Adaptive rule learning of event sequences during the A-not-B task in 9-month-old infants

Abstract

Prior work indicates that infants can use social information to organize simple audiovisual inputs into predictable rules by 8 months of age. However, it is unclear whether infants can use social information to organize more complex events into predictable rules that can be used to guide motor action. To examine these issues, we tested 9-month-old infants using a modified version of an A-not-B task, in which hiding event sequences were paired with different experimenters, who could be used to organize the events into rules that guide action. We predicted that infants’ reaching accuracy would be better when the experimenter changes when the toy’s hiding location changes, relative to when the experimenter stays the same, as this should cue a novel rule used to guide action. Experiments 1 and 2 validated this prediction. Experiment 3 showed that reaching accuracy was better when the toy’s hiding location switched but was consistent with the rule associated with the experimenter, relative to when the toy’s hiding location repeated but was inconsistent with the rule associated with the experimenter. These data suggest that infants can use the identities of experimenters to organize events into predictable rules that guide action in the A-not-B task.

1 |. INTRODUCTION

The ability to flexibly align behavior with relevant goals for learning and action is critical for adaptation and survival in a complex and rapidly changing multisensory world. This flexibility is afforded in part by the ability to organize events into predictable structures or sequences that guide learning and action. Previous work has suggested that infants can learn to parse events across visual (Johnson et al., 2009; Wu, Qian, & Aslin, 2019), linguistic (Gómez, 2002; Marcus, Vijayan, Bandi Rao, & Vishton, 1999; Saffran, Aslin, & Newport, 1996), action (Hernik & Csibra, 2015; Mandler & McDonough, 1995; Sommerville, Woodward, & Needham, 2005; Stahl, Romberg, Roseberry, Golinkoff, & Hirsh-Pasek, 2014), and multimodal domains (Bahrick & Lickliter, 2000; Bahrick, Lickliter, Castellanos, & Todd, 2015; Gogate & Bahrick, 1998). The ability to detect and organize events into flexible representations helps infants learn to perform goal-directed actions (Barr & Hayne, 1996), infer intentional states of others (Baldwin & Baird, 1999; Brugger, Lariviere, Mumme, & Bushnell, 2007), and generalize actions across contexts (Brito & Barr, 2012; Hayne, Boniface, & Barr, 2000; Learmonth, Lamberth, & Rovee-Collier, 2004). Here we expand on prior work to explore whether infants use social information, in the form of the identities of experimenters conducting the task, to organize event sequences into predictable rules that subsequently guide motor action in the A-not-B task.

Our approach is broadly drawn from research in hierarchical action and reinforcement learning (Collins, Cavanagh, & Frank, 2014; Collins & Frank, 2013; Donoso, Collins, & Koechlin, 2014; Frank & Badre, 2012). In reinforcement learning, the goal is to learn stimulus–action–outcome (S-A-O) rules or associations. What makes this learning hierarchical is that the S-A-O rules are cued by a higher-order context. The context can be a space, an object, or a person. In other words, higher-order contexts can be used to cue lower-order rules (S-A-O associations) that are most appropriate for a given context, thereby reducing complexity of learning problems. Prior research reveals that infants as young as 8 months of age can organize audiovisual inputs into predictable rules or sequences of this sort during incidental learning, even in environments where there is no immediate or obvious benefit to doing so (Lewkowicz, Schmuckler, & Mangalindan, 2018; Werchan, Collins, Frank, & Amso, 2015; Werchan & Amso, 2020). Other data show that infants can use social information, in the form of face-voice contexts, to organize different sets of spoken labels for the same objects (Werchan et al., 2015; Werchan et al., 2016). Here, we build off of these findings to examine whether infants can use social information to parse more complex event sequences into predictable S-A-O rules that guide action. To examine this, we used a modified version of the well-documented A-not-B task to explore whether infants attempt to use the identities of the experimenters conducting the task to organize events into predictable S-A-O rules that subsequently guide motor action.

In the A-not-B task, infants experience a specific set of event sequences that require them to integrate inputs and select motor actions over time. They first interact with a friendly experimenter (a person context), a fun toy (stimulus) is then shown and hidden in one of the two locations and they are then allowed to search for the toy after a brief delay (action–outcome). This hiding event sequence is repeated several times, with the experimenter hiding the toy in the same location (A) for multiple trials, after which it is switched and hidden in the alternate location (B). The A-not-B error occurs when infants reach to the previously correct location when the hiding location is switched from the A well to the B well (Piaget, 1954). Key studies in human and non-human primate infants indicate that the PFC plays a critical role in success and failure on the A-not-B task (Diamond & Goldman-Rakic, 1989), as well as the relative strength of working memory representations formed online during learning (Munakata, 1998; Munakata, McClelland, Johnson, & Siegler, 1997). More recent work also suggests that social information present during standard versions of this task, such as the experimenter’s communicative behaviors, also influences infants’ search behavior (Dunn & Bremner, 2019; Topal, Gergely, Miklosi, Erdohegyi, & Csibra, 2008).

Here, we build on these findings and adapt this task to examine whether infants might attempt to use social information offered during the A-not-B task to organize the hiding event sequences into predictable S-A-O rules that can be used to guide action. Specifically, we ask whether infants attempt to use the identity of the experimenter conducting the task to organize the events into predictable S-A-O rules that guide action. As such, our primary manipulations involve the persons hiding the toys.

There is ample evidence motivating the premise that adults and caregivers guide infants’ attention and learning (Csibra & Gergely, 2006; Hood, Willen, & Driver, 1998; Scaife & Bruner, 1975; Tummeltshammer, Feldman, & Amso, 2019; Vygotsky, 1980). For instance, infants prefer infant-directed to adult-directed speech from birth (Cooper & Aslin, 1990) as well faces with direct relative to averted gaze (Farroni, Csibra, Simion, & Johnson, 2002; Farroni, Menon, & Johnson, 2006), and they use these sources of communicative information to support learning (D’Entremont, Hains, & Muir, 1997; Farroni, Massaccesi, Pividori, Simion, & Johnson, 2004; Gelman et al., 1998; Gredeback, Theuring, Hauf, & Kenward, 2008; Thiessen, Hill, & Saffran, 2005). Related work shows that as early as 8 months of age, infants monitor the reliability of an adult’s gaze, and then use this information to make predictions about future events (Tummeltshammer, Mareschal, & Kirkham, 2014). Similarly, other data show that infants are more likely to imitate actions over delays when adults provide social or narrative cues while demonstrating action sequences to infants and toddlers (Brugger et al., 2007; Simcock, Garrity, & Barr, 2011; Zimmermann, Moser, Lee, Gerhardstein, & Barr, 2017). Moreover, in the context of the A-not-B task, there are reports that infants primarily make perseverative errors when adults use communicative cues during the task (Dunn & Bremner, 2019; Topal et al., 2008).

Based on these findings, we hypothesized that infants may attempt to use the identity of the experimenter hiding the toys to organize events into S-A-O rules that guide search behavior. From this vantage point, what may seem like a maladaptive search strategy (reaching to well A when the experimenter now hid the toy in well B) may in fact reflect learning of a predictable sequence that specifies where they should search to find the toy based on the identity of the experimenter hiding the toy (Figure 1). Put simply, infants may use the identity of the experimenter to learn S-A-O rules for where the experimenter is most likely to hide the toy on any given trial (e.g., experimenter 1 hides the toy in well A and does not hide it in well B, so when that experimenter hides a toy on any given trial, it will most likely be hidden in well A). Thus, rather than acting on a trial-by-trial basis with the singular goal of finding a hidden toy, infants may experience this task as a series of events that can be organized into predictable S-A-O rules that support learning and action more generally. This possibility is broadly consistent with prior views suggesting that infants may interpret communicative information in standard versions of the A-not-B task as a cue that adults are attempting to convey generalizable, rather than episodic, knowledge (Topal et al., 2008).

FIGURE 1.

Infants may attempt to use the identities of the experimenters to organize event sequences into predictable rules (S-A-O contingencies), which are then used to guide search behavior

Therefore, in the current study, we used a modified version of an A-not-B task to examine whether infants can use the identities of experimenters conducting the task to organize events into predictable S-A-O rules that guide action. Notably, our goal is not to arbitrate among theories of the A-not-B error, as our predictions are consistent with multiple theoretical approaches (Diamond, 2001; Marcovitch & Zelazo, 1999; Munakata, 1998; Smith, Thelen, Titzer, & McLin, 1999; Topal et al., 2008), but rather to use the event structure of this task to inform the value of experimenter identities as cues for parsing events into predictable S-A-O rules that guide action. We tested 9-month-old infants, an age-group capable of organizing simple audiovisual inputs into predictable rules (Werchan et al., 2015; Werchan et al., 2016; Lewkowicz et al., 2018), and who consistently show perseverative errors in standard versions of the A-not-B task (Diamond, 2002). In Experiments 1 and 2, we examined whether infants can use the identity of the experimenter hiding the toy to organize events into predictable S-A-O rules that guide action (Figures 2 and 5). We hypothesized that a change in the experimenter hiding the toy would cue a novel S-A-O rule, leading to fewer perseverative errors when the toy’s hiding location changes on switch trials. However, when the experimenter hiding the toy remains the same when the toy’s hiding location changes in standard versions of the task, we predicted that infants would search in the location that was previously associated with the experimenter, consistent with using the experimenter to contextualize S-A-O rules that guide search behavior. In this view, reaching to the previously correct location on switch trials does not reflect a perseverative error per se, but instead reflects the rational application of an adaptive search strategy based on using the experimenter to organize learning of predictable rules that guide action. In Experiment 3, we tested the critical prediction that repeat trials that are inconsistent with the S-A-O rule cued by the experimenter (i.e., where the experimenter should have hidden the toy) should result in worse performance than switch trials that are consistent with the S-A-O rule cued by the experimenter. This is even though switch trials require infants to inhibit a prepotent or habitual motor response, whereas repeat trials require infants to repeat a prepotent motor response (Figure 6). In other words, if a change in the experimenter hiding the toy cues a novel S-A-O rule, then this should result in more accurate reaches even on switch trials that require inhibiting a prepotent motor response.

FIGURE 2.

Schematic of the between-subjects design used in Experiment 1. In the standard condition, the experimenter remained the same across all trials (a). Two experimenters conducted the task in the two experimenter standard condition, but the experimenter remained the same across repeat and switch trials (b). In the consistent S-A-O rule condition, a change in the hiding event sequence was paired with a change in the experimenter administering the sequence (c)

FIGURE 5.

Repeat trial search accuracy (a) and number of repeat trials to reach the search criterion (b) across groups in Experiment 2. Group means are shown with individual data points overlaid. Error bars reflect SEM

FIGURE 6.

Schematic of the within-subjects design in Experiment 3. Infants first received two consistent S-AO rule blocks (a), which had two consistent repeat trials followed by one consistent switch trial, in which a change in the hiding event sequence was paired with a change in the experimenter. Infants then received two inconsistent S-A-O rule blocks (b), which had two consistent repeat trials followed by one inconsistent repeat trial, in which the hiding event sequence remained the same but was paired with a change in the experimenter

2 |. GENERAL METHOD

2.1 |. Participants

The final sample across all three experiments consisted of 119 nine-month-old infants (M = 9.21 months, SD = 0.87 months, 66 females, 53 males, 83 white non-Hispanic, 13 black, 8 Hispanic, 7 Asian, 7 Mixed Race/Other, and 1 American Indian). Infants were recruited from the state department of health birth records and through community advertisements. Infants were prescreened for premature birth (<36 weeks), low birthweight (<5 lb), or a history of serious health problems. The Brown University Institutional Review Board approved the study, and parental consent was obtained prior to testing.

2.2 |. Task apparatus

Infants were seated on their parent’s lap across a table from the experimenter(s). The A-not-B apparatus was designed following Diamond and Goldman-Rakic (1989) and consisted of a small blue-felt-covered table that measured 60 cm (L) × 40 cm (W). Two wells were embedded within this table, which were 9 cm in diameter, 11.5 cm deep, and 30 cm apart from center to center. Red felt cloths were used to cover the wells, which measured 22 cm (L) × 15 cm (W). The height of the parent’s seat and the apparatus was adjusted such that the infant could see the hiding wells while they were seated on their parent’s lap. The toys were attractive, brightly colored toys.

2.3 |. Procedure

Each experimental session consisted of a familiarization period followed by two blocks of test trials and was modeled after Diamond and Goldman-Rakic (1989) and Munakata (1998). The familiarization period was used to allow infants to become comfortable with the toys and experimenters. The two test blocks consisted of multiple repeat trials followed by one switch trial. Infants’ reaching accuracy was the dependent measure. Parents were instructed to not encourage, guide, or correct the infant’s actions during the study, and to keep infants seated on their lap facing forward throughout the study.

The familiarization period began with the experimenter introducing the toy to the infant and allowing them to play with it to gain familiarity with the toys and experimenters. The experimenter then partially hid the toy and encouraged the infant to search for the toy. Next, the experimenter placed the toy in one of the wells and covered it with the red felt cloth, and allowed the infant to search for the toy immediately. The infant was encouraged to remove the cover and search for the toy if they did not do so independently. They were then praised and allowed to play with the rewarding toy. This practice trial was repeated in the other well. The side of the well that the toy was first hidden in during familiarization was counterbalanced across subjects.

After familiarization, infants received the test trials. Each trial began with the experimenter waving the toy centered over the task apparatus to capture the infant’s attention. The experimenter then slowly placed the toy in the A well. If the infant looked away while the toy was being hidden, the experimenter repeated the hiding procedure. The experimenter then covered the well with the red felt cloth, and audibly counted to three to draw the infant’s visual attention away from the well. After the three second delay, the experimenter pushed the apparatus to the edge of the table and allowed the infant to search for the toy. If the infant searched for the toy in the correct well, they were praised and rewarded by being allowed to play with the toy for a few seconds. If the infant did not search correctly, the experimenter showed the infant where the toy was hidden, but the infant was not praised or allowed to play with the toy. The experimenter repeated hiding the toy in the A well until the infant searched correctly three consecutive times (repeat trials). Next, a switch trial occurred by hiding the toy in the B well using the same hiding procedure. This block of trials was then repeated by hiding the toy repeatedly in the B well until the infant searched correctly three consecutive times (repeat trials), and then switching back to the A well (switch trial). Thus, there was a total of two testing blocks. The left–right assignment of the A and B hiding wells was counterbalanced across subjects. Also note that since the toy’s location is reversed more than one time in the present study, we use the term “repeat trial” to refer to trials when the toy is repeatedly hidden in the same location, and the term “switch trial” to refer to trials when the hiding location switches to the other well.

2.4 |. Coding

All testing sessions were videotaped for subsequent coding by a trained observer. A reach was defined as an action that resulted in contact and removal of the red felt cloth. The first location that infants reached to on each trial was scored as the response. A subset of the videos (25%; n = 30) were rated by a second independent observer. Reliability of the ratings was 100%.

3 |. EXPERIMENT 1

In our first experiment, we examined whether infants can use the identity of the experimenter conducting the task to organize the hiding event sequences into predictable S-A-O rules that guide action (Figure 2). We tested three different conditions to examine this hypothesis. In a standard condition, only one experimenter conducted the task. In a two-experimenter standard condition, two different experimenters conducted the task, and the same experimenter hid the toy when the toy’s hiding location changed on switch trials. This condition was tested to ensure that the inclusion of two experimenters did not impact typical performance on standard versions of the A-not-B task. Finally, in a consistent S-A-O rule condition, two different experimenters conducted the task, and a change in the toy’s hiding location was paired with a change in the experimenter hiding the toy on switch trials. We predicted that if infants use the identity of the experimenters to organize the hiding event sequences into rules for action, then we expected to find superior performance on the switch trials in the consistent S-A-O rule condition relative to the standard conditions.

3.1 |. Method

3.1.1 |. Participants

The final sample consisted of 63 nine-month-old infants (M = 9.16 months, SD = 0.90 months, 35 females, 28 males, 43 white non-Hispanic, 7 black, 4 Hispanic, 5 Asian, and 4 Mixed Race/Other).

Sample size was determined based on an a priori power analysis with a large effect size (f = 0.4) estimated from prior work (Werchan et al., 2015) at 80% power, which indicated that approximately 22 infants per condition would provide sufficient statistical power. Infants (N = 63) were randomly assigned to one of three conditions: a standard condition (n = 21), a two-experimenter standard condition (n = 21), or a consistent S-A-O rule condition (n = 21). An additional 23 infants were tested, but excluded from the sample due to fussiness resulting in a failure to complete the task (n = 19), experimenter error (n = 2), or parental interference (n = 2). Approximately equal numbers of infants were excluded from each condition (standard condition: n = 5, two-experimenter standard condition: n = 9, consistent S-A-O rule condition: n = 9).

3.1.2 |. Procedure

We tested three between-subjects conditions: a standard condition (Figure 2a), a two-experimenter standard condition (Figure 2b), and a consistent S-A-O rule condition (Figure 2c). The procedure was as described in the General Method section with the following modifications relevant to each condition below.

In the standard condition, one experimenter hid a single toy hid the toy in the A well until the infant searched correctly three consecutive times (E1A/repeat trials), and they then hid it in the B well (E1B/switch trial), which was inconsistent with the S-A-O rule associated with the experimenter. This sequence was repeated, with the experimenter hiding the toy in the B well until the infant searched correctly three consecutive times (E1B/repeat trials), after which they again hid it in the A well (E1A/switch trial).

In the two-experimenter standard condition, two experimenters were seated side-by-side. Experimenter 1 hid the toy in the A well until the infant searched correctly three consecutive times (E1A/repeat trials), and they then hid it in the B well (E1B/switch trial), which was inconsistent with the S-A-O rule associated with that experimenter. Experimenter 2 then hid their toy in the B well until the infant searched correctly three consecutive times (E2B/repeat trials), and they then hid it in the A well (E2A/switch trial), again inconsistent with the S-A-O rule associated with that experimenter. Each experimenter hid a unique toy. This condition allowed us to test whether using two experimenters during the task impacts performance for reasons unrelated to associating each hiding well with a unique experimenter.

In the consistent S-A-O rule condition, two experimenters hid the toys, and a change in the toy’s hiding location was paired with a change in the experimenter hiding the toy. Thus, experimenter 1 hid their toy in the A well until the infant searched correctly three consecutive times (E1A/repeat trials). Then, experimenter 2 hid their toy in the B well (E2B/switch trial). They continued hiding it in the B well until the infant searched correctly three consecutive times (E2B/repeat trials), after which experimenter 1 again hid their toy in the A well (E1A/switch trial). In this way, trials involving a switch in the toy’s hiding location were associated with a change in the experimenter hiding the toy, thus cueing a novel S-A-O rule on trials that otherwise require inhibiting a prepotent response. Note that each experimenter hid a unique toy in the consistent S-A-O rule condition, which was counterbalanced across subjects.

The experimenters in all conditions were young Caucasian females, and only differed in identity and not in race, age, or gender. Additionally, the same two experimenters hid the toys across all conditions, and the left–right seating assignments of the experimenters was counterbalanced across subjects.

3.1.3 |. Analysis plan

Our dependent measures were as follows: (a) number of repeat trials to search criterion across blocks; (b) repeat trial accuracy across blocks, defined as the proportion of correct searches on repeat trials; and (c) switch trial accuracy across blocks, which was defined as number of infants who reached to the correct location on the switch trial. We examined the proportion of correct searches on the repeat trials, as there were multiple repeat trials per block. However, for switch trial performance, we examined the number of infants who searched correctly, given that there was only one switch trial per testing block. We predicted to find no differences in repeat trial accuracy across groups; however, if infants attempt to use the experimenters to organize S-A-O rules for action, then we expected that more infants would search correctly on the switch trial in the consistent S-A-O rule condition relative to the two standard conditions.

3.2 |. Results

3.2.1 |. Repeat trial performance

We first examined group differences in infants’ repeat trial accuracy (proportion of correct searches) across the first and second blocks of testing. A mixed-effects ANOVA with repeat trial accuracy as a dependent variable, group as a between-subjects variable, and block as a within-subjects variable revealed a significant effect of Block, F(1, 60) = 7.68, p = .007, η_p² = 0.11, indicating that infants had better repeat trial accuracy on the first relative to the second block of testing (Figure 3a). However, there was no effect of Group, F(2, 60) = 0.51, p = .61, nor was there a Group × Block interaction, F(2, 60) = 1.05, p = .36, indicating that performance did not differ across groups.

FIGURE 3.

Repeat trial performance across groups in Experiment 1. There was no difference between groups in the proportion of correct searches on repeat trials in the first and second blocks of testing (a). There was no difference between groups in the number of repeat trials to reach the search criterion in the first and second blocks of testing (b). Group means are presented with individual data points overlaid. Error bars reflect SEM

We next analyzed whether there were group differences in the number of trials to meet the search criterion (searching correctly across three consecutive trials) across testing blocks. A mixed-effects ANOVA with the number of trials to criterion as a dependent variable, group as a between-subjects variable, and block as a within-subjects factor revealed a trending effect of Block, F(1, 60) = 3.42, p = .07 η_p² = 0.05, indicating that infants as a group required marginally fewer trials to reach the search criterion on the first relative to the second block of testing (Figure 3b). However, there was no effect of Group, F(2, 60) = 0.18, p = .84, nor was there a Group × Block interaction, F(2, 60) = 1.96, p = .15, indicating that performance did not differ across groups.

3.2.2 |. Switch trial performance

We next examined group differences in the number of infants who reached to the correct location on the switch trial across the first and second blocks of testing (Table 1). A Kruskal–Wallis test revealed no significant difference across groups during the first block of testing, χ² (2, N = 63) = 3.77, p = .15, but there was a significant difference across groups during the second block of testing, χ² (2, N = 63) = 8.25, p = .016, r = 0.36. Follow-up analyses on the first switch trial during the second block of testing using Mann–Whitney U tests revealed no significant differences in switch trial errors between the standard and two-experimenter standard conditions, U = 220.5, p = 1.00. However, Mann–Whitney U tests indicated that significantly more infants reached correctly in the Consistent S-A-O Rule condition relative to both the Standard condition, U = 136.5, p = .015, f = 0.31, and the Two-Experimenter Standard condition, U = 136.5, p = .015, f = 0.31.

TABLE 1.

Number of infants (out of 21) who searched correctly on the first switch trial in Experiment 1

Condition	Block 1	Block 2
Standard (n = 21)	5	6
Two-experimenter standard (n = 21)	7	6
Consistent S-A-O rule (n = 21)	11	14

Finally, we examined whether there was a difference within groups in the number of infants who searched in the correct location on the switch trial during the first relative to the second block of testing. McNemar exact tests indicated that there were no significant differences in the number of infants who reached correctly on the first relative to the second switch trial for the Standard condition, p = 1.00, the Two-Experimenter Standard condition, p = 1.00, or for the Consistent S-A-O Rule condition, p = .61. Taken together, these findings suggest that there is lower perseverative behavior when a different experimenter hides the toy on switch trials, consistent with using the identities of the experimenters to organize the hiding events into predictable S-A-O rules used to guide behavior.

3.3 |. Discussion

Our findings provide initial support for our hypothesis that infants can use the identities of experimenters to organize appropriate S-A-O rules for action. Specifically, infants made perseverative errors on switch trials when these trials were inconsistent with the S-A-O rule cued by the experimenter (standard conditions), but not when these trials were consistent with the S-A-O rule cued by the experimenter (consistent S-A-O rule condition). Our results also indicated that group differences in switch trial performance were most pronounced during the second block of testing, but these group differences were not statistically significant during the first block of testing. This finding is consistent with the interpretation that infants attempt to use the experimenters to learn S-A-O rules for action. Specifically, a rule learning interpretation of A-not-B performance would predict that infants should have better performance in the second relative to first block of testing, given that infants have increased observations of the experimenters conducting the event sequences. This would provide additional opportunities to contextualize S-A-O rules for action, leading to better performance in the second block of testing. As such, our findings support the hypothesis that infants may attempt to use the identities of the experimenters to contextualize appropriate S-A-O rules for action.

However, there are two alternative perception-level explanations of our results. First, using two different toys, which may help infants differentiate switch trials from repeat trials, could have driven better performance on the consistent S-A-O rule condition for reasons unrelated to using the identities of the experimenters to organize S-A-O rules for action (Figure 2c). In other words, infants could simply use the identities of the toys, rather than the experimenters, to determine where to search. A second possibility is that infants might have used the experimenter’s left–right seating location to determine the correct hiding well to search in, given that each experimenter was seated on the same side as the well that they hid the toy in (Figure 2c). That is, rather than using the experimenters to contextualize S-A-O rules for action, it is possible that infants may simply search in the location directly in front of the experimenter. Thus, Experiment 2 aimed to replicate Experiment 1 while controlling for these alternative perceptual-level explanations.

4 |. EXPERIMENT 2

We tested two additional conditions in Experiment 2 to control for alternative perceptual-level explanations of our findings from Experiment 1. First, to ensure that performance in the consistent S-A-O rule condition was not driven by using two different toys in Experiment 1, we tested a consistent S-A-O rule same-toy condition. This condition was identical to the consistent S-A-O rule condition in Experiment 1, except the two experimenters hid the same toy, rather than different toys, across all trials. We predicted that if infants use the experimenters to organize the event sequences into S-A-O rules for action, rather than using the toys as a cue for where to search, then infants in the consistent S-A-O same-toy condition in Experiment 2 should show superior switch trial accuracy relative to the standard condition in Experiment 1. We also tested a consistent S-A-O rule location-change condition to ensure that infants were not simply searching in the hiding well directly in front of the experimenter on each trial. In this condition, the switch trials consisted of a change in the experimenter’s seating location, rather than a switch in the toy’s hiding location. If better performance in Experiment 1 was due to infants simply searching in front of the experimenter, then we predicted that infants should show poor search performance on the switch trials in the S-A-O rule location-change condition. However, if infants instead use the identities of the experimenters as a latent cue to organize S-A-O rules for action, then we expected infants to search in the hiding location consistent with the S-A-O rule cued by the experimenter on switch trials, rather than in the hiding location directly in front of the experimenter. This condition controls for the possibility that infants might attempt to use the location of the experimenters to determine where to search, rather than the identities of the experimenters hiding the toys to organize S-A-O rules for action.

4.1 |. Method

4.1.1 |. Participants

We tested 42 nine-month-old infants who were randomly assigned to a consistent S-A-O rule same-toy condition (n = 21) or a consistent S-A-O rule location-change condition (n = 21). The sample in the consistent S-A-O rule same-toy condition consisted of 21 nine-month-old infants (M = 9.25 months, SD = 0.81 months, 10 females, 11 males, 17 white non-Hispanic, 1 Hispanic, 2 black, and 1 American Indian). An additional six infants were tested, but excluded from the sample due to parental interference (n = 1) or fussiness (n = 5). The sample in the consistent S-A-O rule location-change condition consisted of 21 nine-month-old infants (M = 9.21 months, SD = 0.83 months, 11 females, 10 males, 15 white non-Hispanic, 1 black, 2 Hispanic, 1 Asian, and 2 Mixed Race/Other). An additional 12 infants were tested, but excluded from the sample due to failure to complete the task due to fussiness.

4.1.2 |. Procedure

The procedure for Experiment 2 was identical to the consistent S-A-O condition in Experiment 1 with minor modifications relevant to each condition. For the consistent S-A-O rule same-toy condition, rather than each of the two experimenters hiding a different toy, the two experimenters hid the same toy on each trial (Figure 4a). Thus, experimenter 1 would hide the toy in the A well until the infant searched correctly three consecutive times (E1A/repeat trials). Then experimenter 2 would hide the same toy in the B well (E2B/switch trial), consistent with the S-A-O rule cued by the experimenter. They would then continue hiding it in the B well until the infant searched correctly three times (E2B/repeat trials). Finally, experimenter 1 would hide the same toy in the A well (E1A/switch trial).

FIGURE 4.

Schematic of the between-subjects design used in Experiment 2. In the consistent S-A-O rule same-toy condition, the experimenters hid the same toy across all trials (a). In the consistent S-A-O rule location-change condition, the switch trials consisted of a change in the left-right seating location of the experimenter (b)

The modification for the consistent S-A-O rule location-change condition was such that on switch trials, rather than the hiding location of the toy changing to the other well, the spatial location of the experimenter changed (i.e., the two experimenters swapped left–right seating locations; Figure 4b). The experimenter and the hiding well remained the same, and the same toy was hidden across all trials. Thus, experimenter 1 would hide the toy in the A well until the infant searched correctly three consecutive times (left-E1A/repeat trials). Then, the experimenters swapped left–right seating locations, and experimenter 1 again hid the toy in the A well (right-E1A/switch trial), consistent with the S-A-O rule cued by the experimenter. The experimenters then returned to their original seating locations. Experimenter 2 then repeated this procedure, hiding the toy in the B well until the infant searched correctly three consecutive times (right-E2B/repeat trials) after which the experimenters swapped left–right seating locations, and experimenter 2 again hid the toy in the B well (left-E2B/ switch trial).

4.1.3 |. Analysis plan

As in Experiment 1, our dependent measures were as follows: (a) number of repeat trials to search criterion across blocks; (b) repeat trial accuracy across blocks, defined as the proportion of correct searches on repeat trials; and (c) switch trial accuracy across blocks, which was defined as number of infants who reached to the correct location on the switch trial. Search accuracy for repeat trial performance was operationalized as the proportion of correct searches, given that each infant received multiple repeat trials per block. Search accuracy for switch trial performance was operationalized as the total number of infants who searched correctly, given that infants only received one switch trial per block.

4.2 |. Results

4.2.1 |. Consistent S-A-O rule same-toy condition

We compared infants’ repeat trial reaching accuracy (proportion of correct searches) during the first (M = 0.87, SD = 0.16) and second (M = 0.67, SD = 0.24) blocks of testing using a two-tailed paired samples t test, which revealed that infants performed significantly better in the first block of testing, t(21) = 3.47, p = .002, d = 0.74 (Figure 5a). We then examined the number of trials to meet the search criterion (searching correctly across three consecutive trials) across the first (M = 4.00, SD = 1.48) and second (M = 7.19, SD = 4.52) blocks of testing. A two-tailed paired samples t test revealed that infants required more trials to meet the search criterion during the second relative to the first block of testing, t(21) = –3.09, p = .006, d = 0.67 (Figure 5b).

Next, we used a McNemar exact test to examine whether there were differences in the number of infants who searched correctly on the switch trial between the first and second blocks of testing (Table 2), which revealed no significant differences, p = .09. We then used binomial tests to examine whether the number of infants who searched correctly on the first switch trial was significantly different from chance (50%) on the first and second blocks of testing. These analyses indicated that the number of infants who searched correctly during the first switch trial was no different from chance during the first block of testing, p = 1.00. However, the number of infants who searched correctly on the switch trial during the second block of testing (17 out of 21 infants) was significantly greater than chance, p = .007, Cohen’s g = 0.31. Note that reaching accuracy above chance on switch trials is particularly notable given that switch trial accuracy is significantly below chance (indicating perseverative behavior) in standard versions of the task. A direct comparison of switch trial performance on this condition relative to the standard condition in Experiment 1 using a Mann–Whitney U test shows that there is better performance on the second switch trial in the consistent S-A-O rule same-toy condition, U = 105.00, p = .001, f = 0.24. This indicates that differences in switch trial accuracy between the consistent S-A-O rule condition and standard versions of the task were not likely due to using two different toys.

TABLE 2.

Number of infants (out of 21) who searched correctly on the switch trial across blocks in Experiment 2

Condition	Block 1	Block 2
Consistent S-A-O rule location change (n = 21)	16	13
Consistent S-A-O rule same toy (n = 21)	9	17
Standard (for comparison; n = 21)	5	6

4.2.2 |. Consistent S-A-O rule location-change condition

We compared infants’ repeat trial reaching accuracy (proportion of correct searches) during the first (M = 0.81, SD = 0.27) and second (M = 0.77, SD = 0.17) blocks of testing using a two-tailed paired samples t test, which revealed no significant difference, t(21) = 0.42, p = .68 (Figure 5a). We then examined the number of trials to meet the search criterion across the first (M = 4.24, SD = 1.84) and second (M = 5.52, SD = 2.99) blocks of testing. A two-tailed paired samples t test revealed that there was no significant difference in the number of trials to criterion, t(21) = −1.61, p = .12 (Figure 5b).

We then used a McNemar exact test to examine differences in the number of infants who searched correctly on the switch trial between the first and second blocks of testing (Table 2), which revealed no significant difference, p = .51. We then asked whether the number of infants who searched correctly on the switch trial was significantly different from chance (50%) on the first and second blocks of testing using binomial tests. These analyses indicated that the number of infants who searched correctly during the first block of testing (16 out of 21 infants) was significantly greater than chance, p = .027, Cohen’s g = 0.26. However, the number of infants who searched correctly was not significantly different from chance during the second block of testing, p = .38. A direct comparison of the first switch trial on this condition relative to the standard condition in Experiment 1 using a Mann–Whitney U test shows that more infants searched correctly on the consistent S-A-O rule location-change condition, U = 105.00, p = .001, f = 0.24. This indicates that differences in switch trial accuracy between the consistent S-A-O rule condition and standard versions of the task were not due to the experimenters acting as spatial cues to guide search behavior.

4.3 |. Discussion

These findings are consistent with Experiment 1 and show that performance on switch trials was not driven either from having multiple redundant cues to guide search behavior or due to the experimenter providing a spatial cue indicating where infants should search. Specifically, in the same-toy condition, both experimenters hid the same toy across trials. This condition served to rule out the possibility that improved performance in Experiment 1 was due to the experimenters hiding different toys, rather than due to infants using the identity of the experimenters to organize learning of S-A-O rules for action. We found that performance gains on switch trials persisted even when the experimenters hid the same toy across all trials. This result rules out the possibility that our findings in Experiment 1 were driven by the experimenters hiding different toys. As in Experiment 1, we also observed that infants had better switch trial performance in the second block of testing relative to the first block of testing. This block effect is consistent with a rule learning interpretation of performance, which would predict that infants should show better rule-guided behavior with additional practice observing how the experimenters contextualize S-A-O rules for action. Additionally, infants had better repeat trial accuracy and required significantly fewer trials to reach the search criterion in the first block relative to the second block of testing. While this effect was not found in Experiment 1, having a shared toy would add additional conflict in the S-A-O rules contextualized by the experimenters. This additional conflict may increase the initial difficulty when switching to a new S-A-O rule, leading to worse repeat trial accuracy and more trials to reach the search criterion during the second relative to first block of testing.

In the consistent S-A-O rule location-change condition, we examined whether infants were simply searching in the hiding well directly in front of the experimenter, rather than using the identity of the experimenter to organize S-A-O rules for action. To control for this possibility, the experimenters swapped seating locations on the critical switch trials, and the toy was hidden the same location by the same experimenter. If infants simply used the spatial locations of the experimenters to determine where to search, then we expected infants to perform poorly on the switch trials. In contrast, if infants use the identities of the experimenters to organize the event sequences into S-A-O rules for action, then we predicted that infants should search in the correct hiding location on the critical switch trials, even though the correct location is on the opposite side of the experimenter hiding the toy. Our results supported this latter prediction, which rules out the possibility that infants were simply searching in the hiding location directly in front of the experimenter on switch trials in Experiment 1.

The findings from these experiments provide further support for the hypothesis that infants can integrate across event sequences and use the identities of the experimenters conducting the task to contextualize S-A-O rules for action. Although our results support this interpretation of infants’ performance on this task, they are consistent with alternative explanations of infants’ A-not-B errors. For instance, prior work has indicated that increasing the saliency or perceptual differences between the two hiding wells leads to reductions in infants’ perseverative behavior (e.g., Clearfield, Dineva, Smith, Diedrich, Thelen, 2009; Bremner, 1978). Thus, it is possible that pairing a change in the hiding event sequence with a change in the identity of experimenter might increase the overall saliency of the switch trials, rather than acting as a contextual cue that can be used to help infants organize the event sequences into S-A-O rules that guide action.

To control for this alternative explanation of our findings, Experiment 3 tests a critical prediction of a rule learning interpretation using a within-subjects task design (Figure 6). We predicted that if infants attempt to use the identities of the experimenters to organize S-A-O rules for action, then switch trials that are consistent with the S-A-O rule cued by an experimenter should result in better performance than repeat trials that are inconsistent with the S-A-O rule cued by an experimenter, even though switch trials require inhibiting a prepotent motor response, whereas repeat trials do not. This condition allows us to more confidently assess whether infants are indeed using the identities of the experimenters to guide search behavior, rather than responding based on a prepotent motor response or other possible low-level perceptual factors.

5 |. EXPERIMENT 3

In our final experiment, we used a within-subjects design to test a critical prediction of this work. If infants use experimenters to organize S-A-O rules for action, then we should observe better performance on trials that are consistent with the S-A-O rule relative to trials that are inconsistent with the S-A-O rule, regardless of whether the trial involves a repeat or a switch in the motor response.

5.1 |. Method

5.1.1 |. Participants

The sample consisted of 14 nine-month-old infants (M = 9.32 months, SD = 0.98 months, 10 females, 4 males, 8 white non-Hispanic, 3 black, 1 Hispanic, 1 Asian, and 1 Mixed Race/Other). An additional four infants were tested, but excluded from the sample due to fussiness (n = 3) or experimenter error (n = 1). Sample size was determined based on a power analysis with a large effect size (d = 0.90; estimated from Experiment 1), which indicated that 14 infants would provide sufficient statistical power (80%) for the within-subjects design.

5.1.2 |. Procedure

The hiding procedure described in the General Method section was used, and the testing session consisted of four blocks: two consistent S-A-O rule blocks followed by two inconsistent S-A-O rule blocks (Figure 6). In the first consistent S-A-O rule block, experimenter 1 hid the toy in the A well two times (E1A/consistent repeat trials), after which experimenter 2 hid the same toy in the B well (E2B/consistent switch trial), consistent with the S-A-O rule associated with the experimenter. This block was then repeated, with experimenter 2 hiding the toy in the B well twice (E2B/consistent repeat trials), after which experimenter 1 hid the toy in the A well (E1A/consistent switch trial). After the two consistent S-A-O rule blocks, the first inconsistent S-A-O rule block occurred, during which experimenter 1 hid the toy in the A well two times (E1A/consistent repeat trials), and then experimenter 2 hid the same toy in the A well (E2A/inconsistent repeat trial), inconsistent with the S-A-O rule associated with the experimenter. This block was then repeated in the other hiding well, with experimenter 2 hiding the toy in well B twice (E2B/consistent repeat trials) after which experimenter 1 also hid the toy in the B well (E1B/inconsistent repeat trial), inconsistent with the S-A-O rule associated with the experimenter. If an infant did not search in the correct location on the second consistent repeat trial in any of the four testing blocks, it was repeated until the infant searched correctly. Also note that we used two consistent repeat trials rather than three to shorten the testing session, given that pilot testing indicated that three repeat trials made the experiment too long for the majority of infants to complete. However, meta-analyses indicate that the number of repeat trials does not significantly impact perseverative errors when it ranges from one to three trials (e.g., Wellman, Cross, Bartsch, & Harris, 1986).

5.1.3 |. Analysis plan

Our dependent measures were as follows: (a) consistent repeat trial accuracy, operationalized as the proportion of correct searches on consistent repeat trials; (b) consistent switch trial accuracy, operationalized as the proportion of correct searches on consistent switch trials; and (c) inconsistent repeat trial accuracy, operationalized as the proportion of correct searches on the inconsistent repeat trials. We expected to find no performance differences between consistent repeat trials and consistent switch trials. However, we expected to find worse performance on the inconsistent repeat trials relative to both the consistent switch trials and consistent repeat trials. Additionally, to examine performance differences by block, we also tested whether the number of infants who searched correctly on the first consistent switch or inconsistent repeat trial differed in the first relative to second block of testing.

5.2 |. Results

We examined whether there were differences in the proportion of infants’ reaching to the location where the toy was hidden across trials. We first verified that there were no multivariate outliers present using Mahalanobis Distances. We then used a repeated-measures ANOVA to compare infants’ average reaching accuracy by trial type (consistent switch trial, consistent repeat trial, and inconsistent repeat trial). Mauchly’s test indicated that the assumption of sphericity had been violated, χ² (2) = 7.13, p = .03. Therefore, degrees of freedom were corrected using Greenhouse–Geisser estimates of sphericity (ε = 0.691). Results indicated a significant difference in reaching accuracy by trial type, F(1.38, 17.96) = 8.88, p = .005, η_p² = 0.41 (Figure 7). We followed up on these analyses using pairedsamples two-tailed t tests (Bonferroni corrected alpha = 0.017). Paired differences were examined for normality and outliers prior to analyses using visual inspection, which indicated no outliers or deviations from normality. These analyses indicated that infants had better performance on consistent repeat trials relative to inconsistent repeat trials, t(13) = 4.04, p = .001, d = 1.06, and marginally better performance on the consistent switch trials relative to the inconsistent repeat trials, t(13) = 2.51, p = .026, d = 0.65. There was no difference in performance between the consistent repeat trials and consistent switch trials, t(13) = 1.26, p = .23. We next compared consistent switch trial, consistent repeat trial, and inconsistent repeat trial reaching accuracy to chance (50%) using three one-sample two-tailed t tests. These tests indicated that accuracy on the consistent switch trials was significantly better than chance, t(13) = 3.80, p = .002, d = 1.01, M = 82.1%, SD = 31.6%, as well as accuracy on the consistent repeat trials, t(13) = 9.46, p < .001, d = 2.52, M = 89.6%, SD = 15.7%. However, accuracy on the inconsistent switch trials did not differ from chance, t(13) = 0.37, p = .72, M = 53.6%, SD = 36.5%.

FIGURE 7.

Infants performed significantly better on the consistent repeat trials and consistent switch trials relative to the inconsistent repeat trials in Experiment 3. Group means are shown with individual data points overlaid. Error bars reflect SEM

Finally, we examined whether there were differences in the number of infants who searched correctly in the first relative to the second consistent switch trial, as well as in the first relative to the second inconsistent repeat trial. McNemar tests revealed that there were no significant differences in the number of infants who searched correctly on the first relative to the second consistent switch trial, p = .25, as well as in the first relative to the second inconsistent repeat trial, p = .73 (Table 3). These findings indicate that infants’ performance on the critical consistent switch and inconsistent repeat trials did not significantly change across testing blocks.

TABLE 3.

Number of infants (out of 14) who searched correctly on the first consistent switch and inconsistent repeat trials in Experiment 3

Trial type	Block 1	Block 2
Consistent switch (n = 14)	10	13
Inconsistent repeat (n = 14)	7	9

5.3 |. Discussion

In our third experiment, we tested a critical prediction of this work: if infants attempt to use the identities of the experimenters to organize S-A-O rules for action, then infants should perform worse on repeat trials that are inconsistent with the S-A-O rule cued by an experimenter, relative to trials that are consistent with the S-A-O rule cued by an experimenter. Our findings aligned with this prediction. We found that infants had better performance on consistent repeat and switch trials relative to inconsistent repeat trials. As such, these results support the hypothesis that infants may attempt to use the identities of the experimenters to guide search behavior, rather than responding based on a prepotent motor response or other low-level perceptual factors.

6 |. GENERAL DISCUSSION

Here, we asked whether infants can use the identities of experimenters to organize events into predictable S-A-O rules that can be used to guide action. We examined this question using a modified version of an A-not-B task, in which a change in the hiding event sequence was paired with a change in the experimenter hiding the toy. We tested 9-month-old infants, an age-group who are both capable of organizing simple audiovisual inputs into predictable rule structures (Werchan et al., 2015; Werchan et al., 2016; Lewkowicz et al., 2018) and who consistently show the A-not-B error (Diamond, 2002). We found that infants’ search behavior was consistent with using the identity of the experimenter to organize events into S-A-O rules that guide subsequent action (see Table 4 for a summary of manipulations and findings). Specifically, we found that when a switch in the hiding event sequence was paired with a change in the experimenter hiding the toy in the consistent S-A-O rule versions of the task, infants were more likely to reach to the new hiding location. However, when a change in the hiding event sequence was not paired with a change in the experimenter hiding the toy in standard versions of the task, infants tended to reach to the previously correct hiding location. These results show that infants’ reaching behavior across all conditions was consistent with using the identity of the experimenter to organize events into predictable S-A-O rules that guide action, leading to fewer perseverative errors when a change in the hiding event sequence was paired with a change in the experimenter hiding the toy. Follow-up experiments indicated that the lower level of perseverative errors observed when a different experimenter hid the toy on switch trials persisted even when the same toy was hidden across the repeat and switch trials. Follow-up results also indicated that the source of this improvement was not due to infants using the experimenter’s left–right seating location as a surface-level visual feature that cues which left–right hiding well to search for the toy in.

TABLE 4.

Summary of the factors manipulated and infants’ search behavior on the critical switch trials in Experiments 1 and 2

Condition	Experimenter identity	Experimenter location	Toy identity	Toy location	S-A-O rule violated?	Search aligned with S-A-O rule?
Standard	Same	Same	Same	Switch	Yes	Yes
Two-experimenter standard	Same	Same	Same	Switch	Yes	Yes
Consistent S-A-O rule	Different	Same	Different	Switch	No	Yes
Consistent S-A-O rule location change	Same	Switch	Same	Repeat	No	Yes
Consistent S-A-O rule same-toy	Different	Same	Same	Switch	No	Yes

These results support our hypothesis that infants may attempt to use the identity of the experimenter to organize event sequences into predictable S-A-O rules. However, it is also possible that a change in the experimenter may have increased the overall saliency of the switch trials relative to the repeat trials, a factor that has been shown to reduce infants’ perseverative errors on trials where the toy’s hiding location switches (Bremner, 1978; Clearfield et al., 2009). As such, we conducted a third experiment to more confidently assess whether infants were indeed using experimenters to organize the event sequences into predictable S-A-O rules that guide action. This third experiment was designed to test a critical prediction of a rule-based account of infants’ A-not-B errors: if infants attempt to use the identities of the experimenters to organize S-A-O rules that guide action, then switch trials that are consistent with the S-A-O rule cued by an experimenter should result in better performance than repeat trials that are inconsistent or violate the S-A-O rule cued by an experimenter. In line with this prediction, we found that infants performed better on switch trials that were consistent with the S-A-O rule associated with the experimenter, relative to repeat trials that were inconsistent with the S-A-O rule associated with the experimenter. This is even as the switch trials require infants to inhibit a prepotent motor response, whereas the repeat trials do not, a condition that would normally elicit a perseverative error at 9 months (Diamond, 2002). As such, the finding that infants made increased errors on repeat trials that violate the S-A-O rule associated with the experimenter supports the hypothesis that infants may attempt to use the identities of the experimenters to organize the event sequences into S-A-O rules that guide action.

Previous findings have shown that infants can learn to parse event sequences across visual, linguistic, and action domains using a variety of cues. For example, infants are capable of using statistical properties of event sequences to determine event boundaries (Stahl et al., 2014). Related findings indicate that infants can also use social information (Brugger et al., 2007) and temporal synchrony (Bahrick & Lickliter, 2000; Bahrick et al., 2015; Gogate & Bahrick, 1998) as cues to parse event sequences. Similarly, other work has shown that when 12-month-old infants observe an adult perform an ambiguous action within a larger sequence of actions, they are capable of using causal relations between actions to relate the ambiguous action to an overarching or higher-order goal (Woodward & Sommerville, 2000). This research indicates that infants can parse the event sequence into individual components, and then organize these components into a causal framework that allows infants to make predictions about the overarching goals of ambiguous actions. Our data expands on these findings by showing that infants are capable of using the identities of the experimenters conducting the task to organize event sequences into predictable S-A-O rules that guide action in the A-not-B task.

Our findings are also consistent with prior research examining how changes in environmental or perceptual information during the A-not-B task impact infants’ search behavior. For instance, altering the posture of the infant on B trials results in fewer A-not-B errors (Smith et al., 1999; Thelen, Schöner, Scheier, & Smith, 2001), as does using a different-colored hiding cover on the B trials relative to the A trials (Bremner, 1978). These changes in task dynamics or perceptual information on B trials may provide an alternative source of contextual information that might signal a novel S-A-O rule. Similarly, our results are also consistent with studies examining how social information might impact infants’ search behavior in the A-not-B task. For example, 10-month-old infants are more likely to make A-not-B errors when the experimenters use communicative cues during the task (Topal et al., 2008). The use of communicative cues may create a stronger bias for infants to use the identities of the experimenter to organize the event sequences into S-A-O rules used to guide action, thereby leading to increased perseverative errors on trials where the toy’s hiding location is inconsistent with the S-A-O rule cued by the experimenter. Similarly, infants are less likely to make A-not-B errors when a mechanical claw hides the toys (Boyer, Pan, & Bertenthal, 2011). However, if infants were first familiarized with the function of the mechanical claws, they made perseverative errors at rates similar to those observed in standard versions of the task. These data suggest that infants might interpret experimenters as performing goal-directed actions during the task, which might increase their bias to use the identity of the experimenters to organize event sequences into S-A-O rules that guide subsequent action, rather than responding based on a prepotent motor response alone. This idea is also supported by data showing that infants make A-not-B errors not only after trials of overt reaching to the A location but also after observing an experimenter retrieve the toy at the A location over multiple trials (Longo & Bertenthal, 2006).

While our intention was not to arbitrate among theories of the A-not-B error, this work does inform a broader body of research on such. Our findings also raise the possibility that infants’ perseverative errors on standard versions of the A-not-B task may, in part, reflect an adaptive response strategy as infants attempt to use the identity of the experimenter conducting the task to organize appropriate S-A-O rules used to guide action. When the experimenter does not change when the toy’s hiding location changes, infants align behavior with the S-A-O rule associated with the experimenter, thus searching in the previously correct location rather than in the new location that violates the S-A-O rule associated with the experimenter. In other words, infants’ reaching to the A well on B trials may reflect negative transfer of a learned S-A-O rule on standard versions of the task, until such time that a new rule is learned (Perkins & Salomon, 1992). This interpretation of infant performance on the A-not-B task is consistent with predictions from a recent model of PFC functional development (Werchan & Amso, 2017), which suggests that the infant PFC is adapted to support learning, such as through the structuring of inputs into predictable rules or sequences used to guide learning and action. There are two important benefits to this type of environmental structuring (Collins et al., 2014; Collins & Frank, 2013; Donoso et al., 2014). First, it helps prevent catastrophic interference so that new learning does not interfere with prior learning in different contexts. Second, prior rules can be transferred to novel contexts without having to learn them anew. Thus, organizing inputs into predictable rules helps narrow the space of learning problems by allowing infants to set up expectations in anticipation of future events, as well as by helping prevent new learning from interfering with prior learning. This type of environmental structuring may be particularly beneficial during early life when infants are faced with learning incredible amounts of information from noisy environmental input, despite having relatively limited cognitive and physical resources to accomplish this feat. This suggests that what appears to be an error may reflect an otherwise adaptive mechanism that supports efficient learning and helps prevent catastrophic interference when new information conflicts with prior learning in different contexts. However, future work is needed to determine whether these S-A-O rules are abstract and support generalization, or whether they are concretely tied to the specific experimenter context in which they were learned in.

In sum, our findings suggest that infants can use the identities of experimenters conducting the A-not-B task to organize events into predictable S-A-O rules that guide action. Moreover, these results raise the possibility that the A-not-B error may be a byproduct of an adaptive system that helps infants learn stable and generalizable representations of their environment. Importantly, these findings also have broader impact on how we have typically studied the PFC and infant behavior more generally. We tend to impose our own constructs and cognitive biases on infants, and conduct studies that are inherently designed to look for failures and limitations (Karmiloff-Smith, 1995). Yet, to have a fuller understanding of the developing mind and brain, our findings illuminate the importance of considering infants as different organisms with unique goals for learning and behavior.