When speakers produce words, the process of selecting and retrieving the target word is sometimes helped by prior naming events, and sometimes hindered. In some contexts, adults experience interference from previously named words that are semantically related to the target word (e.g., Belke, Meyer, & Damian, Reference Belke, Meyer and Damian2005; Howard, Nickels, Coltheart, & Cole-Virtue, Reference Howard, Nickels, Coltheart and Cole-Virtue2006). A handful of studies have also described lexical interference effects in young children (Booth & Vitkovitch, Reference Booth and Vitkovitch2008; Gershkoff-Stowe, Reference Gershkoff-Stowe2002; Stemberger, Reference Stemberger1989). This literature is much smaller and less systematic than that focusing on adults, but the available studies raise interesting questions about how lexical interference manifests in childhood and how the dynamic activities that support language use may or may not change with development. This study fits within that literature. It examines the influence of semantic context and previous trials on interference in 3-year-old children during a naming task.

In adults, interference from recently named items has been studied within blocked cyclic naming and continuous naming paradigms. The general pattern of findings is that naming pictures with shared meaning (e.g., cat, dog) produces interference, measured in response latencies and/or naming errors. In blocked cyclic naming, participants name small sets of items, repeating a set several times within a block. Blocks are composed of semantically homogeneous sets (e.g., pig, cow, horse, sheep), or mixed sets. Interference manifests as slower responding and/or more errors within the homogeneous blocks relative to the mixed blocks (Abdel Rahman & Melinger, Reference Abdel Rahman and Melinger2007; Belke et al., Reference Belke, Meyer and Damian2005; Damian & Als, Reference Damian and Als2005; Schnur, Schwartz, Brecher, & Hodgson, Reference Schnur, Schwartz, Brecher and Hodgson2006). In continuous naming (Brown, Reference Brown1981; Howard et al., Reference Howard, Nickels, Coltheart and Cole-Virtue2006), participants name a series of pictures, and items from different semantic sets (e.g., farm animals, fruits) are dispersed throughout the list. Interference manifests as a slowing of successive responses within a semantic set. Interference has also been observed in repetitions or perseverations of previous responses (e.g., pig, cow, horse, cow) in speakers with aphasia and in healthy adult speakers under speeded naming conditions (Hsiao, Schwartz, Schnur, & Dell, Reference Hsiao, Schwartz, Schnur and Dell2009; Vitkovitch, Kirby, & Tyrrell, Reference Vitkovitch, Kirby and Tyrrell1996; Vitkovitch, Rutter, & Read, Reference Vitkovitch, Rutter and Read2001). In healthy adult speakers, these errors do not occur as immediate repetitions, but surface with several trials intervening (Vitkovitch et al., Reference Vitkovitch, Kirby and Tyrrell1996, Reference Vitkovitch, Rutter and Read2001).

Interference effects provide insight into the nature and organization of lexical processing, and into the ways in which “experience changes language processing” (Oppenheim, Dell, & Schwartz, Reference Oppenheim, Dell and Schwartz2010, p. 228). Of interest for the current study is the mechanism responsible for producing such effects. One possibility is that semantic interference may be caused by residual or ongoing activation of past targets (e.g., Belke et al., Reference Belke, Meyer and Damian2005; Vitkovitch et al., Reference Vitkovitch, Kirby and Tyrrell1996). In recent years, however, researchers have interpreted interference from past naming responses as effects that are instantiated in relatively long-term changes to the strength of connections along lexical pathways, rather than short-term residual activation (Damian & Als, Reference Damian and Als2005; Howard et al., Reference Howard, Nickels, Coltheart and Cole-Virtue2006; Hsiao et al., Reference Hsiao, Schwartz, Schnur and Dell2009; Oppenheim, Dell, & Schwartz, Reference Oppenheim, Dell and Schwartz2007; Oppenheim et al., Reference Oppenheim, Dell and Schwartz2010). This shift is driven by several important observations about the conditions that do and do not constrain interference. As noted above, interference is semantically based, occurring among words that share overlapping patterns of activation via shared features or categories. It is important to note that interference effects do not increase with shorter response–stimulus intervals (RSI). Given that residual activation is thought to last for several seconds at most (Bock & Griffin, Reference Bock and Griffin2000; Dell, 1986), the observed insensitivity to RSI makes explanations based in short-term lingering activation unlikely (Hsiao et al., Reference Hsiao, Schwartz, Schnur and Dell2009; Oppenheim et al., Reference Oppenheim, Dell and Schwartz2010; Schnur et al., Reference Schnur, Schwartz, Brecher and Hodgson2006; Vitkovitch et al., Reference Vitkovitch, Kirby and Tyrrell1996). Interference is, however, sensitive to the number of prior related responses: in blocked cyclic naming studies, the effects typically do not appear until the second block of naming (Abdel-Rahman & Melinger, Reference Abdel Rahman and Melinger2007; Belke et al., Reference Belke, Meyer and Damian2005; Schnur et al., Reference Schnur, Schwartz, Brecher and Hodgson2006). In continuous naming, the effects are cumulative: response latencies become incrementally longer with each exposure to items from a given semantic set (Howard et al., Reference Howard, Nickels, Coltheart and Cole-Virtue2006). In addition, interference effects are robust to intervening items (Damian & Als, Reference Damian and Als2005; Howard et al., Reference Howard, Nickels, Coltheart and Cole-Virtue2006). Thus, naming horse is slower after naming cow, even if several unrelated items intervene (although note that interference does dissipate when many items intervene; Schnur, Reference Schnur2014). In light of these findings, researchers (Damian & Als, Reference Damian and Als2005; Howard et al., Reference Howard, Nickels, Coltheart and Cole-Virtue2006; Hsiao et al., Reference Hsiao, Schwartz, Schnur and Dell2009; Oppenheim et al., Reference Oppenheim, Dell and Schwartz2010) have interpreted and modeled interference as arising from experience-driven incremental learning or strengthening of connections along lexical pathways. In some contexts, experience-driven change in the connection weights to words sharing features or connections can interfere with activation and selection of the target word, leading to slower or less successful naming. For example, in the model implemented by Howard et al. (Reference Howard, Nickels, Coltheart and Cole-Virtue2006), interference develops over time because the activation of target words strengthens the connections along the pathways to those words. When speakers attempt to name related items (i.e., other words within the semantic set), semantically related words also receive activation. Those words that have strengthened connections also become stronger competitors, delaying selection of the target word.Footnote ¹ An interesting implication of this perspective on interference is that although it is a disruption to time or accuracy that is measured, this disruption is indexing learning and change within the lexical system.

A handful of studies have also reported interference effects in very young children's speech, primarily through the lens of perseveration errors. There is some, albeit limited, evidence for changes with development in susceptibility to interference during lexical access, or in the temporal dynamics that govern interference effects. Stemberger (Reference Stemberger1989), for example, reported a diary study of spontaneous speech errors produced by his two children, and observed that the children produced a higher rate of between-clause perseverative errors than that observed for adults. Vousden and Maylor (Reference Vousden and Maylor2006) reported that 8- and 11-year-old children produced a higher rate of perseveratory errors than adults in a tongue twister task. Focusing on very young speakers, Gershkoff-Stowe (Reference Gershkoff-Stowe2002) examined toddlers’ picture-naming errors during book sharing, and reported that perseverative errors occurred more frequently in the responses of participants who were approximately 16 to 23 months old (peaking at the onset of their vocabulary explosion) than 28-month-old infants. For both age groups, the vast majority of these errors were repetitions of the immediately preceding response, a notable difference from the pattern produced by healthy adults (Vitkovitch et al., Reference Vitkovitch, Kirby and Tyrrell1996, Reference Vitkovitch, Rutter and Read2001). The absence of immediate perseverative errors in adults has been interpreted as reflecting an immediate but temporary inhibition of activation of the just-named target (Vitkovitch et al., Reference Vitkovitch, Kirby and Tyrrell1996, Reference Vitkovitch, Rutter and Read2001). Booth and Vitkovitch (Reference Booth and Vitkovitch2008) noted the possibility that the different patterns observed for children and adults might indicate a lack of immediate suppression in very young children, and looked for an age-related shift from immediate to delayed repetition when 4- and 6-year-old children named a set of 30 pictures. However, both groups showed above-chance tendencies to repeat the immediately preceding response. For the older group, there was also an above-chance tendency to repeat the response that had been produced two trials prior. The results hinted at, but did not demonstrate, a difference in the temporal dynamics of interference by age from preschool to early school age.

Thus, the evidence with respect to the manifestation of interference from past responses and the potential for developmental change is limited and scattered across age ranges. Moreover, it is not clear how interference phenomena in children should be interpreted. As in earlier studies with adults, lexical interference in children has been attributed to residual activation mechanisms (Booth & Vitkovitch, Reference Booth and Vitkovitch2008; Gershkoff-Stowe, Reference Gershkoff-Stowe2002; Stemberger, Reference Stemberger1989). By extension, the possibility that there may be age-related differences in the amount, control, or time course of residual activation has been considered (Booth & Vitkovitch, Reference Booth and Vitkovitch2008; Stemberger, Reference Stemberger1989). Researchers have also doubted the likelihood that changes occur with development in the efficiency or timing of “shutting down” of residual activation, arguing instead that any differences in susceptibility are more likely due to age-related changes in the robustness of lexical activation that arise as children develop richer and stronger connections within their lexical systems (Booth & Vitkovitch, Reference Booth and Vitkovitch2008; Budd, Hanley, & Griffiths, Reference Budd, Hanley and Griffiths2011; Dell, Burger, & Svec, Reference Dell, Burger and Svec1997; Gershkoff-Stowe, Reference Gershkoff-Stowe2002). When lexical representations are more fragile, activating the target word will be more difficult and more vulnerable to interference. However, to date, researchers have not explored the possibility that interference effects might arise as the result of learning and change rather than short-term residual activation.

If production-driven incremental learning occurs in adults and manifests as semantic interference, then such a mechanism may also account for the interference effects that occur in early childhood. However, the systematic comparisons that have informed understanding of the mechanisms of interference in adulthood have not been pursued with children, and it is not known whether interference effects are subject to the same influences as in adults. Several important pieces of information are needed to develop a clearer picture of how interference from prior naming events manifests in childhood, namely, whether such interference is a semantic phenomenon, whether interference effects are cumulative, and whether interference effects are insensitive to RSI and intervening material. The current study addresses the first two of these needs.

With respect to semantic effects, recent work (Snyder & Munakata, Reference Snyder and Munakata2013) has demonstrated that 6-year-old children show slower responding to semantically homogenous sets in blocked cyclic naming, supporting the view of a semantic effect. More information is needed about early development, however. In the study reported by Gershkoff-Stowe (Reference Gershkoff-Stowe2002), the 28-month-old children produced perseverative errors for semantically related targets as well as unrelated targets (e.g., car/ball, p. 674). These children named pictures in two books, one with a mixed set of items and one containing animal pictures (Polar Bear, Polar Bear by Bill Martin Jr.). Although the error rate was higher with the animal book, perseverations did occur for both stimulus sets. The stimulus books differed on two counts: the animal book presented pictures from a homogeneous semantic set. It also presented words that were much less likely to be familiar to the 2.5-year-old children in the study (e.g., flamingo, walrus, leopard). Thus, it is possible that lexical familiarity rather than semantic context was responsible for the higher rate of interference in the animal book. Finally, in the research reported by Booth and Vitkovitch (Reference Booth and Vitkovitch2008), 4-year-old children named 30 animals and produced perseverative naming errors, as well as semantically related, nonperseverative errors. There was no semantically heterogeneous set, and so it is not known whether the children in this age range would have also shown interference effects that were not semantically constrained.

With respect to questions about the cumulative nature of interference effects, Snyder and Munakata (Reference Snyder and Munakata2013) reported that, unlike adults, interference effects in blocked cyclic naming emerged from the first experimental block, rather than only emerging in later blocks. Such a result could point to different mechanisms or faster buildup of interference; however, the authors also noted that the difference in outcome may have been a methodological artifact. Other studies have not examined interference from the perspective of growth of interference across responses.

The purpose of the current study was to examine whether some of the influences on interference that hold for adults also hold for young children. The goal was to gain insight into the possibility that an incremental learning mechanism may also be an appropriate account of interference in early childhood. Three-year-old children named a series of pictures presented one at a time on a computer monitor. There were two sets of pictures, a semantically homogeneous set (animals, consistent with prior child-focused work; Booth & Vitkovitch, Reference Booth and Vitkovitch2008; Gershkoff-Stowe, Reference Gershkoff-Stowe2002) and a mixed set. The study resembled previous work with children (Booth & Vitkovitch) and adults (Vitkovitch et al., Reference Vitkovitch, Kirby and Tyrrell1996) in that it presented a continuous list of animals for naming. The two stimulus sets were well equated for likely difficulty of naming. Measures of response accuracy, response latency (response time [RT]), and perseverative errors were taken. RSI was not manipulated, but was controlled to assure relatively constant RSI across conditions and participants. Finally, this study was not designed to examine the robustness of interference to intervening material, but to provide a foundational picture for future investigations of interference phenomena. The research questions were as follows:

1. 1. Do 3-year-old children show semantically constrained interference effects?

   1. a. Given two sets of stimuli that are well balanced for likely familiarity and difficulty of the items, are they slower and/or less successful in naming items from a semantically homogenous set? I expected that the children would show more difficulty with the semantically homogeneous picture set, measured via response accuracy and RT.

   2. b. Are perseverative errors more frequent among semantically related targets, given well-matched stimulus sets? I expected that repetitions of immediately preceding responses would occur, and would be more frequent in the semantically homogeneous set. Repetitions among semantically related items, but not unrelated items, would be consistent with a semantic interference effect. However, given prior evidence that somewhat younger children also produce unrelated perseverations (Gershkoff-Stowe, Reference Gershkoff-Stowe2002), the possibility of such effects in 3-year-olds could not be ruled out.

2. 2. Do 3-year-old children show semantic interference effects that are cumulative? I anticipated that interference would increase across blocks, and that the increase would be greater for the semantically homogenous condition. Because it involves change over time, a cumulative semantic interference effect would be more consistent with a learning mechanism than a residual activation mechanism and would lay the foundation for future confirmatory work. Because residual activation is thought to be rather short lived, it would not be expected to accrue beyond a very small number of trials, and thus consistent growth of interference would not be expected. The strongest evidence for cumulative semantic interference would be a result showing similar outcomes for the two conditions at the start of the experiment, followed by decrements in performance in the semantically homogenous condition across time, measured as increases in errors and/or increases in RT.

3. Alternatively, other outcomes might more easily be interpreted as being due to residual activation or other factors. For example, a stable pattern (i.e., not increasing across the experiment) of worse performance in the semantically homogeneous condition might be indicative of short-term residual activation effects within that condition (or less interestingly, poor matching of the conditions resulting in the homogenous condition containing more challenging items). A pattern of frequent, immediate repetition errors in one or both conditions might also be interpreted as more likely based in residual activation than learning.

RESULTS

Children's response accuracy and perseverative errors are addressed first, before turning to RTs. For the analyses of response accuracy and RT, the effects of condition, block, and their interaction on performance were investigated using linear mixed-effects models with participants and items as crossed random effects. Block (1–4) was included in these analyses as a numerical predictor for the analysis of change over time. Treatment coding was used for the categorical predictor, condition, with the animal condition as the reference. Analyses were implemented using the lme4 package (Bates, Maechler, Bolker, & Walker, Reference Bates, Maechler, Bolker and Walker2015) in R (R Core Team, 2015), with random intercepts for participants and target items. Models were compared using the Akaike information criterion and likelihood ratio (χ²). The figures present the observed means and standard errors calculated across all subjects and trials (i.e., not averaged within each subject).

Accuracy

Out of 1,047 usable responses (animal condition = 522; mixed condition = 525), there were 190 errors (18%). Of these, 106 (56%) contained a hesitation; 30 (16%) were a nonresponse, timeout or unintelligible; and 54 (28%) contained a naming error. Of the naming errors, 44 (81%) involved the production of another item from the same stimulus set. Response set errors accounted for virtually all of the naming errors in the animal condition (99%) and approximately half (53%) of the naming errors in the mixed condition. The remaining 9 errors in the mixed set varied: naming part of the target (e.g., toe/foot), and providing related but incorrect responses (e.g., coffee/cup, climb up/ladder, shirt/dress, moon/balloon). The likelihood of an accurate response was analyzed with mixed logit modeling, using the glmer function within the lme4 package (Bates et al., Reference Bates, Maechler, Bolker and Walker2015). Note that the analysis produces results in log odds of the outcome. For ease of interpretation, however, descriptive data are given in the original unit of measurement, accuracy. Figure 1 presents the mean response accuracy for each combination of condition and block. As can be seen, Figure 1 points to somewhat higher accuracy for the animal condition than the mixed condition in Block 1, followed by higher accuracy for the mixed condition than the animal condition in subsequent blocks.

Figure 1.

Mean response accuracy by condition and block.

Table 1 presents the outcomes of successive model comparisons. Neither condition nor block showed a significant main effect, nor did the interaction between the two. An analysis by trials rather than blocks also showed no significant effects.

Table 1.

Model outcomes for accuracy

Note: AIC, Akaike information criterion; M1–4, Models 1–4 included random intercepts for participants and targets, in addition to the fixed effects of interest.

Perseverative responses

Perseverative responses were calculated as the number of repetitions out of the total number of naming errors that named an item within the current response set. Thirteen of the 19 participants produced at least 1 such repetition error. Overall, there were 34 response set errors in the animal condition, and 10 response set errors in the mixed condition. Within the animal condition, 56% of response set errors were repetitions of prior responses (n = 19); within the mixed condition, this value was 60% (n = 6). By necessity, all of the repetitions in the animal condition were semantically related to the target by the category of animal. Four of the six repetitions in the mixed condition were judged to be semantically related to the target, based on thematic (hand/mitten; hand/glove) or taxonomic relationships (firetruck/tractor; chair/table). Of the remaining two, one shared a phonological relationship (moon/balloon). The other, although coded as not semantically related, could be argued to be related within a child's experience based on event schema (clown/birthday cake; coded as a repetition of cake). Given the low number of repetition errors, inferential analyses were not conducted. For comparison to the prior literature, however, Figure 2 presents the distribution of repetition errors in the animal condition as a function of lag. As can be seen, there were no immediate repetitions, with the first repetition emerging at a lag of 4 (i.e., with three responses intervening). The vast majority of the repetition errors occurred at lags ranging from 4 to 16, with one error at a lag of 23. The distribution of repetitions in the mixed condition was not notably different from that for the animals. Repetition errors occurred at lags of 2, 6, 8, 10, 21, and 23. Again, there were no immediate repetitions.

Figure 2.

Total number of perseverative response errors in the animal condition, by lag.

Response latencies (RT)

The analysis of RTs included all correct responses that were produced without a vocal hesitation or response time out. Two children had usable RTs from less than 30% of trials, and so they were excluded from the RT analysis. For trials containing errors in the use of the voice key (e.g., early triggers due to lip pops), RTs were recaptured manually by offsetting the time between the event that had triggered the voice key and the actual onset of the response. In the event that the triggering event could not be unambiguously identified, the response was discarded as unusable. Prior to analysis, the RTs from usable trials were screened for extreme values. RT values that were more than 3 SD above the mean for each Condition × Block cell were removed as potential outliers. In total, 26 data points were trimmed as potential outliers (3.2% of RTs), leaving 781 data points for analysis.

Figure 3 presents the mean RTs for each combination of condition and block. As can be seen, the response latencies are very similar in the two conditions in Block 1. Response latencies appear to increase over the successive blocks, with a greater apparent increase in the animal condition.

Figure 3.

Mean response time by condition and block.

The analyses supported these observations. Table 2 presents the outcomes of successive model comparisons for RT. Both condition (model M1) and block (M2), when added as main effects, produced significant effects and were preferred over a model containing only the random intercepts for participants and items. A model containing both main effects (M3) was preferred over models containing only one of the two predictors. The final model, containing the interaction between condition and block (M4), was preferred over the model containing main effects only. The final model revealed statistically significant predicted gains in RT as a function of block for each condition, but with a greater degree of slowing in the animal condition: 65 ms (p < .001) for the animal condition versus 23 ms (p = .005) for the mixed condition. An analysis by trial rather than block revealed a similar pattern of results. The final model containing the interaction between condition and trial was preferred over a model containing main effects only, χ² (1) = 4.76, p = .03. The slopes predicted by this model were statistically significant for both conditions, with a predicted increase in response latency of 8.2 ms per trial in the animal condition (p < .001) and 3.7 ms for the mixed condition (p = .001).

Table 2.

Model outcomes for RT

Note: AIC, Akaike information criterion; M1–4, Models 1–4 included random intercepts for participants and targets, in addition to the fixed effects of interest.