J Psycholinguist Res (2009) 38:43–64 DOI 10.1007/s10936-008-9082-2 ORIGINAL ARTICLE
The Effect of Feedback Schedule Manipulation on Speech Priming Patterns and Reaction Time Dana Slocomb · Kristie A. Spencer
Published online: 12 September 2008 © Springer Science+Business Media, LLC 2008
Abstract Speech priming tasks are frequently used to delineate stages in the speech process such as lexical retrieval and motor programming. These tasks, often measured in reaction time (RT), require fast and accurate responses, reflecting maximized participant performance, to result in robust priming effects. Encouraging speed and accuracy in responding can take many forms, including verbal instructions and feedback, and often involves visually displayed RT feedback. However, it is uncertain how manipulation of the schedule of this RT feedback influences speech RT speed and, ultimately, the priming effect. This experiment examined the effect of visually presented RT feedback schedules on priming patterns in 20 older healthy adults. Results suggested that feedback schedule manipulation had a differential effect on reaction time, depending on the interstimulus interval between the prime and the target, but no effect on response priming patterns. Keywords
Reaction time · Feedback · Priming · Speech
Introduction Priming tasks are frequently used to delineate stages in the speech process such as lexical retrieval and motor programming. These tasks, often measured in reaction time (RT), require fast and accurate responses, reflecting maximized participant performance, to result in robust priming effects. Encouraging speed and accuracy in responding can take many forms, including verbal instructions and feedback, and often involves visually displayed RT feedback. However, it is uncertain how manipulation of the schedule of this RT feedback influences RT speed and, ultimately, the priming effect. There is evidence to suggest that different feedback schedules may motivate different levels of speed and accuracy and that the outcome may vary depending on the age of the participants. Therefore, this experiment
D. Slocomb · K. A. Spencer (B) Department of Speech and Hearing Sciences, University of Washington, 1417 NE 42nd Street, Seattle, WA 98105, USA e-mail:
[email protected]
123
44
J Psycholinguist Res (2009) 38:43–64
aims to assess the effect of visually presented RT feedback schedules on priming patterns specifically in older healthy adults. Speech RT Priming Priming tasks can vary considerably in terms of experimental and response parameters. A unifying theme, however, is that a bias is created toward a particular response (Lepine et al. 1989; Olivier and Rival 2002). When the expected target is encountered, participants respond faster than when the expected target does not appear. This increase in RT that occurs when an individual must switch from a prepared response to an unexpected response is referred to as an RT cost or switch cost (Lepine et al. 1989; Meyer and Gordon 1985). For example, when a bias is created for a prime–target match, RTs are typically faster for this condition than for a prime–target mismatch condition. This bias is often established through response probabilities. That is, the high probability of a prime–target match would prompt participants to prepare the primed response (Olivier et al. 1998; Olivier and Rival 2002). In general, “response priming” is the term used to refer to priming procedures where the prime and the target are either identical or unrelated (Versace and Nevers 2003) and a bias is created for the prime stimuli through the ratio of prime–target matches to mismatches (Rosenbaum and Kornblum 1982). The theoretical underpinnings of the priming effect are debated but generally include motor preparation or programming factors (Craighero et al. 1996; Dirnberger et al. 2000; Lepine et al. 1989; Leuthold and Jentzsch 2002; Meyer and Gordon 1985; Nakata et al. 2005; Olivier et al. 1998; Olivier and Rival 2002; Yaniv et al. 1990) and/or cognitivelinguistic factors (Arnott and Chenery 2001; McDonald et al. 1996; Milberg et al. 1999; Nakata et al. 2005; Shenaut and Ober 1996). An accurate and sensitive priming response requires participants to remain vigilant and to respond as quickly as possible. Evidence supporting this notion has been reported by Rogers and Storkel (1998) in their form-based priming study where the provision of feedback was manipulated in two experimental conditions. The authors found that speech RTs were significantly faster across both prime–target match and prime–target mismatch conditions when participants were provided with average RT feedback between experimental sets compared to receiving no feedback at all. Importantly, the presence of the phonologic similarity effect, which refers to interference in processing with similar sound structure among words, was contingent upon the feedback schedule. That is, in the no-feedback condition, the phonologic similarity effect was not found (Rogers and Storkel 1998). This finding suggests that response rate plays a role in the detection of such priming effects, as the slower responses in the experiment with no feedback did not yield the phonological similarity effect. The authors argue that only maximally rapid responses will allow observation of this priming effect because the primed response may be subject to rapid decay. These results indicate that further research into manipulation of feedback schedules in speech RT priming tasks is warranted. Unfortunately, no other speech priming investigations consider feedback schedule manipulation as a variable. Feedback Schedule Manipulation: Evidence From the Limb Motor Learning Literature Given the paucity of research on the influence of feedback schedule in speech RT studies, it is necessary to turn to the wealth of information in the limb motor learning literature. Augmented feedback, which is feedback that is not a natural result of the task (Salmoni et al. 1984), is generally divided into two types: knowledge of results (KR), or the degree to which specific aspects of motor performance were achieved, and knowledge of
123
J Psycholinguist Res (2009) 38:43–64
45
performance (KP), or information about movement quality (Schmidt and Wrisberg 2000; Young and Schmidt 1992). Though some authors have narrower definitions of KR that exclude visually displayed RT from this category (Salmoni et al. 1984), others indicate that RT does fall under KR (Hardesty and Bevan 1964; Ishikura 2005; Strang 1983). Regardless of type, limb control research strongly suggests that manipulation of feedback schedule has an effect on performance outcome (Anderson et al. 2001, 2005; Salmoni et al. 1984; Winstein and Schmidt 1990; Yao et al. 1994). Particularly germane to the motor learning literature is the influence of feedback schedules on the acquisition of a skill versus the retention or long-term learning of that skill. Though speech priming tasks do not reflect learning of a new skill, these tasks more closely parallel the acquisition phase in motor learning as the RT measure reflects performance during the task at hand, rather than how performance varies over time. Therefore, the notion that different feedback schedules can alter performance during acquisition in the limb motor learning literature suggests a possible influence of feedback schedules on speech RT priming patterns. In addition, studies of repetition priming (i.e., priming that occurs following repeated exposure to a stimulus) suggest that the underlying neural mechanisms of repetition priming may overlap with those of skill acquisition (Dennis and Schmidt 2003; Kirsner et al. 1993; Poldrack and Gabrieli 2001). Feedback schedules can vary considerably. Feedback given after each trial is termed 1:1 feedback. Summary feedback provides the participant with information about every trial in a graph or list format after a specified number of trials is completed or at the end of a trial set (as opposed to after each trial). Average feedback, in contrast, gives just one indicator that is considered to be the average performance on multiple trials after a specified number of trials are completed or at the end of a trial set. Delayed feedback refers to information about a specific, earlier trial provided after intervening trials. For example, feedback for trial one might be displayed after trial three. With faded feedback, the frequency of feedback decreases as the trial block progresses. Feedback can also be provided after a specified number of trials, for example, every five trials, and consist of information regarding the trial directly preceding the feedback, delayed feedback from earlier trials, summary feedback of the previous five trials, or average feedback of the previous five trials (i.e. a combination of feedback types or methods). A distinction is made between absolute and relative frequency of feedback, with absolute feedback being the total number of times feedback is provided during a session and relative feedback being the ratio of total number of times feedback is provided over the total number of trials (Salmoni et al. 1984). Taking this into account, 1:1 feedback can also be referred to as 100% feedback and feedback every specified number of trials can be referred to by the specific relative percentage (number of feedback trials over the total number of trials) or as lower frequency feedback. Though it is possible to keep the absolute frequency of feedback consistent while manipulating the relative frequency, this alters the total number of trials, confounding the effects of feedback schedule with the amount of practice (e.g. more trials) (Adams et al. 2002; Salmoni et al. 1984; Winstein and Schmidt 1990). Feedback provided after every trial, or 1:1 feedback, was historically thought to promote the best motor learning outcomes (Bilodeau 1966). However, more recent research that differentiates between acquisition and retention periods reveals divergent effects of feedback schedules on acquisition versus retention (Ishikura 2005; Lavery 1962; Salmoni et al. 1984; Schmidt et al. 1990; Winstein and Schmidt 1990; Yao 2003; Yao et al. 1994). Aiming and striking tasks are commonly used in feedback schedule manipulation experiments, with accuracy outcome measures often reflecting better performance during acquisition with 100% feedback versus lower frequency feedback. These studies typically reveal an inverse
123
46
J Psycholinguist Res (2009) 38:43–64
relationship between performance during acquisition and performance during retention in that lower frequency feedback during acquisition promotes better performance long term (retention and/or transfer phases) despite poorer performance with these schedules during acquisition (Anderson et al. 2001, 2005; Lavery 1962; Salmoni et al. 1984; Schmidt et al. 1990; Winstein and Schmidt 1990; Yao 2003; Yao et al. 1994). Some studies reveal an optimized level of lower frequency feedback (e.g. after 3 trials or after 5 trials) with particular motor tasks (Ishikura 2005; Schmidt et al. 1990; Yao et al. 1994). These studies, as well as others revealing patterns of acquisition and retention that do not follow this general norm, suggest that additional factors, such as task complexity, the participant’s previous experience with the task, type of task, and/or age, interact with feedback schedule (Anderson et al. 2005; Guadagnoli et al. 1996; Ishikura 2005; Rice 2003; Wulf et al. 1998). Age Effects: Evidence From Limb Motor Learning Literature Age emerges as a potential significant variable with respect to motor learning. Carnahan et al. (1996) investigated the use of KR by younger (average age 22.5 years) versus older (average age 75 years) adults in learning a key press task in which participants received feedback after every trial or summary feedback after every five trials. In terms of retention, the results indicated that younger and older adults use KR similarly (Carnahan et al. 1996). However, the authors noted that during the acquisition phase, the older adults’ performances were more variable with 1:1 feedback than with summary feedback. In a related study, Jarus (1995) manipulated feedback schedule with older (average age 59 years) and younger (average age 23 years) healthy adults to determine the effects on sensory awareness calibration, which involved assessing the incline of ramps. Results substantiated limb motor learning findings in that better retention and transfer was achieved by participants who had lower frequency feedback during acquisition (Jarus 1995). However, the results diverged from the literature in that the younger group with lower frequency feedback performed better than the younger group with 100% feedback during the acquisition phase, while the older groups showed the reverse (yet more common) pattern of better performance in the group with 100% feedback in comparison with the group that received lower frequency feedback during acquisition (Jarus 1995). Moreover, it is well known that older adults have slower overall RTs than younger adults (Light and Spirduso 1990; Spirduso 1975; Yan et al. 1998), which may indirectly influence priming patterns. In contrast, many studies support the notion that older adults use feedback similarly to younger adults (Rice 2003; Swanson and Lee 1992; Wishart and Lee 1997), even when controlling for overall cognitive slowing (Giffard et al. 2003; La Voie and Light 1994). Motivational Properties of RT Feedback: Evidence from Limb RT Literature Beyond the limb motor learning literature, information regarding feedback schedule manipulation can be gained from limb RT tasks. Worell and Worell (1963) investigated the effect of general instructions on RT, by standardizing two sets of instructions, one with speed emphasized and one with speed not mentioned at all. Results indicated that verbal instructions that emphasized speed motivated significantly faster RT performance (Worell and Worell 1963). Strang (1983) explored the motivational properties of KR, defined as median RT provided during a baseline phase, during a key press task. One group of participants was given one-time KR, and the other group was given no-KR; neither group was given instructions regarding speed. He found that participants who received the one-time KR had significantly faster RTs
123
J Psycholinguist Res (2009) 38:43–64
47
(Strang 1983). The addition of a third condition, called “the promise of future knowledge of results (PKR),” in which participants were shown a display informing them that they would be shown their median RT at the end of the set, was included to further explore this aspect of motivation. The results indicated that those in the PKR group performed faster than the no-KR group and similarly to the one-time KR group. Thus, the author hypothesized that the one-time KR exposure served to create anticipation of future feedback thus motivating the participants to increase their speed (Strang 1983). In contrast, Crabtree and Antrim (1988) argue against the use of RT feedback in limb control studies. The authors posit that [I]n a reaction time assessment, feedback must be kept to a minimum because learning through direction, reinforcement, and motivation should not be reflected in the RT measures. Although true RT is not generally improved by feedback, no knowledge of results should be given to the subject because the effects on the motivation and/or stress levels of the subject vary. Since motivation is often reduced without some information, however, subjects should be told at standard times that (1) they are performing well and (2) they must continue to do their very best. (p. 365) Despite idiosyncratic responses to reaction time feedback, it is clear that verbal instructions and/or KR can significantly influence RTs (Rogers and Storkel 1998; Strang 1983; Worell and Worell 1963). Feedback Schedule Manipulation: Evidence from the Speech Motor Learning Literature Limb motor learning and RT investigations have provided insight into the effects of feedback schedules on RT and overall task performance. Further support for the examination of feedback schedule manipulation in speech priming paradigms stems from the speech motor learning literature. Adams and Page (2000) investigated the manipulation of feedback schedule using a novel speech task (i.e. the use of a predetermined slow rate of speech). Consistent with findings in the limb motor learning literature, the authors found that feedback given every five trials elicited better retention than feedback given after every trial. However, little difference was found in performance between the two feedback schedules during the acquisition phase. The authors indicate that this outcome, which diverges from the limb motor learning literature, could reflect differing measurement techniques. In a follow up study, Adams et al. (2002) explored the effects of the manipulation of feedback during a similar speech motor learning task in participants with Parkinson’s Disease. At the beginning of the acquisition phase, error scores were lower for the participants receiving feedback every trial. By the end of the acquisition phase, participants receiving feedback every trial and every five trials performed similarly (Adams et al. 2002). The authors note that the effects of relative versus absolute frequency could have affected this outcome, as early on, the group receiving feedback less frequently was exposed to significantly less feedback overall. These results are consistent with the limb motor learning literature, in that reduced frequency of feedback led to better retention (Adams et al. 2002). However, the authors acknowledge the need for further research to determine whether this pattern applies to other speech tasks, the role of task complexity may play, the application of the findings to other motor speech disorders, and whether these findings can be applied to treatment paradigms (Adams et al. 2002). Steinhauer and Grayhack (2000) investigated the manipulation of feedback in the learning of a vowel nasalization task. Ten healthy participants were randomly assigned to one of three groups, receiving feedback for 100% of trials, for 50% of trials, or for none at all. The feedback was a nasalance score, indicating the percentage of nasalance achieved
123
48
J Psycholinguist Res (2009) 38:43–64
during the previous trial. The results were similar to the limb motor learning literature, as a decrease in performance in the retention phase correlated with the highest frequency of feedback schedule (Steinhauer and Grayhack 2000). However, the results diverged from the limb motor learning literature in that this decrease in performance with the highest frequency feedback was also found during the acquisition phase. Possible explanations for reduced performance with increased KR include that it may (1) detract from intrinsic feedback, (2) lead to increased variability in performance due to overcorrection, or (3) create an overdependence on KR (Steinhauer and Grayhack 2000). The findings also revealed similar performance during acquisition in the groups that received 50% feedback or none at all, representing an even more significant deviation from the limb motor learning literature (Steinhauer and Grayhack 2000). That is, providing some feedback produced the same results as providing no feedback, indicating that this feedback had no influence on performance. The authors postulated that the similar performance in groups that received 50% feedback and no feedback, as well as the detrimental effects of 100% feedback, may be explained by dynamical theories of motor control; for this particular task the feedback was not necessary for learning so the feedback was either ignored (50% feedback) or actually interfered (100% feedback). These discrepancies underscore the need for further research into feedback manipulation in speech tasks. Though the speech motor learning literature does correspond with the limb motor learning literature in some respects, there is evidence to suggest that feedback schedule manipulation in the speech realm may yield different outcomes (Adams and Page 2000; Steinhauer and Grayhack 2000). Potential Influences on Speech Priming Patterns Another factor affecting priming patterns is the amount of time between the prime and the target, or interstimulus interval (ISI). In a study investigating the effects of ISI on response priming patterns (Spencer and Wiley 2008), the results suggested that priming patterns did indeed vary as a result of ISIs. Specifically, the robustness of the priming effect decreased with an increase in ISI. This finding appeared to be due primarily to faster reaction times in the prime–target match condition during the short ISI conditions (i.e., 50 and 150 ms) versus the longer ISI conditions (i.e., 1,000 and 2,000 ms). Thus, speed of response may have a pronounced effect on priming patterns if the prime–target match and prime–target mismatch conditions are differentially affected. The age of the participant might also have a marked effect on outcome, as younger adults have faster overall RTs than older adults (Light and Spirduso 1990; Spirduso 1975; Yan et al. 1998). Moreover, during skill acquisition, there is some evidence to suggest that younger participants may perform faster with lower frequency feedback while older participants may perform faster with higher frequency feedback. The limb and speech motor learning literature align to suggest that if the current task more closely resembles the retention phase, faster RTs are expected to correspond with lower frequency feedback, even across age groups. For the present study, targeted participants were older healthy adults (40+), as this is the typical age range used to match individuals with neurological impairment who are studied with response priming (Spencer and Rogers 2005). Currently, there is a dearth of knowledge regarding the effects of feedback schedule on speech RT performance. This study will investigate the effect of feedback on overall RT and response priming patterns in older healthy adults. As there is no direct parallel to speech reaction time priming tasks from the limb or speech motor learning literature, it is difficult to establish hypotheses based strictly on this research. There is some indication that results will
123
J Psycholinguist Res (2009) 38:43–64
49
more closely parallel findings from acquisition stages (Dennis and Schmidt 2003; Kirsner et al. 1993; Poldrack and Gabrieli 2001). Looking to the speech motor learning literature, we would expect that RTs might be similar across feedback schedules (Adams and Page 2000) or be faster with lower frequency feedback (Steinhauer and Grayhack 2000) which would thus lead to the potential for more robust priming patterns. These predictions are the opposite of what might be proposed from acquisition data for limb motor learning tasks. That is, faster RTs would then correspond with the highest frequency feedback schedule (i.e., 1:1 feedback). Methods Participants Twenty healthy adult volunteers participated in this study, consisting of 14 females and six males, ages 44–79 years (M = 57.6, SD = 9.7), with 12–21 years of education (M = 16.2, SD = 1.8). Participants reported no history of neurological disease or speech and language difficulties, and had normal or corrected-normal vision. Fifteen participants passed a hearing screening at 35 db at 500, 1,000, 2,000, and 4,000 Hz. Three participants had elevated thresholds at one frequency and two participants had elevated thresholds at two frequencies. All participants followed instructions without difficulty, so these elevated thresholds were not a concern for this study. All participants passed a screening to confirm absence of stroke history using the Questionnaire for Verifying Stroke-Free Status (QVSFS) (Jones et al. 2001). Participants were consented before study participation in accordance with the Institutional Review Board of the University of Washington. Procedural Overview The protocol was an extension of the work of Spencer (2003) and Spencer and Rogers (2005). Testing occurred in a well-lit, quiet room over one 1 3/4 h session. The questionnaire, hearing and vision screening, and QVSFS were administered first. Next, participants viewed word pairs (primes followed by targets) on a computer screen and were instructed to read the target word aloud as quickly as possible. RTs were measured as the time between the appearance of the target word and the initiation of speech. Practice tasks were used to ensure participant understanding and to familiarize them with the task. To control for the different voice onset times of each target word, speech RTs for each target word were measured during a simple reaction time (baseline) task where participants were asked to read the word on the screen as quickly as possible once it became underlined. The experimenter scored all responses online for errors and taped all sessions for reliability purposes. Apparatus Presentation of the stimulus and response timing were achieved using a personal computer (Micron Millennia Max GS 133 with an 18 inch Trinitron monitor). Word presentation, registration of speech onset, and calculation of reaction times were managed by E-Prime software (version 1.1; www.pstnet.com/E-prime/e-prime/htm) and a Serial Response Box (SRB) (Model #200a) by Psychology Software Tools. The SRB operated with 0.6125 ms timing accuracy at 1,600 characters per second. The stimulus presentation triggered the timing mechanism and stopped with the onset of voicing from the subject’s response. A signal
123
50
J Psycholinguist Res (2009) 38:43–64
was obtained from an accelerometer positioned inferior and lateral to the subject’s thyroid cartilage and connected to an ICP Sensor (PCB Piezotronics model 480C02). The accelerometer detected the voicing onsets which then activated the voice key of the SRB. Participant responses were recorded by a Tascam DA 302 DAT recorder. Stimuli A total of 32 words made up the stimuli set, with half consisting of one-syllable words and the other half consisting of two-syllable words. Sixteen of these words served as prime words (eight one-syllable and eight two-syllable) in both the prime–target match and prime–target mismatch conditions. For the prime–target match condition, the 16 prime words served as both the prime and the target (e.g., shopper-shopper). For the prime–target mismatch condition, 16 novel target words (eight one-syllable and eight two-syllable) were created. Each novel target shared a rime with its prime pair word but had a different onset in the mismatch condition (e.g. shopper-chopper). All stimulus words were low-frequency based on the criterion of less than or equal to 20 instances/million (Francis and Kucera 1982) and were selected based on the notion that more robust priming effects are found with low-frequency stimuli than with high frequency stimuli (Kirsner et al. 1993; Versace and Nevers 2003). As the stimuli were presented orthographically, all words were evaluated for spelling regularity, which could potentially impact speed of retrieval, as it has been suggested that irregularly spelled words bypass phonological grapheme to phoneme conversion rules (Burke et al. 2000). Hanna et al. (1966) orthotactic probability calculations were used to calculate regularity indices, representing the average percentage of the time each grapheme in a word represents a specific phoneme based on word position and syllable stress. The average regularity index for the words used in the present study was 68.4% (range: 46–84.6%). By this method, four words fell below a regularity index of 60% (considered top cutoff for a word to qualify as irregularly spelled) (Calkins 2003). These words (zipper, rake, showing, and rowing) were deemed acceptable in terms of orthography to pronunciation patterns by three independent judges and subsequently remained in the stimuli set. Experimental Design and Procedure Three different feedback schedules were used (1) 1:1 feedback, (2) average feedback after every five trials, and (3) average feedback at the end of the trial set. As Rogers and Storkel (1998) investigation revealed significantly slowed RTs combined with no priming effect in a no-feedback condition, the no-feedback condition was not further explored here. Also, sets were quasi-randomized so that half of the participants received an average feedback at the end condition first and the other half received this condition last. This randomization was done to determine if reaction times and the priming effect for the least frequency feedback condition was influenced by participants receiving earlier sets with higher frequency feedback. For the priming protocol, the participants were asked to prepare to say the word that appeared on a red screen (prime) and read aloud the word that appeared on a green screen (target) as quickly and accurately as possible. They were informed that most of the time, the two words would be the same. Speed was emphasized in these verbal instructions and participants were told when they would receive information about their speed (after every trial, after every five trials, or after all of the trials). Participants also were informed whether there would be a delay between the red screen and the green screen (i.e., two ISI conditions).
123
J Psycholinguist Res (2009) 38:43–64
51
Each trial began with an alerter signal on a red screen followed by a prime word which remained on a red screen for 1,000 ms. A blank red screen followed for an ISI of 250 ms in the “no-delay” condition and 2,000 ms in the “delay” condition. Next, the target word was presented on a green screen for 1,000 ms. In the 1:1 feedback condition, the green screen was followed by a white screen (700 ms duration) showing response RT in milliseconds for the previous trial. In the 1:5 feedback condition, the green screen went directly to the next trial, beginning with the red screen and the alerter. After the fifth trial, a white screen (2,000 ms duration) followed the green screen displaying the average RT in milliseconds for the past five trials. In the average feedback at the end of the trial set, the green screen was followed by the red alerter screen of the next trial for all trials except the last one of the set. That trial was followed by a white screen (2,000 ms duration) displaying average RT in milliseconds for all of the trials in the set. During development of the trial sets, the duration of the feedback display was 700 ms in all conditions. However, during pilot testing, it was determined that this duration worked well for the 1:1 condition, when feedback was frequent, but during the lower frequency feedback conditions, 700 ms was too rapid for participants to process. The protocol consisted of six sets of 65 trials, two sets for each of the three feedback schedules (1:1, 1:average 5, average at the end of the set), for the delay (ISI 2,000 ms) and no delay (ISI 250 ms) conditions. Order of presentation was quasi-randomized across participants. To bias participants toward the match condition, the target matched the prime in 49 trials (75%) and did not match the prime in 16 trials (25%). See Appendix for the list of stimuli and distribution of matched versus mismatched pairs. Word pair presentation was pseudo-randomized to ensure that primes were not repeated over any two successive trials. To familiarize participants with the task, they were given practice with 10-pairs of nonexperimental stimuli, with the same to switch ratio approximately equivalent to the experimental tasks (mismatch condition on 3/10 trials). One set of 10 practice trials combined the delay (ISI of 2,000 ms) condition and the 1:1 feedback conditions. The second set of 10 practice trials combined the no-delay (ISI of 250 ms) and average feedback after five trials conditions. An explanation of the feedback and ISI conditions was provided to the participants. Speech Baseline Task Due to variation in voice onset time for each stimulus word, a simple reaction time baseline task was administered to allow for normalization of the data. In this task, participants were shown a word on a white screen and asked to say it as quickly as possible once it became underlined and the letters changed from black to green (the underlining and turning green occurred simultaneously, referred to as the go-signal). Each of the experimental target words (N = 32) were presented in random order in each of four sets to generate sufficient exemplars of each token for each participant for normalization. The go-signal was given at random intervals from 750–2,000 ms to foil anticipatory reactions (Deger and Ziegler 2002). Feedback was given after every trial for the baseline task in order to encourage vigilance and speed in responding. As the study was designed to explore the effects of feedback schedule manipulation on reaction time, the baseline task was given after the experimental tasks to prevent bias toward the 1:1 condition. The four sets were presented in random order. Scoring and Reliability Participant speech errors during experimental and baseline tasks were recorded online by the author and were defined using a multidimensional scoring system. Thirty-three percent
123
52
J Psycholinguist Res (2009) 38:43–64
of total trials (two random sets for each participant) were independently scored by a trained graduate student from audio taped recordings. For the experimental sets, speech RTs of less than 250 ms were considered outliers and were removed from the data set. The baseline task was a simple RT task which enabled participants to respond more quickly; therefore, speech RTs of less than 150 ms were considered outliers for these sets. Next, individual outliers, calculated as three standard deviations above or below each participant’s mean reaction time, were removed from the merged baseline sets as well as each experimental set. Data Normalization For data normalization purposes, at least three baseline RTs for each token were required to calculate the mean and standard deviation for each token. Due to speech errors, some participants had only two RTs for a token. To remedy this, a third RT, or dummy variable, was created by averaging the two RTs. All subjects had at least two tokens for each baseline word. Three of the 20 participants required the use of at least one dummy variable and none of these required the use of more than two dummy variables. Once errors and outliers were removed from the data, the data was normalized to minimize token onset variability as well as individual variability (Kessler et al. 2002). Z-scores were calculated by subtracting the experimental RT for each token from the mean RT for that token in the baseline data and dividing by the standard deviation for that token in the baseline data. These transformed RTs (T-RTs) then represent the difference between the RT under the various experimental priming conditions and the simple RT as measured during baseline. Thirty-three percent of all normalized data (two sets for each participant) were re-calculated for reliability purposes by a trained graduate student.
Results Error Rates The highest occurring error rates in the experimental sets included self-corrections (41%), early responses (21%), late responses (18%), and perseverations (14%). Vowel errors, sound errors, missed trials, sound insertions, manner errors, sound omissions, and abandoned responses each made up 1% or less of total errors. The accelerometer failed to detect a response on
