picture combinations that were particularly low in frequency (e.g., wench, laggard). Prins et al. (1997) conjectured that slow processing during beginning stages of lexi calization, specifically semantic processing of words, was at-play in PWS, and went further to specula te that slow semantic activation of words might be to blame for disrupted fluency in PWS. All three findings outlined here
57 indirectly point to inefficien t activation of semantic word networks, in line with our N400 results. Noteworthy is our finding that fr ontal N400 priming but not parietal N400 priming was observed for the PWS. This implie s that PWS encounter difficulty not at the level of concept formation but specifically involving the activation of words in semantic networks. A different interpretation is that PWS exhibit over-activation of semantically associated words on the path to picture naming, resulting in disproportionately high competition between words comprising seman tic word networks. One study conducted by Newman and Ratner (2007) used pictures of highly familiar words to assess naming speed using RT, accuracy, and fluency in PWS and PWNS. They found that while various lexical factors (i.e., word fr equency, neighborhood density, and neighborhood frequency) had a similar effect on the naming speed of PWS and PWNS, more naming errors were made by PWS (AWS 94.3%, AWDNS 97.6%). Word frequency also had a particularly negative effect on fluency for the PWS. In addition, PWS were shown to supply very low-frequency responses to re latively common stimuli (e.g., Â“patella" for knee). Newman and Ratner suggest that PWS Â“could have a fundamental difference in a basic level of language processingÂ” (p.208) This fundamental difference could be explained via over-activation of semantic ne tworks, an effect that might possibly be learned. PWS often learn to keep multiple synonyms in-mind in order to readily substitute words as a strategy for avoidi ng moments of stuttering. That is, they circumlocute or substitute words as a copi ng mechanism when stuttering is anticipated, which may inadvertently cause an initial ove r-activation of Semantically-Related words
58 (Guitar, 2006, p.158). In the contex t of our task, this may have manifested in maintained activation of multiple potential word entries at the presentation of the pictures. Having more than one possible word entry in mind may have induced an interference effect at the presentation of Semantically-Related auditory probe words, resulting in parietal N400 activation not seen for these items in the PWNS group. The pattern of naming errors in PWS reported by Newman and Ratner (2007) supports th is conclusion. Both interpretations of the N400 effects en tertained here are tentative and require further study. Although the PWS did not displa y a typical posterior N400 priming effect for Semantically-Related probe words, their picture naming ability was grossly similar to that seen for the PWNS (see Results, Behavior al Data). This finding aligns with those of Weber-Fox (2001), who found that PWS exhi bited atypical N400 activations while participating in a visual sentence processing ta sk. At the same time, behavioral data did not differentiate the PWNS and PWS in he r study. PWS and PWNS performed similarly on the Test of Adolescent and Adult Language a standardized language assessment, and had similar accuracy for detecting seman tic anomalies embedded within sentences (Weber-Fox, 2001). Weber-Fox and Hampt on (2008), too, reported atypical N400 activations for PWS compared with PWNS, wh ile behavioral linguistic performance was not found to differ between groups. Findings from both studies align w ith our results that although behaviorally PWS may perform simila rly to PWNS, covert aspects of lexical processing may differ as evidenced by ERPs. Th erefore, it appears to be important to look beyond behavior, to covert processes, when investigat ing clinical phenomena such as stuttering.
59 Discussion of Phonologi cally Related Findings Phonologically Driven Fi ndings at ~268 Milliseconds Several ERP effects were observed for the Phonology task that differentiated processing of auditory probe words from Fille r trials, among them anterior P1, anterior and posterior N1 components, anterior P2 a nd P3 components, and a posterior standard N400 effect. However, none of these co mponents differentiated processing of Phonologically-Related from Phonologically-Unr elated stimuli, and none differentiated the PWNS versus PWS groups. An additional ERP component was observed that did differentiate processing of Phonologically-R elated versus Pho nologically-Unrelated words. This was the Phonological Mismatch Negativity, a negativegoing wave elicited only by Phonologically-Unrelated words, which peaked in amplitude at ~268 milliseconds post stimulus onset. The Phonological Mismatch Negativity was first observed by Praamstra and Stegeman (1993) and was later labele d by Praamstra, Meyer and Levelt (1994). Praamstra and Stegeman (1993) had ten particip ants complete two tasks requiring them to listen to pairs of words and non-words. Some trials contained word pairs or non-word pairs that rhymed, while other trials contained word pairs or non-word pairs that did not rhyme. Participants were instructed to judge whether the two auditorily presented words comprising each trial rhymed. Praamstra and Stegeman (1993) observed a significant modulation in ERPs during the time wi ndow spanning 300 to 600 milliseconds after second word onset, with ERPs to non-rhyming stimuli more negative in amplitude than
60 ERPs to rhyming stimuli. This ERP modulation was larges t at central and temporoparietal electrode locations on the scalp. In a later study (Praamstra et al., 1994), 24 participants completed two experiments both involving a delayed or imme diate response task. In one experiment, participants heard pairs of words or non-words that rhymed. In the second experiment, participants heard pairs of words or non-words that alliterated. For each trial, in both experiments, participants were required to judge whether auditorily presented words rhymed or alliterated. Similar to the result s reported by Praamstra and Stegeman (1993), here analysis of the ERP data revealed that unrelated word pairs elicited a larger, more negative-going wave between 450-700 milliseconds after second word onset for the rhyming words and between 250-450 milliseconds for alliterating real word pairs. Nonword pairs did not elicit N400 priming effect s for either experiment. Praamstra et al. (1994) interpreted this effect as Â“Â…simila r enough to the Â‘classi calÂ’ N400 to be provisionally placed in the same categoryÂ” (p.215). These above mentioned studies both support the notion that phonologi cal priming of words can m odulate a late negative ERP component. We, too, observed a negative going ERP component for Phonologically-Unrelated items that was not observed for Filler trials or for Phonologically-Related trials. This effect had a peak latency of ~268 milliseconds after probe word onset, and was localized to the right temporal-parietal region of the scalp. Our data in conjunction with the above mentioned studies helps to confirm that at least one ERP component is sensitive to phonological priming between words. To our know ledge, we are the first to report that a
61 Phonological Mismatch Negativity can be elicit ed via a picture-word priming task. The studies by Praamstra and colleagues, reviewed above, both utilized auditory match-tosample tasks to elicit and modulate the Phonological Mismatch Negativity. It is noteworthy, too, that in the pres ent study we did not observe the same phonological priming effects reported by Jescheni ak et al. (2002). In that study, they showed phonological priming for an N400 comp onent that was widespread across the scalp at regions more consistent with a tr aditional N400 effect th an with a Phonological Mismatch Negativity effect. This difference in ERP manipulations may be related to the modifications of two aspects in our phonol ogical priming task from that used by Jescheniak et al. (2002). Firs t, in that study, picture labe ls and auditory probe words shared multiple overlapping phonemes. In cont rast, picture labels and auditory probe words in our Phonologically-Related condition shared only the initial phoneme. As noted in Appendix A, we decided to prime initi al phoneme only, because stuttering tends to occur on initial sounds. Second, Jescheniak et al. (2002) required participants to explicitly remember and judge auditory pr obe words. In contrast, we instructed participants to ignore probe words, due to concern that a dual-task requirement would induce disproportionately hi gh rates of stuttering duri ng testing. Stronger phonological priming coupled with a requirement to hold auditory probe words in phonological working memory seems to induce stronger phonological N400 primi ng effects, although the contribution of each factor is not specifically known at this time. In contrast to our results for semantic network activation, we did not observe any difference between PWS and PWNS on phonologi cal processing aspects of our task. A
62 recently-published review of the literature on psycholinguistic ability in PWS (Broklehurst, 2008) concluded that PWS ma y have a slower rate of phonological encoding than PWNS, but only under increased cognitive load. As reviewed in the Introduction, Weber-Fox et al. (2004) conducte d a study wherein PWS and PWNS had to perform rhyme judgment tasks. Participants were shown words that either rhymed and looked similar (e.g., thrown, own), did not rhyme and did not look similar (e.g., cake, own), looked similar but did not rhyme (e.g., gown, own), or did not look similar but rhymed (e.g., cone, own). They found that although ERPs and response accuracy were similar for both groups, RTs were slower for PWS as cognitive load was increased (i.e., on trials for which phonology and orthography of the target words was incongruent). Since our naming task was not particularly demanding, participants may not have faced enough cognitive load to have affected the efficiency of phonological network activation in our PWS. Summary, Conclusions and Directions for Future Research Speech production begins with the formation of a concept, followed by two levels of premotor planning. One level involves gene rating a linguistic utte rance plan while the second involves generating a speech motor pl an and program. Activation and processing of words plays a key role in generating a li nguistic utterance plan. As a speaker forms a concept-to-be-named, words in mental lexicon begin to activate, and activation spreads to a cohort of Semanticallyand Phonologically -Related words. The words compete for activation until a Â“winnerÂ” emerges.
63 The activation of semantic and phonologica l word networks was explored in PWS via a picture-word priming task, during which ERPs were recorded. Results indicate that the activation of semantic word networks ope rates differently for PWS versus PWNS on the path to picture naming. One interpretation of these results was that semantic network activation is under-active or inefficient in PWS. This may be because PWS have weak associations among Semantically-Related words in mental lexicon; an effect that other, behavioral research involving PWS seems to confirm. If true, an under-activation of words at the earliest stages of linguistic u tterance planning may affect efficiency of processing in some other stages of speech planning that follow. One method for treating this level of function might be to focus on vo cabulary learning and st rengthening of word associations, which in turn may Â“primeÂ” th e linguistic utterance planning system to operate more efficiently. A different interpretation was that our re sults reflect over-activ ation of semantic word networks. We speculate that if PWS ha ve over-activation of semantically related words this may have induced an interference effect at the presentation of SemanticallyRelated auditory probe words, resulting in parietal N400 activation as opposed to attenuation. If true, then therapy might instead need to be aimed at decreasing circumlocution and word substitution behavior s, i.e., in order to reduce the amount of Semantically-Related words active on the path to speech production. Still another possibility is that over-activati on of semantic word networks, if real, is not strategic but reflects a developmental problem. That is, the architecture of the mental lexicon in PWS may be disrupted, due to genetic predisposi tion to weaker language function (Guitar,
64 2006), which may spur disorganization in how lexical knowledge is represented and access in PWS; at least semantically. Because we saw frontal N400 priming e ffects for both groups and did not see similar posterior N400 priming effects for PWS, we can speculate that differences between PWS and PWNS do not lie within th e concept formation stage of planning; rather, the problem seems to involve the processi ng of lexical items, at least at a semantic level. Although differences in phonological network activation were not observed, one may still ask whether processing differs purely at a lexical-semantic level in PWS, or whether other linguistic pro cessing deficits can be found in PWS. Cuadrado & WeberFox (2003) compared processing of syntactic (specifically, verb agreement) violations in PWS and PWNS by investiga ting the morphology of the P600 ERP elicited from individuals in each group. The P600 is a la te, positive-going ERP component elicited by phrase structure and agreement violati ons. Cuadrado & Weber-Fox (2003) observed atypical P600 ERPs for PWS, evidence of a linguistic processing deficit beyond lexicalsemantics. More research is needed to better understand the breadth and depth of atypical linguistic processing in PWS. Although we did not observe group diffe rences at the level of phonological network activation, it is still possible such differences exis t. For example, further research could be conducted using probe words w ith a stronger phonological relationship with their corresponding picture labels. Anot her way to further examine phonological encoding in PWS may be to force participants to actively attend to auditory probe words during testing. A task such as this would be more taxing on the system which, as noted
65 above, may draw-out differences in phonol ogical processing ability between PWS and PWNS. The current results are consistent with the hypothesis that PWS do not execute linguistic utterance planning in the same way as PWNS. This was evidenced by the atypical N400 effects displayed by PWS while performing Semantically-Related naming tasks. Further research is needed to explore the significance and extent of psycholinguistic processing differences in PWS and PWNS, and the specific manner in which such differences set the st age for moments of stuttering.
66 LIST OF REFERENCES Andrews, G., Craig, A., Feyer, A.M., H oddinott, S., Howie, P.M., & Neilson, M.D. (1983). Stuttering: A review of res earch findings and theories circa 1982. Journal of Speech and Hearing Disorders 48 226-246. Bell, A. J. & Sejnowski, T. J. (1995). An information maximization approach to blind separation and blind deconvolution. Neural Computation, 7 1129-1159. Bloodstein, O. (1995). A Handbook on Stuttering (ed 5). San Diego, CA: Singular Publishing Group Inc.. Bosshardt, G. & Fransen, H.J.M. (1996). On -line sentence processing in adults who stutter and who do not stutter. Journal of Speech and Hearing Research, 3, 785Â– 797. Bothe, A.K., Davidow, J.H., Bramlett, R.E., & Ingham, R.J. (2006). Stuttering treatment research 1970Â–2005: I. Systematic review incorporating trial quality assessment of behavioral, cognitive, and related approaches. American Journal of SpeechLanguage Pathology, 15 321Â–341. Brocklehurst, P. (2008). A re view of the covert repair hypothesis of stuttering. Contemporary Issues in Communica tion Sciences and Disorders, 35 25-43. Burger, R. & Wijnen, F. (1999). Phonological en coding and word stress in stuttering and nonstuttering subjects. Journal of Fluency Disorders, 24 91Â–106.
67 Butterworth, B. (1989). Lexical access in speech production. In W. Marslen-Wilson (Ed.), Lexical representation and process (pp. 108 Â–135). Cambridge, MA: The MIT Press. Conture, E.G. (1996). Treatment efficacy: Stuttering. Journal of Speech and Hearing Research 39 18-26. Craig, A.R., Franklin, J.A., & Andrews, G. ( 1984). A scale to measure locus of control of behavior. British Journal of Medical Psychology, 57 173-180. Crowe, K. & Kroll, R. (1991). Response late ncy and response class for stutterers and nonstutterers as measured by a word-association task. Journal of Fluency Disorders, 16 35-54. Cuadrado, E.M. & Weber-Fox, C.M. (2003) Atypi cal syntactic processi ng in individuals who stutter: Evidence from event-relate d brain potentials and behavioral measures. Journal of Speech, Language, and Hearing Research, 46, 960Â–976. Cutler, A. (1982). Slips of the Tongue and Language Production The Hague, Germany: Mouton de Gruyter. Deacon, D., & Shelley-Tremblay, J. (2000). How automatically is meaning accessed: A review of the effects of a ttention on semantic processing. Frontiers in Bioscience, 5 E82Â–E94. Dell, G.S. (1986). A spreading-activation th eory of retrieval in sentence production. Psychological Review, 93 283-321. Dell, G.S. (1990). Effects of frequenc y and vocabulary type on phonological speech errors. Language and Cognitive Processes, 5 313-349.
68 Dell, G.S. & OÂ’Seaghdha, P.G. (1991). Medi ated and convergent lexical priming in language production: A comment on Levelt et al. (1991). Psychological Review, 98 604-614. Dell, G.S. & OÂ’Seaghdha, P.G. (1992). Stages of lexical access in language production. Cognition, 42 287Â– 314. Dien, J. (2005). PCA Toolbox [Computer software] (Version 1.93). Lawrence, KA. Available from http://www.peopl e.ku.edu/~jdien/dowloads.html. DiLollo, A., Neimeyer, R.A., & Manning, W.H. (2002). A personal construct psychology view of relapse: Indications for a narr ative therapy component to stuttering treatment. Journal of Fluency Disorders, 27 19-42. Eldridge, K.A. & Felsenfeld, S. (1998). Differentiating mild and recovered stutterers from nonstutterers. Journal of Fluency Disorders, 23 173Â–195. Escera, C., Yago, E., Corral, M., Corber a, S., & Nunez, M.I. (2003). Short communication attention capture by audito ry stimuli: Semantic analysis follows attention switching. European Journal of Neuroscience, 18 2408-2412. Ferree, T.C., Luu, P., Russell, G.S. & Tucker D.M. (2001). Scalp electrode impedance, infection risk, and EEG data quality. Clinical Neurophysiology, 112 536Â–544 Ferree, T.C., Eriksen, K.J., & Tucker, D. M. (2000). Regional head tissue conductivity estimation for improved EEG analysis. IEEE Transactions on Biomedical Engineering, 47 ,1584Â–1592.
69 Fischler, I. (1990). Comprehendi ng language with event-related potentials. In J. W. Rohrbaugh, R. Parasuraman & R. Johnson, Jr. (Eds), Event-related Brain Potentials: Basic issues and applications (pp. 165-177). New York, NY: Oxford University Press. Franklin, M.S., Dien, J., Neely, J.H., Huber, E., & Waterson, L.D. (2007). Semantic priming modulates the N400, N300, and N400RP. Clinical Neurophysiology, 118 1053-1068. Fromkin, V. A. (1973). Speech Errors as Linguistic Evidence The Hague, Germany: Mouton de Gruyter. Garrett, M.F. (1988). Processes in language production. In F. J. Newmeyer (Ed.), Linguistics: The Cambridge survey (Volume 3, pp. 69-96). Cambridge, MA: Harvard University Press. Greenhouse, S. W. & Geisser, S. (1959). On methods in the analysis of profile data. Psychometrika, 24 95-112. Guitar, B. (2006). Stuttering: An integrated approach to its nature and treatment, third edition Baltimore, MD: Lippincott Williams & Wilkins. Hendrickson, A.E., & White, P.O. (1964). Promax: A quick method for rotation to oblique simple structure. The British Journal of Statistical Psychology, 17 65-70. Hennessey, N., Nang, C., & Beilby, J. (2008). Sp eeded verbal responding in adults who stutter: Are there deficits in linguistic encoding? Journal of Fluency Disorders, 33 180-202.
70 Jensen, P.J., Markel, N.N., & Beverung, J. W. (1986). Evidence of conversational disrhythmia in stutterers. Journal of Fluency Disorders, 1 183Â–200. Jescheniak, J.D., Schriefers, H., Garrett, M. F., & Friederici, A.D. (2002). Exploring the activation of semantic and phonological c odes during speech planning with eventrelated brain potentials. Journal of Cognitive Neuroscience, 14, 951-964. Jezer, M. (1997). Stuttering: A life bound up in words. New York, NY: Harper Collins Publishers inc. Karniol, R. (1995). Stuttering, language, and cognition: A review and a model of stuttering as suprasegmental se ntence plan alignment (SPA). Psychological Bulliten, 117 104-124. Kutas, M. & Van Petten, C. (1991). Influences of semantic and syntactic context on openand closed-class words. Memory Cognition, 19 95-112. Levelt, W.J.M. (1983). Monitori ng and self-repair in speech. Cognition 14 41104. Levelt, W.J.M., Roelofs, A., and Meyer, A.S. (1999). A theory of le xical access in speech production. Behavioral and Brain Sciences, 22 1Â–75. Mahon, B., Costa, A., Peterson, R., Vargas, K.A ., & Caramazza, A. (2007). The effect of semantic distance in the picture-word in terference paradigm: Implications for models of lexical selection. Journal of Experiment al Psychology-Learning Memory and Cognition, 33 503-535. McClure, J.A. & Yaruss, J.S. (2003). Stuttering survey suggests success of attitudechanging treatment. ASHA Leader, 8 19. ( http://www.nsastutter.org//search/dsp_results.php?tbl=material&mixid=176 )
71 Miller, G.A. (1991). The science of words New York, NY: Scientific American Library. Miller, L. & Lee, C. (1993). Construct va lidation of the Peabody Picture Vocabulary Test-Revised: A structural equation mode l of the acquisition order of words. Psychological Assessment, 5 438-441. Naatanen, R., & Picton, T.W. (1987). The N1 wave of the human electric and magnetic response to sound: A review and an an alysis of the component structure. Psychophysiology, 24 375-425. Nelson, D.L., McEvoy, C.L., & Schreiber, T.A. (1998). The University of South Florida word association, rhyme, and word fragment norms. http://www.usf.edu/FreeAssociation/ Newman, R. & Bernstein Ratner, N. (2007). The role of selected lexical factors on confrontation naming accuracy, speed, and fluency in adults who do and do not stutter. Journal of Speech, Language and Hearing Research, 50 196Â–213. Novick, B., Lovrich, D., & Vaughan, H.G. ( 1985). Event-related potentials associated with the discrimination of acoustic and semantic aspects of speech. Neuropsychologia, 23 87-101. Perkins, W.H., Kent, R.D., & Curlee, R.F. (1991). A theory of neuropsycholinguistic function in stuttering. Journal of Speech and Hearing Research, 34 734-752 Peters, H.F.M., Hulstijn, W., & Van Lieshout P.H.H.M. (2000). Recent developments in speech motor research into stuttering. FoliaPhoniatrLogop, 52 103-119.
72 Postma, A. (2000). Detection of errors dur ing speech production: A review of speech monitoring models. Cognition 77 97131. Postma, A. & Kolk, H. (1993). The covert repair hypothesis: Prear ticulatory repair processes in normal and stuttered disfluencies. Journal of Speech and Hearing Research, 36, 472-487. Postma, A., Kolk, H., & Povel, D.J. (1990). Sp eech planning and exec ution in stutterers. Journal of Fluency Disorders, 15 49Â–59. Praamstra, P., Meyer, A., & Levelt, W. (1994). Neurological manifestations of phonological processing: Latency variati on of a negative ERP component timelocked to phonological mismatch. Journal of Cognitive Neuroscience 6 204-219. Praamstra, P. & Stegeman, D. (1993). Phonologi cal effects on the auditory N400 eventrelated brain potential. Cognitive Brain Research 1 73-86. Preisendorder, R. W. & Mobley, C. D. ( 1998). Anatomy of word and sentence meaning. Proceedings of the National Academy of Sciences, 95 899-905. Prins, D., Main, V., & Wampler, S. (1997) Lexicalization in adults who stutter. Journal of Speech, Language and Hearing Research, 40, 373-384. Richman, M. (1986). Rotation of principal components. Journal of Climatology, 6 293335. Sasisekaran, J., De Nil, L., Smyth, R., & Johnson, C. (2006). Phonological encoding in the silent speech of persons who stutter. Journal of Fluency Disorders 31 1-21. Spencer, K.M., Dien, J., & Donchin, E. (2001). Spatiotemporal analysis of the late ERP responses to deviant stimuli. Psychophysiology, 38 343-358.
73 Stemberger, J.P. & MacWhinney, B. (1986). Frequency and the lexical storage of regularly inflected forms. Memory and Cognition, 14 17-26. Szekely, A., Jacobsen, T., D'Amico, S., Devescovi, A., Andonova, E., Herron, D., Lu, C.C., Pechmann, T., Plh, C., Wicha, N., Federmeier, K., Gerdjikova, I., Gutierrez, G., Hung, D., Hsu, J., Iyer, G., Kohnert, K., Mehotcheva, T., OrozcoFigueroa, A., Tzeng, A., Tzeng, O., Arvalo, A.L., Vargha, A.,Butler, A.C., Buffington R., & Bates, E. (2004). A ne w on-line resource for psycholinguistic studies, Journal of Memory and Language, 51 247Â–250. Tataryn, D.J., Wood, J. M., & Gorsuch, R. L. (1999). Setting the value of k in promax: A Monte Carlo study. Educational and Psychological Measurement, 59 384-391. Taylor, W., Lore, J., & Waldman, I. (1970). La tencies of semantic aphasics, stutterers, and normal controls to cloze item s requiring unique and nonunique oral responses. Proceedings, 78th Annual Convention, 75-76. Van der Merwe, A. (1997). A theoretical framework for the characterization of pathological speech sensorimotor control. In M.R. McNeil (Ed.), Clinical management of sensorimotor speech disorders (pp. 1-25). New York: Thieme. Van Petten, C. & Kutas, M. (1990). Interac tions between sentence context and word frequency in event-rela ted brain potentials. Memory Cognitive 18 380393. Watson, B., Freeman, F., Devous, M., Chapma n, S., Finitzo, T., & Pool, K. (1994). Linguistic performance and re gional cerebral blood flow in persons who stutter. Journal of Speech and Hearing Research, 37, 1221-1228.
74 Weber-Fox, C. (2001). Neural systems for sentence processing in stuttering. Journal of Speech, Language, and Hearing Research, 44 814Â–82. Weber-Fox, C., Spencer, R.M.C., Spruill, J.E., & Smith, A. (2004). Phonological processing in adults who stutter: Electr ophysiological and behavioral evidence. Journal of Speech, Language, & Hearing Research, 47, 1244-1258. WeberFox, C., & Hampton, A. (2008) Stuttering and natural speech processing of semantic and syntactic constraints on verbs. Journal of Speech, Language, and Hearing Research, 51 1058-1071. Wijnen, F. & Boers, I. (1994). Phonol ogical priming effect s in stutterers. Journal of Fluency Disorders, 19, 1-20. Wingate, M.E. (1964). A standard definition of stuttering. Journal of Speech and Hearing Disorders 29 484-488. Wingate, M.E. (1988). The Structure of Stuttering : A Psycholinguistic Approach New York, NY: Springer-Verlag. Wolpaw, J.R. & Penry, J.K. (1975). A tem poral component of the auditory evoked response. Electroencephalography and C linical Neurophysiology, 39 609-620. Yairi, E. & Ambrose, N.G. (2005). Early Childhood Stuttering: For Clinicians by Clinicians. Austin, TX: Pro-Ed.
76 Appendix A: Research Design While our research design was similar in many respects to that used by Jescheniak et al. (2002), some important ch anges were made. Like Jesc heniak et al. (2002), we had two general types of trials: Filler trials, and Experimental trials. Filler trials only involved naming a picture at a delayed latency, with no audito ry probe word presentation. Experimental trials consisted of a pict ure-to-be-named, followed immediately by an auditory probe word, followed by a cue to name the picture. Four different probe word conditions were included: 1) Trials for whic h the probe word was Semantically-Related (but not phonologically-related) to its corresponding picture; and 2) Trials for which those same probe words were reassigned, each to a different picture to which it was Semantically(and Phonologically -) Unrelated. Also included were 3) Trials for which the probe word was Phonologically-Related (but not semantically-related) to its corresponding picture; and 4) Trials for which those same probe words were reassigned, each to a different picture to which it was Phonologically(and Semantically-) Unrelated. In summary, we had five conditions in total: Semantically-Rel ated, SemanticallyUnrelated, Phonologically-Related, Phonol ogically-Unrelated, and Filler. Research Design Modifications We modified the task design used by Je scheniak et al. (2002) in two different ways. The first modification concerned the leve l of attention participants were required to pay to the auditory probe words. In Je scheniak et al. (2002) participants were instructed to explicitly remember the aud itory probe words. Their task included
77 Appendix A: (Continued) a Â“word checkÂ” at the end of each experiment al trial, requiring them to see a printed word and decide whether this word was the audi tory probe word they heard for that trial. We removed this requirement, and instructed participants to ignor e the auditory probe words. The task was made passive due to concern that the dual-task nature of naming while remembering a probe word might induce disproportionately high rates of stuttering in at least some of the participants who stut tered. Of key importance, Jescheniak et al. (2002) did not analyze ERPs recorded to Fi llers. However, we incorporated ERPs elicited by these trials into our analysis. Speci fically, the Filler (nami ng-only) trials were used to establish ERP baseline activity, i.e., ERPs elicited on Filler trials were seen as reflecting processing activities underway while participants waited to name pictures but did not hear auditory probe words. The critic al test of our ERP analysis was to determine whether each of the four experimental (probe word) conditions elicited ERP activity that differed from ERPs elicited by Filler trials. ERP differences between Experimental and Filler trials should reflect activity specifical ly related to processing the auditory probe words. The second change concerned the nature of our Phonologically-Related condition. For this condition we elect ed to use only auditory probe word s that shared the initial phoneme with their corresponding pictures In contrast, Jescheniak et al. (2002) used auditory probe words that heavily rhymed with their corresponding pictures. We changed the degree of phonological overlap because word-initial phonemes usually
78 Appendix A: (Continued) attract more stuttering than phonemes occ upying any other word position (Bloodstein, 1995). Priming word-initial phoneme s allowed us to investigate how the activation of phonologically-related words sharing only the initial phoneme operates in PWS versus PWNS.
79 Appendix B: Picture labels and priming words Picture name Freq. # of phon. Semantic probe words # of phon. Unrelated probe words Phonological probe words # of phon. Unrelated probe words balloon 1.946 5 pop 3 cave bagpipe 6 frame bat 2.708 3 cave 3 wood beach 3 hill Bed 5.136 3 sleep 4 game bank 4 frost bell 3.332 3 ring 3 open bruise 4 moon book 6.075 2 read 3 toad barrel 4 dive broom 2.197 4 sweep 4 ape bone 3 dent camel 3.258 5 desert 5 spoon concert 6 wheat cannon 1.946 5 ball 3 water key 2 drought cheese 3.466 3 mouse 3 neck chime 3 flag comb 1.792 4 brush 4 pop kite 3 trial desk 4.522 4 chair 2 movie drought 4 pole dog 4.754 3 cat 3 ring dent 4 melt door 5.958 3 open 4 time dive 3 patch fish 5.1 3 water 4 glass frost 5 concert football 3.526 6 game 3 chair frame 4 shrink fork 2.773 3 spoon 4 web flag 4 sand frog 2.303 4 toad 3 hand phone 3 bank giraffe 1.099 4 neck 3 cut juice 3 cycle glove 2.996 4 hand 4 desert ghost 4 beach hammer 2.485 4 nail 3 ball hill 3 bone map 3.714 3 road 3 bird mold 4 bruise match 4.06 3 fire 3 brush moon 3 key monkey 2.944 5 ape 2 road melt 4 barrel nose 4.407 3 face 3 camp nickle 4 chime penguin 1.792 7 bird 3 fire plate 4 mold pizza 1.099 5 food 3 dirt patch 3 walrus popcorn 0.693 6 movie 4 mouse pole 3 bagpipe
80 Appendix B: (Continued) saw 2 wood 3 cat sand 4 Kite scissors 1.609 5 cut 3 foot cycle 4 phone shoe 4.382 2 foot 3 read shade 3 twine shovel 1.609 4 dirt 3 car shrink 4 stand snake 3.178 4 bite 3 bathroom staff 4 web spider 2.079 5 web 3 nail stand 5 nickle tent 3.807 4 camp 4 face trial 4 juice toilet 3.367 5 bathroom 6 sweep twine 4 shade watch 3.714 3 time 3 food wheat 3 plate wheel 3.807 3 car 2 bite web 3 ghost window 5.303 5 glass 4 sleep walrus 6 staff
81 Appendix C: Behavioral Data Subjec t # Fille r Trial Unrelate d Semantic Semanticall y Related Unrelated Phonologica l Phonologicall y Related Stutterer / Fluent 1 75 37 38 38 37 s 2 76 38 38 38 38 s 3 76 38 38 38 38 s 4 75 38 38 38 38 s 5 76 38 38 38 38 s 6 76 37 38 37 38 s 7 75 38 38 38 38 s 8 72 37 37 36 37 s 9 76 38 37 37 38 s 11 76 38 38 38 38 s 13 76 38 38 38 38 s 14 76 38 38 38 38 s 15 74 37 37 37 37 s 33 76 38 38 38 38 s 16 76 38 38 38 38 f 17 75 38 38 38 38 f 18 76 38 38 38 38 f 19 76 38 38 38 38 f 20 73 38 37 35 37 f 22 75 38 36 37 37 f 23 76 38 38 38 38 f 24 76 38 38 38 38 f 26 76 37 38 38 38 f 27 76 38 38 38 38 f 29 76 38 38 38 38 f 31 73 37 37 37 38 f 34 76 38 38 38 38 f 35 76 38 38 38 38 f