A randomized crossover single-case series comparing blocked versus random treatment for anomia

ABSTRACT

The motor learning literature has demonstrated that blocked practice facilitates better acquisition of motor skills, whereas random practice facilitates retention and transfer. The verbal learning and memory literature offers similar evidence. The purpose of this study was to investigate effects of blocked versus random practice in treatment for anomia. The study used a single site, randomized crossover design, with two replicated experimental phases (two blocked and two random) for each of 10 individuals with anomia. Each phase consisted of a cued picture-naming treatment. Individual treatment and maintenance effects, as well as weighted averages and group effects, were calculated using Tau-U based on the proportion of correctly named probes. Nine of 10 participants demonstrated treatment effects during each of the four phases. Acquisition was comparable for blocked and random practice. Maintenance effects were observed following seven blocked phases of treatment and 12 random phases of treatment across participants. For four of 10 participants the random schedule resulted in better maintenance of trained items. Although further research is needed, the present data suggest that for word retrieval treatment with multiple repetitions of the same items, a random presentation may benefit maintenance of treatment gains.

KEYWORDS:

• Aphasia
• anomia
• therapy
• treatment
• naming

Introduction

Treatment schedules in aphasia therapy

Recently, there has been growing interest in systematically investigating ingredients of rehabilitation treatments to determine which ingredients are essential and which are nonessential (Hart et al., 2019; Van Stan et al., 2019; Zanca et al., 2019). For example, effects of within- and between-session treatment schedules on aphasia therapy outcomes have been evaluated. Previous studies have examined factors such as intensity (Godecke et al., 2021; Dignam et al., 2015; Cherney, 2012), dosing (Harvey et al., 2021; Pierce et al., 2021; Harnish et al., 2014), and feedback (Fillingham et al., 2006; McKissock & Ward, 2007). However, it is still unclear which factors or combinations of factors are most impactful. Moreover, given time constraints on clinicians, it is especially important to maximize the efficiency of therapeutic tasks. Thus, researchers can help determine the optimal aphasia therapy practice schedule(s) that yield the greatest rehabilitation outcomes while maximizing efficiency.

Practice schedules in motor rehabilitation

The influence of different practice schedules has been thoroughly studied in the learning and motor rehabilitation literature. The principles of motor learning explain how people acquire and retain motor skills (Maas et al., 2008). These principles address rehabilitation concepts including distribution of practice (i.e., massed versus distributed/spaced), practice schedule (i.e., blocked versus random), target complexity (i.e., simple versus complex movement), and feedback frequency (i.e., frequent versus sporadic), among others.

A principle of motor learning that holds particular relevance for rehabilitation is whether practice schedules for treatment targets should be presented in blocked or random sets. Blocked practice refers to training a specified target or set of targets repetitively before moving on to new targets for subsequent trials. Random practice refers to training targets in a random fashion from trial to trial. Thus, in blocked practice, the individual may anticipate the next target(s) to be trained in most trials, but in random practice, the individual can never anticipate the next target.

The effects of these different practice schedules have been investigated in skills related to sports (Cheong et al., 2012; Menayo et al., 2010; Travlos, 2010) and dance sequences (Bertollo et al., 2010), among other activities (Dwyer et al., 2004; Lavis & Mitchell, 2006; Price et al., 2008; Rivard et al., 2015). In addition to examining the effects of practice schedule on performance during training, several studies have also investigated retention or transfer of treatment effects. Previous research demonstrates that blocked practice facilitates acquisition, whereas random practice enhances retention of motor skills (Shea & Morgan, 1979). Although a random practice schedule may eventually facilitate superior retention of motor skills over a blocked practice schedule, random practice may also result in poorer performance and slower acquisition during training (Magill & Hall, 1990). This effect of random practice slowing acquisition but ultimately enhancing retention of treatment gains is known as contextual interference, developed out of Battig’s findings of “intratask interference” during verbal learning (1966; 1972; Magill & Hall, 1990).

There have been three nonexclusive theoretical explanations for contextual interference in the motor learning literature: the elaboration-distinctiveness (Wright et al., 1992), deficient processing (Callan & Schweighofer, 2009), and forgetting-reconstruction hypotheses (Lee & Magill, 1985). The elaboration-distinctiveness hypothesis posits the more elaborate processing that would occur in random practice, whereby similarities and differences between trial items are highlighted during inter-trial intervals, allows for more complete learning (Wright et al., 1992). The deficient processing hypothesis asserts that blocked repetitions would result in reduced attention and reduced rehearsal in later presentations (Callan & Schweighofer, 2009), whereas random practice would not allow for habituation, resulting in better retention. Most notably, the forgetting-reconstruction hypothesis asserts that motor plans are called up (i.e., reconstructed) each time a new presentation in random practice occurs because it has been forgotten since the previous presentation. This processing during random trials is in direct opposition to that occurring during blocked practice where the same motor plan may be maintained in a motor memory buffer, avoiding the need for complete recall and reconstruction to take place. It is the repeated recall of motor plans during random practice that is believed to enhance motor learning and skill retention by resulting in stronger memory representations.

Despite the consensus of blocked practice facilitating acquisition and random practice facilitating maintenance in the motor learning literature, the results of studies examining blocked versus random practice schedules in motor-speech rehabilitation have been mixed. Knock and colleagues (2000) used an alternating treatments design pairing different targets with random or blocked practice schedules in two individuals with co-occurring severe apraxia of speech and aphasia. They found that random practice facilitated retention of acquired speech sounds at one and four weeks post-treatment. However, two additional case studies involving participants with apraxia of speech found no clear pattern of blocked or random practice schedules facilitating better retention of accurate speech sound production in trained words (Wambaugh et al., 2014; 2016). A follow-up group-level study found significantly higher percent change scores for items treated in the random condition compared with the blocked condition, but no difference in the effect sizes for the two conditions was observed (Wambaugh et al., 2017). Maas and Farinella (2012) found similarly mixed results in their treatment study involving four children with childhood apraxia of speech. Taken together, these findings demonstrate that the impact of blocked versus random practice schedules has not been as consistent when applied to motor-speech activities as in non-speech motor activities. Moreover, it remains unclear how practice schedule may impact rehabilitation of linguistic skills, such as word-retrieval in individuals with aphasia.

Practice schedules in verbal learning and memory

It is feasible that rehabilitation of confrontation naming may benefit in different ways from blocked and random practice schedules due to intratask (i.e., contextual) interference (Battig, 1966; 1972), much like motor rehabilitation. However, it is unclear precisely how and to what extent these different schedules may facilitate recovery. Knowing which target is coming next in a blocked schedule may help strengthen associations between the presented prompt and the target, thus improving automatic spreading activation (e.g., Collins & Loftus, 1975; Dell et al., 1997; Indefrey & Levelt, 2004) and ultimate selection of the target. However, consecutive trials with the same target may only initially result in activation of the target through lexical retrieval (e.g., Dell et al., 1997; Indefrey & Levelt, 2004), whereas trials following the first may not require complete lexical retrieval, instead accessing the target from a temporary phonological buffer where it may be stored after at least partial activation (Baddeley & Hitch, 1974; 2019; Gathercole & Baddeley, 1993). If the latter is true, a blocked schedule may waste treatment time by limiting the number of times a target is fully retrieved, whereas random practice may provide greater benefit by increasing the number of opportunities to fully retrieve the target.

Evidence from the verbal learning and memory literature lends support to the idea that blocked and random practice schedules may differentially impact word retrieval rehabilitation. Perhaps the most relevant evidence comes from studies of spaced retrieval. Cepeda et al. (2006) found in their meta-analysis that spaced practice is better for retrieval of lexical items than massed practice in neurotypical adults and that longer inter-study intervals (ISIs; the time between successive exposures to the same item) favour longer retention. Although practice distribution and schedule are different concepts, blocked schedules lend themselves to massed practice, whereas random schedules lend themselves to distributed practice. In a blocked schedule, successive trials for the same target occur immediately after one another. In a random schedule, the ISIs for each target item are longer due to intervening trials. Therefore, a random schedule, with its spaced effect, should yield longer ISIs and thus improved retention.

Cepeda et al. (2006) also found that ISI length corresponds with the length of the retention interval (i.e., a shorter retention interval would benefit from shorter ISIs). As such, they discounted the deficient processing hypothesis to explain their results because it suggests that longer ISIs always lead to improved recall, regardless of retention interval length. Although the authors did not discuss elaboration-distinctiveness (Wright et al., 1992) or forgetting-reconstruction (Lee & Magill, 1985), they suggested three separate theoretical accounts for their findings: encoding variability (Glenberg, 1979; Melton, 1970), consolidation (Wickelgren, 1972), and study-phase retrieval (Braun & Rubin, 1998; Thios & D’Agostino, 1976).

Most notably, consolidation theory (Wickelgren, 1972) suggests that later presentations of an item are facilitated by the previous presentation’s consolidation into long-term memory by inheriting its state or progress toward consolidation. Accordingly, longer ISIs lead to greater consolidation between study episodes. This may fit with a linguistic version of the forgetting-reconstruction hypothesis (Lee & Magill, 1985), as longer ISIs would provide an opportunity to sufficiently forget the target word by clearing the phonological buffer (Baddeley & Hitch, 1974, 2019; Gathercole & Baddeley, 1993) and therefore necessitate full retrieval of the target on subsequent trials, while still allowing those subsequent trials to benefit from previous retrievals that have made progress toward consolidating into long-term memory. Unlike forgetting-reconstruction theory (Lee & Magill, 1985), the consolidation hypothesis (Wickelgren, 1972) indicates that if too much time passes between study episodes, there will be no consolidated trace of the item left in long-term memory for the subsequent study episode to inherit. Blending the theories, an ideal ISI would be an amount of time that sufficiently clears the phonological buffer, requiring full retrieval upon subsequent recall attempts, and maximizes consolidation into long-term memory without degrading the trace (e.g., the word-retrieval path) completely.

To summarize the previously presented studies, blocked schedules have promoted acquisition of nonspeech motor skills, while random practice schedules have been shown to facilitate retention. There has been limited, mixed evidence from studies targeting motor speech rehabilitation. Although principles of motor learning may inherently apply more to the motor aspects of communication, there is evidence that phonological, semantic, and lexical retrieval processes are interactive (Dell & O’Seaghdha, 1992) and integrate with motor control during speech production (Walker & Hickok, 2016). Moreover, evidence from the verbal learning and memory literature suggests that blocked and random practice schedules may have a similar effect on lexical retrieval. Blocked schedules may initially facilitate acquisition due to the efficiency of retrieving targets from a phonological buffer, but random schedules may facilitate maintenance due to the repeated completion, or reconstruction, of the lexical retrieval process. The use of similar practice schedules in naming treatment has been explored in both the semantic dementia and post-stroke aphasia populations, though to a limited extent.

Practice schedules in naming therapy for semantic dementia

Semantic dementia, also known as the semantic variant of primary progressive aphasia, is a neurodegenerative condition primarily affecting semantic knowledge and leading to anomia, among other deficits (Gorno-Tempini et al., 2011). For a review of word-retrieval treatments in semantic dementia and other forms of primary progressive aphasia, see Jokel et al. (2014). In an anomia treatment study for three participants with semantic dementia, Hoffman et al. (2015) found that introducing variability into the naming treatment by altering the presentation sequence led to greater retention of trained items at follow-up for two participants, despite a similar benefit of the variable and fixed-order presentation sequences during training. Although this study was not structured to specifically assess blocked and random practice schedules, the results support the notion that a random schedule may facilitate maintenance due to greater and variable spacing between trial items, requiring repeated lexical retrieval. However, a study that directly assessed the benefit of spaced retrieval on naming in a single participant found no benefit of spacing recall intervals (Bier et al., 2009).

Blocked vs. random practice in aphasia therapy

One key study exploring within-session spacing of trials during a naming treatment for individuals with post-stroke aphasia demonstrated that spaced (e.g., random) practice facilitated both better acquisition and maintenance than massed (e.g., blocked) practice following a single training session for each treatment condition (Middleton et al., 2016). These results are consistent with the notion that random practice should facilitate maintenance, but contradict the idea that blocked practice should facilitate acquisition. However, their treatment timeline was brief, with post-treatment assessment the day after the single training session and follow-up assessment one week later. Therefore, it is possible that the limited time between testing acquisition and maintenance may have contributed to the similar results at both time points.

The present study will contribute to the limited body of evidence regarding treatment schedules during naming treatment for individuals with post-stroke aphasia, providing further insight on the immediate and long-term benefits of blocked versus random practice schedules. Further investigation of blocked and random practice schedules in aphasia treatment is important for two reasons. First, we must better understand how individuals with aphasia optimally acquire and maintain lexical items that are initially inaccessible to them. The verbal learning and memory literature discussed above is based on findings from neurotypical performance. We cannot assume that individuals with aphasia will perform the same way. The second reason to test our hypothesis is that identifying the best schedule for a naming treatment to facilitate acquisition and maintenance will improve our ability to maximize efficiency during the limited time available in clinical therapy sessions. Therefore, the present study investigated the effects of blocked and random stimulus presentation during a cued picture-naming treatment (CPNT) to determine whether the order in which items are targeted within a session influences participant outcomes. Based on the converging evidence from the fields of motor rehabilitation and verbal memory, we predict that the blocked schedule will facilitate initial acquisition of trained items, while the random schedule will facilitate maintenance of items following a CPNT designed to improve word-retrieval.

Materials and methods

Participants

This study received approval from the local Institutional Review Board before enrolling participants, who were recruited from the community. Inclusion criteria included 1) age between 19 and 90, 2) a raw score greater than 3 and less than 40 out of 60 (between the 10th and 60th percentile rankings) on the Boston Naming Test (BNT; Kaplan et al., 1983), 3) presence of aphasia as determined by an Aphasia Quotient score of less than 94.7 on the Western Aphasia Battery (WAB; Kertesz, 1982), 4) a score no less than two standard deviations below the norm on the Auditory Verbal section of the WAB (Kertesz, 1982), 5) six or more months post-onset from the most recent stroke, 6) pre-morbid right handedness as ascertained by the Edinburgh Handedness Inventory (Oldfield, 1971), and 7) status as a native English speaker. Based upon review of medical records and interview of the patient and a caregiver, participants were excluded if they had 1) a history of diffuse injury or disease of the brain (e.g., closed head trauma with more than 6 h of unconsciousness, probable Alzheimer’s disease), 2) significant depression as indicated by the Beck Depression Inventory (Beck et al., 1961) or 3) uncorrected vision or hearing. Individuals participated in a brief examination of apraxia of speech (AOS; McNeil et al., 1997). Two or three certified speech-language pathologists (SLPs) otherwise unaffiliated with the study provided ratings of AOS presence and severity based on clinical judgement. Individuals who were rated with profound AOS (i.e., a score of seven out of seven) by two SLPs were excluded. All assessments and treatments were administered in a quiet university laboratory setting by an SLP or one of two SLP graduate students, supervised by a certified SLP.

Design and assessment procedures

The present research was approved by the university’s institutional review board. The study used a randomized crossover design across behaviours and participants, with one replication of each of two experimental treatment phases (i.e., blocked and random). Participants were randomly assigned via a random number generator to begin with either blocked or random treatment and alternated thereafter. Prior to beginning treatment, all participants underwent informed consent with study personnel and baseline assessment of speech and language abilities. In addition to standardized language assessment, all individuals participated in naming a corpus of 575 black and white pictures on two separate occasions. Pictures included line drawings from Snodgrass and Vanderwart (1980), which were supplemented with additional stimuli consisting of drawings with varying degrees of shading and realism as well as photographic images. Items consistently named correctly and incorrectly (at both sessions) were identified, the latter of which served as candidate items for treatment and probing. Participants were given 12 s to name each image before advancing to the next item. Based on clinical judgement, the 12-second limit was chosen to capture most responses while reducing frustration that may occur during a longer response period.

Therapy

All therapy stimuli were presented by the researcher on a laptop computer (Dell Latitude 4630, 14-inch screen) via Eprime 2.0 Professional (Psychology Tools, www.pstnet.com). Stimuli included black and white images of nouns from the corpus of 575 pictures described above. Participants were asked to name the pictured object for each presentation. There were four phases of treatment: two blocked phases alternating with two random phases. Participants were randomly assigned to begin with a blocked or random treatment phase, and order of the first phase was counterbalanced across participants.

From the list of pictures participants were unable to name on two occasions during the assessment phase, 20 unique pictures were chosen as treatment stimuli for each of four phases of treatment (80 total words). Treatment and probe lists were balanced for word frequency, number of living versus nonliving items, and number of syllables.

For each of the four treatment phases, therapy occurred four days per week for two weeks, resulting in 32 therapy sessions across eight weeks. Dosing for each phase of treatment was constant (see Harnish et al., 2014). Dosage parameters were 160 teaching episodes per session (8 presentations of 20 pictures) and 1280 teaching episodes per phase (160 episodes x 8 sessions). The cumulative intervention intensity of all four phases was 5120 teaching episodes (1280 × 4 phases). The duration of therapy sessions was dependent upon the time required for the participant to name all eight presentations of each of the twenty items.

The blocked phases of CPNT were a modified version of the anomia platform treatment created by Kendall et al. (2014). The therapy was developed as a platform tool to be used with individuals with anomia resulting from a variety of possible impairments (e.g., semantic access, phonological encoding). The cueing in the therapy is not a hierarchy in the traditional sense, whereby cues are faded out to encourage more independence from the participant. Instead, cues at multiple levels of the naming process are consistently given throughout therapy to attempt to stimulate a correct production in individuals who potentially have different impairments at various levels in the naming process. It is this repeated correct production by the participants tied to the exposure of multiple cues or associations with the target that is thought to be the mechanism of action for behavioural and neuroplastic changes. Indeed, this treatment has produced improved naming abilities of trained items in previous studies (Harnish & Lundine, 2015; Harnish et al., 2014; Harnish, Rodriguez, et al., 2018a). During the blocked treatment phases, participants saw eight consecutive presentations of the same picture, some of which were supplemented by cues on the computer or by the therapist. The participant attempted to name the picture 1) independently, 2) with the addition of the written word below the picture, 3) after the therapist spoke the word, 4) after seeing a blank screen for three seconds, 5) after three scripted semantic cues spoken by the therapist, 6) after a letter and phonemic cue spoken by the therapist, 7) after the therapist spoke the word again, and 8) after seeing another blank screen for three seconds. The treatment was considered a blocked paradigm because all eight presentations of each of the 20 stimuli were presented in consecutive order (e.g., eight consecutive presentations of “book,” followed by eight consecutive presentations of “tree”); however, the order in which treatment items were selected was random. That is, the presentation software was set to randomly select the order of presentation of treatment items, but once an item was selected, all eight presentations occurred sequentially in a blocked fashion. All eight presentations of each of the 20 pictures occurred in the same order regardless of the correctness of the participant’s response. In the event the participant could not name the item with the cue, they were given the opportunity to repeat the word after the therapist’s model for each trial.

The random phases of treatment trained the same number of items with the same cues as the blocked phases. Instead of presenting all eight trials of each item consecutively with the same cueing pattern, pictures and cues were randomized (e.g., “tree” – semantic cue, “book” – phonemic cue, “hanger” – repetition, “book” – orthographic cue, “flower” – phonemic cue; see Figure 1). Thus, the total number of therapeutic trials was the same, but the order of presentation was randomized. Encouragement and feedback (i.e., knowledge of results and modeling) were provided for each phase of treatment.

Figure 1. Treatment protocol.

A pictorial display of the blocked and random treatment protocols. At the top, the blocked protocol shows eight consecutive pictures of a chair, each accompanied by the eight different naming opportunities during treatment, with or without cues, described in the methods. At the bottom, the random protocol shows eight different pictures of treatment items (i.e., chair, basket, leaf, and table in random order and sometimes repeated) accompanied by random naming opportunities with or without cues (i.e., repetition, semantic cue, orthographic cue, independently name, name after a 3-second delay).

Probes

All treatment items were probed (i.e., tested in a confrontation naming paradigm) during baseline, before the start of treatment each day, after all treatment phases were finished, and at three months following completion of treatment. Six picture naming probe lists were created from items that were named incorrectly during initial assessment. The first four lists each contained items trained in phases one through four, respectively. List five probed untrained items as an experimental control and was administered with the same frequency as trained lists. We incorporated this untrained list because we wanted to demonstrate that treatment effects were a result of the therapy as opposed to frequent probing. Since many anomia treatments, including the treatment in the present study, tend to yield item-specific treatment effects without much generalization to untrained items¹ (Harnish, Schwen Blackett, et al., 2018b; Harnish et al., 2014; Nickels, 2002), we did not expect generalization of treatment items to untrained lists. However, if list five showed improvement over time, it is possible that this would represent generalization of trained to untrained items. To differentiate between true generalization effects and improvements in naming untrained items due to frequent probing (Nickels, 2002), a sixth list of untrained items was created for infrequent probing. List six was administered on only three occasions: before treatment phase one, at the completion of treatment phase four, and at follow-up. See the supplemental materials for a visual representation of our probe schedule (Figure S1).

During the initial baseline phase, lists one through five were probed according to a pre-determined, randomly assigned order. Participants were randomly selected to receive between five and eight baseline probes. Using a different number of baseline probes across participants ensures that the number of baseline probes is not responsible for treatment effects (Kratochwill et al., 2013). During treatment phases, the list of items trained in that phase of treatment was probed daily. The same list was also probed in the five to eight sessions prior to that treatment phase (as a phase-level baseline) and in the two sessions immediately following that treatment phase (as a phase-level post-treatment measure). Then, probing for each list occurred only once weekly. This reduction in probing was used to decrease the likelihood of reinforcing potential errors that may occur with frequent probing (Howard et al., 2015). The overall schedule allowed for collection of baseline, treatment, and post-treatment probes for each phase and reduced probing of items not trained in a given phase.

Baseline probes occurred across two or three days depending on scheduling and participant fatigue. Prior to each baseline naming probe session, individuals participated in the Spatial Span assessment from the Wechsler Memory Scales (Wechsler, 1997) to divert attention from the naming process and distract participants from potential mental rehearsal of the words prior to the next administration of the naming probes (as some sessions were completed during the same day). Researchers controlled for any trend detected during baseline probes using Tau-U (Parker et al., 2011; see Statistical Analysis). Follow-up probes were collected three months after treatment concluded. All probe lists were administered three times during the same follow-up visit, with administration of the Spatial Span again used between administrations to decrease mental rehearsal. Picture naming probe items were delivered on a laptop computer screen and were scored as correct or incorrect by the therapist. Responses that differed by one phoneme or more were counted as incorrect unless the variation was dialectical. Minor phoneme distortions (e.g., likely due to the influence of dysarthria or AOS) were counted as correct so long as the target was still intelligible. Similarly, plural/singular variations (e.g., tacos for taco) were counted as correct so long as the change did not result in a different root word or impact intelligibility (e.g., checker for checkers).

A list of acceptable synonyms was created and discussed with the research team until consensus was reached. Ongoing consensus meetings occurred to discuss potentially acceptable responses that varied from the target word. Participants were allowed 10 s to name each picture on probe lists while the picture remained on the computer screen. The clinician indicated accuracy by pressing a key on the keyboard. Self-corrections were scored as correct. No cueing or feedback occurred during probing. Each probe item on the list was presented only once per session, and each list was presented in a random order, regardless of the phase during which the probe occurred. That is, there was no blocked or random presentation structure for the probes to mimic treatment phases because each item was presented only once. Moreover, the purpose of the probes was not to assess the efficacy of the treatment schedule in real time but rather to determine how well training improved short-term lexical retrieval for a given item from one session to the next.

Statistical analysis

Single-subject data were analyzed using Tau-U (Parker et al., 2011), a statistical analysis technique derived from Mann–Whitney U and Kendall rank correlation. Tau-U measures effect size as nonoverlap between two phases (i.e., a baseline and intervention phase) to complement visual analysis. Tau-U can account for treatment phase trend while also correcting for undesirable baseline trend by adding trend data to the nonoverlapping data (e.g., in the case of positive trend during treatment) or subtracting it (e.g., to control for positive trend during the baseline phase). These are two factors that make Tau-U ideal for single-case experimental design. Experimental control was determined by comparing baseline performance on untrained items with performance on these same untrained items probed at every session during the first treatment phase as well as the three sessions immediately post-treatment. We also reviewed performance on a seldom probed untrained list via visual inspection to gauge the presence of true generalization. Per Vannest and Ninci (2015), Tau-U effect sizes of < 0.20 were considered small; 0.20 to < 0.60, moderate; 0.60 to < 0.80, large; and ≥ 0.80, very large. These benchmarks were used to interpret all the analyses described above. Additional interpretation of the Tau-U statistic will be provided in the discussion.

Results

Participants and session length

Of twenty-eight individuals recruited for participation, 10 met inclusion criteria and completed the study. Demographic data are presented in Table 1. There were three female and seven male participants in the sample. Age ranged from 23 to 76 years. All participants had 12 years of education or more. BNT and WAB scores were variable, though six of 10 participants were classified as having Broca’s aphasia. AOS ratings were also variable. All participants except one (P5) were rated as having at least mild AOS. P1, P3, P6, P7, and P10 began with random treatment, whereas P2, P4, P5, P8, and P9 began with blocked. All participants completed all sessions without adverse events. The length of each session was recorded throughout the study. Across participants, time in treatment did not significantly differ between blocked (M = 24 min; range: 10 min to 56 min) and random (M = 26 min; range: 10 min to 65.5 min) phases of treatment (t(9) = −0.682, p = .512).

Table 1. Demographic and linguistic information for enrolled participants.

Participants	Gender	Age	Education (years)	Months post-CVA	BNT score	WAB aphasia quotient	WAB classification	AOS ratings
P1	M	70	20	154	35	69.3	Broca's	6.33
P2	F	75	12	42	4	32.1	Broca's	1.33
P3	M	34	17	10	26	64.1	Broca's	3.67
P4	F	23	14	22	38	81.1	Anomic	2
P5	M	76	18	164	38	75.8	Transcortical Motor	0
P6	M	59	21	25	11	49.9	Broca's	4
P7	M	62	17	138	39	92.6	Anomic	2.5
P8	M	65	14	6	11	52.4	Broca's	3
P9	F	50	16	40	15	56.2	Transcortical Motor	4
P10	M	70	13	52	18	37.5	Broca's	4.83

Note. AOS ratings are averaged across all raters and were rated on a scale of 0 (absent) – 7 (profound). M = male, F = female; CVA = cerebrovascular accident; BNT = Boston Naming Test (Kaplan et al., 1983); WAB = Western Aphasia Battery (Kertesz, 1982); AOS = Apraxia of Speech.

Reliability and fidelity

All probe and treatment sessions were video- and audio-recorded to evaluate reliability and treatment fidelity. Intra- and inter-rater reliability were separately calculated on accuracy via percent agreement on 20% of randomly selected probe sessions by either the therapist who delivered the treatment or a therapist who did not deliver the treatment, respectively. Both intra- (99.3%) and inter-rater (98.5%) reliability were found to be high.

Undergraduate student raters otherwise unaffiliated with the study evaluated treatment fidelity on 10% of randomly selected treatment sessions. Raters used a checklist for each session that asked them to indicate if 1) the picture was shown, 2) the participant was given the correct cue by the therapist, and 3) the participant was asked to attempt repeating the word if they were unable to produce it correctly after the initial cue. Each of these three criteria was worth one point for every trial. Each session was scored for number of points earned out of number of total possible points. Treatment fidelity was calculated at 97%.

Individual treatment effects

To investigate whether blocked or random presentation of treatment stimuli enhanced naming acquisition and/or maintenance, individual effect sizes were calculated for each treatment phase for each participant using Tau-U (Parker et al., 2011) via a web-based application (the Tau-U Calculator, singlecaseresearch.org; Vannest et al., 2016). To assess acquisition, baseline performance was compared with performance during and immediately following treatment, including the post-treatment sessions (as in Lee & Cherney, 2018). All participants responded to the aphasia treatment in both phases of both conditions with large to very large effect sizes except P1. P1 demonstrated a large effect of the first blocked and first random treatment phase but no significant effect of the second blocked or random phase (Table 2). Tau-U averages, weighted for the number of baseline probes, were calculated for the two blocked treatment scores and the two random treatment scores for each participant. Five participants (P1, P3, P4, P6, and P9) demonstrated a better overall effect of the random treatment schedule via Tau-U averages in a larger effect size category (e.g., very large vs. large), whereas one participant (P2) improved more during the blocked treatment. Four participants showed a roughly equivalent response to both schedules (P5, P7, P8, and P10). Of the participants that showed greater benefit from the random treatment, P1, P3 and P4 were the most notable, demonstrating the greatest difference in response between the two conditions.

Table 2. Participants’ Tau-U effect sizes comparing baseline and treatment + post-treatment performance (acquisition).

Participants	Blocked Phase 1	Blocked Phase 2	Blocked Weighted Avg.	Random Phase 1	Random Phase 2	Random Weighted Avg.
P1	0.75* (L)	−0.02	0.36	0.77* (L)	0.56	0.67† (L)
P2	0.96† (V)	0.82* (V)	0.89‡ (V)	0.80* (V)	0.73* (L)	0.76‡ (L)
P3	0.71* (L)	0.75* (L)	0.73† (L)	0.91† (V)	0.78* (L)	0.85‡ (V)
P4	0.73† (L)	0.85† (V)	0.79‡ (L)	0.85† (V)	0.95† (V)	0.90‡ (V)
P5	0.89† (V)	0.91† (V)	0.90‡ (V)	0.89† (V)	0.80* (V)	0.85‡ (V)
P6	0.87† (V)	0.64* (L)	0.75‡ (L)	0.98‡ (V)	0.67* (L)	0.84‡ (V)
P7	0.87† (V)	0.96† (V)	0.92‡ (V)	1.00‡ (V)	0.98† (V)	0.99‡ (V)
P8	0.96‡ (V)	0.96† (V)	0.96‡ (V)	1.00† (V)	0.98† (V)	0.99‡ (V)
P9	0.84† (V)	0.69* (L)	0.77‡ (L)	0.73* (L)	0.95† (V)	0.84‡ (V)
P10	0.85† (V)	1.00† (V)	0.93‡ (V)	0.83† (V)	0.93† (V)	0.88‡ (V)

Note. S = small effect (< 0.20); M = moderate effect (0.20 – < 0.60); L = large effect (0.60 – < 0.80); V = very large effect (≥ 0.80). Tau-U values are rounded to the nearest hundredth decimal place. Effect strength determined by standards described by Vannest and Ninci (2015).

*p < .05. †p < .01. ‡p < .001.

To examine maintenance effects, baseline and follow-up performance (3 months after completing treatment) of trained items were compared (Table 3). Two participants (P8 and P9) showed very large significant effects of both blocked phases of treatment, and three additional participants (P2, P3, and P10) showed a very large effect in one of the two blocked phases. Meanwhile, three participants (P7, P8, and P10) demonstrated very large significant effects of both random phases of treatment, and an additional six participants (P1, P2, P3, P4, P5, and P9) demonstrated very large effects in at least one of the two random phases. Tau-U averages revealed a similar pattern to the individual effect sizes across treatment phases. Five participants (P2, P5, P8, P9, and P10) showed large or very large significant effects of the blocked schedule overall. Eight participants (P1, P2, P3, P4, P7, P8, P9, and P10) showed large or very large significant effects of the random schedule overall. Of the participants with significant treatment effects based on Tau-U averages, four demonstrated better maintenance resulting from the random schedule (P1, P3, P4, and P7), as represented by an average Tau-U value in a larger effect size category. Two participants demonstrated better maintenance resulting from the blocked schedule (P5 and P9). Three participants showed a roughly equivalent response to both schedules (P2, P8, and P10). One participant (P6) demonstrated no significant maintenance effects across any of the four individual treatment phases or either Tau-U mean value.

Table 3. Participants’ Tau-U effect sizes comparing baseline and follow-up performance (maintenance).

Participants	Blocked Phase 1	Blocked Phase 2	Blocked Weighted Avg.	Random Phase 1	Random Phase 2	Random Weighted Avg.
P1	0.60	−0.87	−0.13	0.39	1.00* (V)	0.69* (L)
P2	1.00* (V)	0.33	0.67* (L)	0.93* (V)	0.47	0.70* (L)
P3	0.13	1.00* (V)	0.57	1.00* (V)	0.73	0.87† (V)
P4	0.25	0.53	0.39	0.53	1.00* (V)	0.77* (L)
P5	0.87	0.47	0.67* (L)	1.00* (V)	−0.73	0.13
P6	0.33	0.13	0.23	0.50	0.00	0.26
P7	0.60	0.33	0.47	1.00* (V)	1.00* (V)	1.00† (V)
P8	1.00* (V)	1.27† (V)	1.13‡ (V)	1.00* (V)	1.00* (V)	1.00† (V)
P9	1.00* (V)	1.00* (V)	1.00† (V)	0.53	1.00* (V)	0.77* (L)
P10	0.87	1.00* (V)	0.93† (V)	0.94* (V)	1.00* (V)	0.97† (V)

Note. S = small effect (< 0.20); M = moderate effect (0.20 – < 0.60); L = large effect (0.60 – < 0.80); V = very large effect (≥ 0.80). Tau-U values are rounded to the nearest hundredth decimal place. Effect strength determined by standards described by Vannest and Ninci (2015).

*p < .05. †p < .01. ‡p < .001.

Finally, across participants, there was little evidence of generalization or practice effects on the untrained word lists. Three of the 10 participants (P4, P8, and P9) demonstrated significant moderate to large Tau-U effect sizes on the untrained, frequently probed word list (list 5) when comparing baseline performance with performance during and immediately following treatment (Table 4); however, it is unclear whether these effects are due to generalization or practice effects given the frequency of probing. The remaining participants showed no significant effects. Raw scores on the seldom probed word list (list 6) revealed four participants (P1, P3, P6, and P8) with small improvements from pre- to post-treatment probing and five participants (P1, P3, P4, P6, and P8) with either additional improvement or maintenance of some acquired items at follow-up probing (Table 5). A plot of P1’s performance throughout the study is demonstrated in Figure 2. Additional individual participant plots can be found in the supplementary materials (Figures S2-S10).

Figure 2. Cued picture naming treatment probe scores for blocked and random phases for P1.

Four plots of the percent of probes correctly named for each of 44 sessions for participant 1. Each plot shows the percent of probes correctly named for the untrained and seldom probed sets compared to either the first or second random or blocked set of words. Baseline, treatment, post-treatment, and follow-up phases are indicated. Participant 1 shows notably better performance on the trained probes during the first and second phase of random treatment as well as the first phase of blocked treatment. Maintenance of trained items appears better following the random phases.

Table 4. Participants’ Tau-U effect sizes comparing baseline and treatment + post-treatment performance on untrained items (control).

Participants	Untrained
P1	−0.21
P2	−0.07
P3	0.40
P4	0.55* (M)
P5	0.25
P6	0.51
P7	0.05
P8	0.75† (L)
P9	0.79† (L)
P10	0.58
Weighted Avg.	0.37‡ (M)

Note. S = small effect (< 0.20); M = moderate effect (0.20 – < 0.60); L = large effect (0.60 – < 0.80); V = very large effect (≥ 0.80). Tau-U values are rounded to the nearest hundredth decimal place. Effect strength determined by standards described by Vannest and Ninci (2015).

*p < .05. †p < .01. ‡p < .001.

Table 5. Participants’ raw scores on seldom-probed untrained items (generalization).

Participants	Pre-treatment correct (n = 20)	Post-treatment correct (n = 20)	Follow-up correct (n = 20)
P1	1	5	2
P2	3	0	1
P3	2	4	4
P4	5	3	6
P5	10	7	9
P6	0	1	1
P7	10	6	10
P8	1	4	3
P9	1	1	1
P10	5	5	1

Group treatment effects

As stated above, acquisition effects were calculated via Tau-U by contrasting baseline performance with performance during and immediately following treatment. Comparing the average of all participants’ acquisition effects, weighted by the number of baseline sessions, revealed very large and approximately equal Tau-U effect sizes for random (0.86, p < .01) and blocked (0.80, p < .01) schedules. An advantage of the random schedule was seen over the blocked schedule when examining the effect of maintenance from baseline to follow-up, which revealed a large effect of random practice (0.72, p < .01) and a moderate effect of blocked practice (0.59, p < .01).

Discussion

Treatment results

The purpose of this study was to conduct a preliminary analysis exploring differences in acquisition and maintenance of picture naming abilities between blocked and random presentations of CPNT for persons with post-stroke aphasia. Overall, results showed that both blocked and random phases of treatment produced significant treatment gains. According to the Tau-U results, 95% of treatment phases showed a large or very large significant increase in picture naming probe performance when compared to baseline performance, indicating a treatment effect. Nine of 10 participants demonstrated significant treatment effects for acquisition during all phases of treatment. The remaining participant did show significant gains in the first phase of each condition. These findings are in agreement with Harnish et al. (2014), demonstrating that CPNT improves retrieval of trained items in individuals with anomia.

Comparison of blocked and random practice schedules

In comparing the two schedules, the random treatment schedule produced slightly better outcomes than the blocked schedule in terms of maintenance, whereas the schedules were roughly equivalent in terms of acquisition effects. The advantage of the random schedule for maintenance of trained items is consistent with the motor learning literature. It is also consistent with evidence from the verbal learning and memory literature and recent studies investigating distribution of practice in aphasia treatment, where our random and blocked schedules align with spaced and massed practice, respectively (Middleton et al., 2016; Middleton et al., 2020). Principles of motor learning suggest that random practice facilitates maintenance due to the phenomenon known as contextual interference (Magill & Hall, 1990), which may be attributed to the forgetting-reconstruction hypothesis (Lee & Magill, 1985) and stems from intratask interference in verbal learning (Battig, 1966; 1972). Additionally, longer ISIs in verbal memory tasks (like those in a random schedule) have been shown to facilitate longer periods of retention (Cepeda et al., 2006), potentially attributable to consolidation theory (Wickelgren, 1972). However, the lack of difference in impact on acquisition for the blocked and random schedules is inconsistent with the forgetting-reconstruction hypothesis (Lee & Magill, 1985), which would predict an advantage of blocked practice on acquisition due to the use of a temporary memory buffer (i.e., phonological buffer; Baddeley & Hitch, 1974; 2019; Gathercole & Baddeley, 1993) during training.

To understand why the blocked treatment schedule did not result in greater acquisition effects, we consider multiple possibilities. First, we consider errorless learning. Repeated studies have demonstrated that errorless and errorful learning approaches provide similar benefits for individuals with aphasia receiving anomia treatment (Fillingham et al., 2003; 2005; 2006; McKissock & Ward, 2007). The blocked practice schedule utilized in the present study may offer a more errorless style with fewer demands on retrieval due to the activity of a phonological buffer (Baddeley & Hitch, 1974; 2019; Gathercole & Baddeley, 1993), whereas a random schedule may require retrieval attempts on every trial (Battig, 1966; 1972), akin to an errorful approach. Considering the errorless or errorful nature of the practice schedules may explain similarities on acquisition; however, it is worth noting that previous studies found similar results of errorless and errorful learning on maintenance as well (Fillingham et al., 2006; McKissock & Ward, 2007), which is inconsistent with our finding an advantage of the random condition on maintenance.

A second possibility is that impairment of the phonological buffer may have precluded some participants from benefiting from the blocked schedule during acquisition. If the phonological buffer were impaired, activation of a target word would decay more quickly, creating similar demands on the word retrieval process to the random schedule. Therefore, acquisition could look similar between the random and blocked conditions for any participants with an impairment of the phonological buffer. However, the random schedule could still provide an advantage over the blocked schedule on maintenance due to the longer ISIs and greater consolidation into long-term memory.

Six participants did not show an individual advantage of the random practice schedule on maintenance: P2, P5, P6, P8, P9, and P10. Three of these (P2, P8, and P10) demonstrated roughly equivalent maintenance effects with both schedules. P5 demonstrated a significant very large maintenance effect for the first random phase of treatment but a loss after the second random phase (Table 3). The reason for this response is unclear, but perhaps due to a loss of novelty or an effect of fatigue. P6 on the other hand, showed no significant benefit across treatment phases and conditions, indicating poor maintenance in response to CPNT, regardless of condition. P9 demonstrated a significant very large maintenance effect for the second random phase of treatment, but no significant maintenance effect for the first random phase of treatment (Table 3). This response, as with P5, is unclear, but may demonstrate that P9 required more time to acclimate to the random schedule.

Although the group-level advantage of the random treatment on maintenance is relatively small, this may in part be attributed to the fact that our measure of maintenance (i.e., contrasting baseline and follow-up performance) does not control for effects of acquisition. Of course, the number of acquired items dictates the total possible number of items that can be maintained. Therefore, if random practice has a limited effect on acquisition for a given participant, this would also truncate their overall maintenance effect for the random schedule. Future studies may benefit from utilizing a criterion-based treatment (i.e., requiring a specified number of correct productions before moving on) to obtain better measures of maintenance, since this approach could lead to a greater number of opportunities to maintain acquired words. Of note, Knock et al. (2000) showed larger motor speech treatment effects for a random schedule than blocked using a criterion-based treatment.

The length of ISIs may further explain the small advantage of the random practice schedule on maintenance. In their meta-analysis, Cepeda et al. (2006) proposed that longer ISIs favour longer retention periods. However, the authors note that the ISI should correspond to the desired length of retention as much as possible (e.g., if the goal is to retain the target for a day, the ISI should be approximately one day). In the current study, each treatment session lasted between 24 and 25 min on average across participants, indicating that even the longest ISIs for the random condition lasted only minutes. Meanwhile, follow-up was assessed 3 months after the final treatment phase. Therefore, it is possible that the minutes-long ISI did not yield as strong of a maintenance effect on the months-long retention period as a longer, perhaps days-, weeks-, or months-long ISI may have yielded. Future studies should investigate whether increasing the length of ISIs (to match the length of the retention period more closely) could produce greater benefits of random practice on maintenance.

Impact of the results

There are both theoretical and practical implications for determining whether blocked or random practice differentially affects language outcomes in terms of acquisition and maintenance. Theoretically, our results partially support the literature from verbal memory and motor rehabilitation that random practice facilitates maintenance perhaps due to the effects of contextual interference. However, blocked practice did not produce an advantage in terms of acquisition as hypothesized, leaving room for further investigation and theoretical explanation.

Practically, our results may encourage SLPs to trial a random treatment schedule when targeting word retrieval with patients with aphasia. Nevertheless, it is worth reiterating that not all individual participant responses were consistent with our group-level results. Some participants benefited more from one practice schedule than another in terms of acquisition and some participants demonstrated better maintenance in response to the blocked schedule (P5 and P9). As a result, individual performance and personal preference remain important factors in choosing treatment schedules for patients with aphasia. For example, patients may appreciate the increased challenge of the random task structure or may prefer to avoid excess frustration via a more repetitive blocked design. Future studies should gather information about patient preference for blocked and random treatment schedules, as well as variables that may influence their preferences and their success with each.

Limitations and future directions

The present study has some limitations that should be addressed in future research. First, order effects may have contributed to the results. Despite attempts to control for order effects by conducting both treatment conditions twice for each participant and counterbalancing the initial treatment among participants, four out of 10 participants demonstrated greater improvement during the treatment they were administered first, evidenced by a mean Tau-U value in a larger effect size category. Five participants responded roughly equally to both treatment schedules and one participant demonstrated greater improvement during the treatment administered second. Although order effects alone cannot fully explain performance across all participants, future research should make greater attempts to control for them.

A possible practice effect is an additional limitation of the present study. A moderate overall effect was observed on the untrained frequently probed lists (list 5) across participants (Table 4), which suggests that a practice effect may have influenced our results. However, this moderate effect was potentially driven by a small number of participants with true generalization effects. P4 and P8, two of the three participants who demonstrated moderate to large improvement on the untrained, frequently probed list (list 5) also showed improvements on the untrained, seldom probed list (list 6; Table 5). P8 improved on the seldom probed list from pre- to post-treatment and maintained one of three acquired words at follow-up. P4 declined in performance from pre- to post-treatment, but accurately named more words on the seldom probed list at follow-up than at pre-treatment. Future studies should continue to control for practice effects and distinguish practice effects from generalization.

Another limitation of the present study was the timing of the follow-up measures used to determine maintenance effects. Follow-up probing was completed three months after the final treatment phase. This meant that items from earlier treatment phases had to be maintained considerably longer than items from later treatment phases (e.g., items from the first treatment phase would need to be maintained six weeks longer than items from the final treatment phase). Future studies should assess maintenance in a staggered fashion commensurate with the treatment phases of the experiment.

Many studies of aphasia or similar disorders have difficulty controlling for the unique features associated with participants’ specific neurologic injuries, including comorbidities such as AOS, as well as premorbid and current cognitive-linguistic levels of functioning. This study was no exception, and therefore, was limited in its ability to distinguish whether treatment effects were indeed caused directly by the treatment condition or may have been informed by individual variability. For example, P1 demonstrated the lowest average response to both treatment schedules due to the absence of a treatment effect in the second phase of each condition. P1 also had the highest rating of AOS, leading to the possibility that AOS severity contributed to this participant’s treatment outcomes. However, despite differences in AOS severity among the other participants, they all demonstrated large to very large effects on acquisition with both schedules and variability in their maintenance does not appear to correspond to AOS severity. We attempted to control for participant variability using stringent inclusion criteria and a case-series design with replication of experimental phases, but the heterogeneity of aphasia remains an inherent challenge. We were also limited in drawing group-level conclusions based on the small sample size. Larger group studies may show trends that were not identified through the present single-subject design.

Future studies could also look more closely at related cognitive-linguistic skills. For example, a study by Schweighofer et al. (2011) examined the role of visuospatial working memory in motor learning in blocked and random treatment phases. For individuals with intact visuospatial working memory, random practice led to good retention, but blocked practice led to poorer retention. For individuals with impaired visuospatial working memory, random and blocked practice both led to good retention. The authors suggested that for those with poor working memory, slower processing and increased errors during learning led to better maintenance. Since there is literature to suggest that working memory is frequently affected in persons with aphasia (Burgio & Basso, 1997; Lang & Quitz, 2012; Mayer & Murray, 2012; Potagas et al., 2011; Seniow et al., 2009; Wright & Fergadiotis, 2012) and informs treatment response (Harnish & Lundine, 2015), this could be an interesting line of future research. To explore the potential impact of working memory in the present study, we performed a post-hoc analysis of the relationship between participants’ scores on the backward spatial span (Wechsler, 1997) and the weighted average of acquisition effect sizes for the blocked and random treatment conditions (Table S1 and S2). The results of our linear regressions were not statistically significant (Blocked: F(1, 8) = 1.237, p = .298; Random: F(1, 8) = 2.871, p = .129), but future studies with larger samples may have more power to observe such an effect, if one exists.

Finally, in order to maintain experimental control to investigate the effect of differing practice schedules, we did not capitalize on specific principles known to facilitate generalization. Previous research has demonstrated that there are principles that guide generalization to untrained items in the same semantic category, such as training less prototypical items to generalize to more prototypical items (Kiran et al., 2011) and training abstract words to promote generalization to concrete words (Sandberg & Kiran, 2014). Moreover, there is a body of literature demonstrating increased latency of picture naming when items are grouped according to semantic category than when they are randomly presented (Belke et al., 2005; Martin et al., 2004; Navarrete et al., 2014), suggesting that there may be an increased processing load associated with inhibition of competing representations. Future research should investigate the effects of semantic grouping in blocked and random practice during aphasia therapy.

Conclusions

The present study investigated the effects of blocked versus random practice in treatment for word retrieval deficits in individuals with aphasia. We hypothesized that blocked practice would foster acquisition because of the repeated consecutive exposure to the word after the initial attempt at naming it independently. We suspected that this would increase the likelihood of consecutive correct responses because the word may be held in a phonological buffer without the demands of recalling it during each presentation. Additionally, a blocked schedule would avoid the distracting nature of alternating between targets at random. However, we hypothesized that random practice would facilitate maintenance because of the necessity to repeatedly practice retrieving the lemma and phonological code for each presentation. In partial support of our hypotheses, we found that both blocked and random schedules facilitated large acquisition effects across participants and that the random practice schedule produced a slightly greater effect of maintenance.

Nearly all participants in the present study responded with large to very large treatment effects in both conditions, regardless of delivery in a blocked or random fashion. Although at the group level random practice facilitated slightly greater maintenance, there were subtle individual differences in treatment response across participants, which may be accounted for by preference or comorbid impairments. Therefore, the present data suggest that for word retrieval treatment with multiple repetitions of the same items, a random presentation may be of benefit, particularly to maintenance of treatment gains; however, the decision to deliver treatment items in a random fashion should not be applied universally, considering some participants performed best with a blocked schedule. Instead, deciding on a treatment schedule may be best informed by the patient’s preference and performance, perhaps following a trial period of both blocked and random schedules. Future research should investigate the extent to which cognitive abilities may modulate response to the different practice schedules, to determine for whom blocked or random schedules are best.

Acknowledgements

We would like to acknowledge Jennifer Brello, Nadine Whiteman, and Angela Dubis-Bohn for their assistance with data collection. Authors of this article were supported in part by the National Institute on Deafness and Other Communication Disorders of the National Institutes of Health under award number R01DC017711 (SMH). The first author (VAD) was also supported in part by the National Center for Advancing Translational Sciences of the National Institutes of Health under Grant Numbers TL1TR002735 & UL1TR001450. The content is solely the responsibility of the authors and does not necessarily represent the official views of the National Institutes of Health.

Disclosure statement

No potential conflict of interest was reported by the author(s).

Notes

1 Some semantic treatments show generalization from trained to untrained items when they share semantic features (Boyle, 2004; Boyle & Coelho, 1995, Coelho et al., 2010, Kiran & Thompson, 2003). Also, there is some evidence that typicality plays a role in generalization whereby training atypical prototypes generalizes to more typical untrained prototypes (Kiran, 2007).