Examining the role of external language support and children’s own language use in spatial development

Abstract

This research investigated whether an experimental manipulation providing children with external language support reflects developmental processes whereby children come to use language within spatial tasks. One hundred and twenty-one 3- to 6-year-old children participated in language production and spatial recall tasks. The Production task measured children’s task-relevant descriptions of spatial relations on the testing array. The Recall task assessed children’s delayed search for hidden object locations on the testing array relative to one or more spatial reference frames (egocentric, room-centered, and intrinsic). During the Recall task, the experimenter provided children with either descriptive or non-descriptive verbal cues. Results showed that children’s task-relevant language production improved with age and the effects of language support on spatial performance decreased with age. However, children’s production of task-relevant language did not account for effects of language support. Instead, children benefited from language support, irrespective of their task-relevant language production. These results suggest that verbal encoding is not a spontaneous process that young children use in support of their spatial performance. Additionally, experimental manipulations of language support are not fully reflective of the ways in which children come to use language within spatial tasks.

Classic questions in developmental psychology center on what children can accomplish independently versus with external support (Vygotsky, 1978). Experimental studies often manipulate external factors, such as experimenter instructions or task contexts, to test whether these factors facilitate performance. The processes thought to underlie the effects of experimental studies are at times interpreted as the same processes children come to use without external support (e.g., Dessalegn & Landau, 2008; Richland et al., 2006). However, there is a need to more thoroughly investigate factors underlying experimental effects and evaluate whether such factors map onto mature performance without external support. We investigate this issue within spatial development, specifically focusing on the role of language.

Spatial skills involve perceiving and remembering locations, navigation, perspective taking, and mental rotation (Hegarty & Waller, 2005; Vasilyeva & Lourenco, 2010). Individual differences in spatial skills are predictive of achievement in math and science (Verdine et al., 2017; Wai et al., 2009). There are large developmental changes in spatial skills during the preschool and elementary school years (Vasilyeva & Lourenco, 2010; Verdine et al., 2017). Given the importance of spatial skills, it is critical to investigate mechanisms involved in developmental change. Language is one potential mechanism. Language effects are found across tasks, including those assessing spatial reference frame selection (Miller et al., 2016; Miller, Kirkorian, et al., 2020), reorientation (Hermer-Vazquez et al., 2001; Shusterman et al., 2011), feature-binding (Dessalegn & Landau, 2008, 2013), analogical reasoning (Loewenstein & Gentner, 2005; Miller, Andrews, et al., 2020; Pruden et al., 2011; Simms & Gentner, 2019), and mental rotation (Casasola et al., 2020; Miller, Andrews, et al., 2020; Pruden et al., 2011). Some theorists proposed a verbal encoding account in which, as children acquire task-relevant language, they use such language within spatial tasks (Dessalegn & Landau, 2008; Hermer-Vazquez et al., 2001; Loewenstein & Gentner, 2005; Shusterman & Spelke, 2005). When children are provided with language support, such as by the experimenter during the spatial task, they can use such language to facilitate their performance (Dessalegn & Landau, 2008; Loewenstein & Gentner, 2005; Miller, Andrews, et al., 2020). However, it is unknown whether proficiency in using task-relevant language actually leads children independently to adopt verbal encoding strategies within spatial tasks. To address this question, we investigated 3- to 6-year-olds’ spatial performance with external language support versus on their own to further understand the role of verbal encoding.

Experimental studies demonstrate that external language support before or during spatial tasks facilitates children’s spatial performance (e.g., Dessalegn & Landau, 2008; Loewenstein & Gentner, 2005; Miller et al., 2016; Shusterman & Spelke, 2005). For example, in a spatial reference frame recall task, 4-year-olds needed to use landmarks to locate hidden objects in an array after the child moved to a new location or the array rotated (Miller et al., 2016). Providing 4-year-olds with task-relevant spatial language (e.g., “I hid the toy by the frog”) facilitated their performance relative to a control group (i.e., “I hid the toy here”). Similar effects of language supports have been found with other spatial words (e.g., “left/right”, “top/bottom”, “between”, Dessalegn & Landau, 2008; Loewenstein & Gentner, 2005; Miller, Kirkorian, et al., 2020) and with non-spatial but task-relevant language (i.e., “the yellow is prettier”, “the red wall can help you”, Dessalegn & Landau, 2013; Shusterman et al., 2011).

External language support enhances spatial performance, but does it reflect the mechanisms whereby children come to use verbal encoding independently? Many studies have focused on only one age group and cannot directly speak to developmental mechanisms (e.g., Farran & O’Leary, 2016; Miller et al., 2016; Shusterman et al., 2011). The studies that focused on multiple age groups showed that the effects of language support differ based on age and task difficulty (Dessalegn & Landau, 2013; Loewenstein & Gentner, 2005). For example, Loewenstein and Gentner found that, on a relational mapping task with no perceptual distractors, 3- but not 4-year-olds benefited from language support, but with perceptual distractors, 4-year-olds showed benefits. Relatedly, Dessalegn and Landau found age effects on a feature-binding task whereby children needed to recall the relative position of one colored side of a square. Four-year-old children benefited from hearing relational language, but 6-year-old children performed highly with or without the language support. Together, these results suggest that spatial language helps children perform spatial tasks at earlier ages.

If the age effects of external language support reflect developmental processes whereby verbal encoding comes to influence spatial performance, then once children acquire and can effectively use language, the effects of language support should be reduced. In addition to their experimental manipulation described above, Dessalegn and Landau (2013) included a post-test vocabulary task assessing children’s productive spatial vocabulary for the terms associated with the feature-binding task. They found that 6-year-olds were at ceiling for all terms (“left/right”, “above/below”), but 4-year-olds were only at ceiling for some terms (“above/below”). The researchers endorsed a verbal encoding account, theorizing that 6-year-old children’s high performance in both the spatial and vocabulary tasks reflected their automatic language use. Four-year-old children benefitted temporarily from language support but did not use it during the spatial task. This interpretation also aligns with research finding that children’s spatial language production correlates with their spatial performance (Hermer-Vazquez et al., 2001; Pruden et al., 2011; Simms & Gentner, 2019).

Whereas some past research provides support for verbal encoding, other research focusing on children’s task-relevant language production, both spatial and non-spatial, within spatial tasks suggests that verbal encoding may not be a primary mechanism (Miller, Andrews, et al., 2020; Miller & Simmering, 2018). For example, Miller, Andrews et al. had 4- to 5-year-olds participate in two spatial tasks and then explain their solutions to each spatial problem. The strongest predictor of spatial performance was not whether children produced spatial words, but whether they represented task-relevant spatial and non-spatial information, either verbally through speech or non-verbally through gesture. Using speech to represent task-relevant information was not more predictive of spatial performance than non-verbally using gestures. Miller and colleagues (Miller, Andrews, et al., 2020; Miller & Simmering, 2018) proposed that children need to attend to and encode task-relevant information, but such encoding is not dependent on using language within the task. According to this perspective, external language support is effective in directing children’s attention to task-relevant spatial information (Miller, Kirkorian, et al., 2020). However, it would predict that age-related differences in the effectiveness of such support are not driven by changes in children’s use of verbal encoding.

The contrast in theoretical perspectives and results related to the role of verbal encoding in the development of children’s spatial skills may result from different methodological approaches and unanswered questions. Both theoretical perspectives investigated the effect of children’s language production on their spatial performance; however, they assessed language production using different methods. Research aligning with verbal encoding accounts typically assessed language production using vocabulary tasks and did not assess production within the spatial tasks (e.g., Hermer-Vazquez et al., 2001; Simms & Gentner, 2019). In contrast, research theorizing that verbal encoding is not a primary mechanism assessed both spatial and non-spatial language within the context of the spatial task (Miller, Andrews, et al., 2020; Miller et al., 2017). Assessing language use within the spatial task is important because young children may produce spatial words in vocabulary tasks but not apply such language on their own within more complex tasks (Hund et al., 2017). Relatedly assessing both spatial and non-spatial task-relevant language production is also important. This is because non-spatial cues, similarly to spatial cues, can be relevant for spatial tasks (e.g., color cues in the feature-binding task). However, non-spatial cues may not always differentiate relevant spatial information and children do not always use the most relevant terms within the spatial tasks (Miller et al., 2017; Plumert & Nichols-Whitehead, 2007). Thus, to understand children’s potential for using verbal encoding effectively, it is critical to assess children’s production of both task-relevant spatial and non-spatial language within the spatial task.

An additional difference in methodological approaches is that only research endorsing a verbal encoding account has investigated age-related differences in the effectiveness of external language support on children’s spatial performance. Research questioning the verbal encoding account has investigated individual differences within a more constricted age range. However, a limitation with research analyzing age-related differences is that such research focused on children’s spatial vocabulary and not did not investigate task-relevant language production within the spatial task, nor did it evaluate how such effects of language support vary over a continuous age range. These differences in methodological approaches leave unanswered questions as to whether the reduced effects of language support associated with age result from improvements in children’s use of verbal encoding.

Current Study

In the current study, we investigated the role of language in children’s spatial development by assessing the relative role of external language support and children’s task-relevant language production on their spatial performance. We extended past research by testing the effects of external language support over a continuous age range between 3- to 6-years of age, by measuring children’s task-relevant language production within the spatial task, by extending the age range for testing the effects of language support, and by assessing whether the age-related effects of language support generalize to performance in a different spatial task.

We used a spatial recall task designed to assess reference frame selection (Miller et al., 2016; Miller, Kirkorian, et al., 2020; Nardini et al., 2006). Multiple types of reference frames can be used to encode locations, including egocentric (relative to oneself), room-centered (relative to a large space), and intrinsic (relative to layout of nearby objects) (Levinson, 2003; Simons & Wang, 1998). Using a spatial reference frame task to evaluate the role of language was ideal for several reasons. First, spatial reference frame selection during recall improves between 3- and 6-years, with reliable selection of an intrinsic reference frame, which is most relevant for this task but the latest to develop, emerging around 5- to 6-years (Nardini et al., 2006). This is the same age range in which children’s spatial word production develops substantially (Kuperman et al., 2012). Second, developmental changes in language use may be a mechanism driving developmental change. External language support facilitates 4- and 5-year-olds’ selection among spatial reference frames, including intrinsic reference frames (Miller et al., 2016; Miller, Kirkorian, et al., 2020). Additionally, 4-year-old children’s use of relational terms within the spatial reference frame task was associated with their intrinsic reference frame selection (Miller et al., 2016).

We used the same procedures as Miller et al. (2016). Children participated in the Production task followed by the Recall task, using the same spatial array. The array was a short rotating table with three red and two yellow cups used as hiding locations and four landmarks (Figure 1). Using this layout provided children with multiple different cues for describing and encoding task-relevant spatial information, which included cup color, landmark, and relational cues (e.g., next to, behind, across from). Using this layout reduced the dependency on children producing and encoding complex spatial terms (e.g., “left/right” in the feature-binding task, Dessalegn & Landau, 2013) that typically emerge later in development (Rigal, 1994; Shusterman & Li, 2016).

Figure 1.

Layout of the testing array with numbers only used here as reference to specific hiding locations (not displayed during the experiment). This picture is taken from the viewpoint of the child during the hiding events. The table was 1.0 m in diameter and 0.4 m in height. It was centered within the testing space and was approximately 1 m from each side of the curtained wall. The cups were 9 cm in diameter and 11 cm tall. The plates were approximately 16 cm in diameter. Distances between hiding locations are described from the center of one cup to the center of the other cup (note that equivalent distances are reported together: Location 1–2 and 1–5, 33 cm; Location 1–3 and 1–4, 40 cm; Location 2–3 and 4–5, 25 cm; Location 2–4 and 3–5, 48 cm; Location 2–5, 62 cm; Location 3–4, 26 cm). Distances between hiding locations and the nearest stuffed animal landmark are described from the center of the cup to the center of the stuffed animal’s head and were approximately 20 cm each. Four X-marks (20 cm per side) were centered along each curtained wall of the testing space with gaps of approximately 55 cm to the curtain and 25 cm to the edge of the table.

In the Production task, children were asked to describe the location of a toy hidden under one of the cups. We assessed whether children’s speech disambiguated the exact hiding location (similar to Miller, Andrews, et al., 2020). In the Recall task, children searched for an object hidden under one of the cups following a short delay and misalignment of reference frames. Similar to Nardini et al. (2006), across all trials children could use an intrinsic reference frame. However, the alignment of the egocentric and room-centered reference frames was manipulated across trials by having the child move (manipulating the egocentric reference frame) and/or array rotate (manipulating room-centered reference frame) between the hiding and search events (Figure 2). In the Recall task, children were assigned to either the Language or Control Condition. The conditions differed based on the cues that the experimenter provided while hiding the toy. In the Language Condition, children heard descriptive cues specifying the hidden object’s location relative to one or two nearby landmarks (e.g., “by the frog”) and in the Control Condition children heard non-descriptive cues (“here”).

Figure 2.

Display of Four Rotation Types, specifying whether a reference frame is aligned (+) or misaligned (−). Diagram not drawn to scale.

The Production task matched the Recall task’s hiding event, enabling comparison between children’s task-relevant language production (as an index of verbal encoding) and their spatial performance. The central difference between tasks was that in the Production task children were asked to describe locations and in the Recall task children were asked to remember locations. Both tasks also had similar goals. In the Production task, to accurately disambiguate locations required that children describe the hiding location relative to other objects (e.g., “the yellow cup next to the dog”, “the cup between the dog and the cow”, Figure 1). In the Recall task, to accurately identify the hidden object’s location across trials, children needed to remember the hidden object’s position relative to other objects on the table (i.e., when the table rotated the only cue children had to identify the hidden location was the location’s relative position to other cups and the landmarks on the array).

In using this paradigm, we asked three central research questions related to the role of language in young children’s spatial development. First, are there age-related differences in children’s production of task-relevant terms within the spatial task? Prior research found effects of age on children’s production of spatial terms (e.g., Dessalegn & Landau, 2013), but we know of no research that has assessed differences by age on children’s use of task-relevant spatial and non-spatial language within the spatial task. If children’s production of task-relevant terms improves with age, that would suggest that children have increased potential in using verbal encoding over time.

Second, are there age-related differences in the effectiveness of external language support on children’s spatial reference frame selection during recall? This question is both a conceptual replication of and extension of past research. It tests for a conceptual replication of age differences in children’s spatial reference frame selection during recall, specifically for trials misaligning spatial reference frames (Nardini et al., 2006). It also tests for a conceptual replication of age differences in the effectiveness of external language support on spatial performance in general (Dessalegn & Landau, 2013; Loewenstein & Gentner, 2005). It extends past research by assessing whether these age-dependent effects of external language support generalize to performance in a large-scale spatial task assessing children’s spatial reference frame during recall, as prior research investigating this effect assessed relational reasoning (Loewenstein & Gentner, 2005) and feature-binding (Dessalegn & Landau, 2013), using smaller scale tasks.

Third, do effects of language support on children’s spatial performance, expected to be associated with age, relate to age-related differences in children’s task-relevant language production? This third question differentiates between hypotheses on the role of verbal encoding in children’s spatial development. Specifically, if verbal encoding underlies age-related differences in children’s spatial performance (Dessalegn & Landau, 2013; Loewenstein & Gentner, 2005), then higher use of task-relevant language within the spatial task should result in smaller effects of external language support. Alternatively, external language support may be one way to facilitate spatial performance (Miller, Kirkorian, et al., 2020), but is not reflective of the mechanisms by which children spontaneously use language in support of their spatial performance (Miller, Andrews, et al., 2020; Miller & Simmering, 2018). In this case, we should see independent effects of external language support and children’s task-relevant language production. That is, external language support should predict children’s spatial performance even in cases when children produce task-relevant terms.

Method

Participants

Participants were 121 3- to 6-year-old children (M age = 4.55, SD = .87, range = 3.33–6.56, females = 57). There were 62 participants in the Control Condition (M age = 4.55, SD = .88) and 59 in the Language Conditions (M age = 4.55, SD = .86). An additional 25 children participated but were excluded due to caregiver interference in the Production task (13, described below), incomplete data (3), non-compliance (5), not reaching Recall task criterion (3, described below), and experimenter error (2). Four participants were excluded from analyses of the Production task only due to not talking. Data was collected in a university-affiliated research center in the Midwestern United States. Participants were primarily from White-middle class backgrounds (individual demographic data was not collected). Caregivers completed informed consent according to the University’s Institutional Review Board. Children received a small prize for participation. Data from the 4-year-olds (39 participants out of the sample of 121) was published previously (Miller et al., 2016, Experiment 1). The authors investigated the effects of language manipulations that occurred during and prior to the spatial task as well as a visual cue manipulation on children’s spatial reference frame selection. The data from Experiment 1 are included here to address new research questions with new analyses and to provide continuity between the 3- and 5- to 6-year-old sample (see Appendix A for further details).

Apparatus

Figure 1 displays the testing space layout and dimensions. The testing array was a rotating table centered in a square-shaped testing space. The testing space was partitioned by curtains that hung from ceiling to floor. The testing array was a short rotating table centered in the middle of the space. On top of the table were 5 cups (3 red and 2 yellow) arranged in a pentagon shape and 4 stuffed-animal landmarks (dog, cow, pig, and frog), in between or behind each cup. There were 4 duct tape X-marks on the floor: 2 indicated standing positions during the hiding and/or search events and the other two provided symmetry. The hiding object was a small toy. A ceiling-mounted camera recorded the sessions.

Design and Procedure

Children completed a Production task followed by a Recall task. One experimenter (E1) ran the protocol while a second (E2) recorded responses and rotated the table.

Production task.

The Production task measured children’s descriptions of the hidden object’s locations within the array. Children typically participated with their caregiver, although E2 completed this role on rare occasions. Before the task, E1 instructed the caregiver not to talk to the child about the task or materials, specifically instructing them to avoid mentioning the cups or the stuffed animal landmarks during the session. E1 highlighted the importance of control in the study and explained that we are interested in children’s spontaneous responses. Caregivers were told that they could prompt the child up to two times to give more clues (e.g., by saying “can you tell me more?”) when the child gave a vague response (e.g., “it is under a cup”). However, caregivers were told to not give directive prompts that would help the child in describing the toy’s location (e.g., not asking “what color is the cup?” or “what animal is it by?”). However, in some cases, a caregiver did not follow our instructions and the children were excluded from analyses of the Production task (see Participants section) due to the possible influence on children’s task-relevant language production (Hund et al., 2017) and their spatial performance (Loewenstein & Gentner, 2005; Miller et al., 2016). We had caregivers participate to help children feel comfortable talking and to give children motivation to provide precise descriptions.

Before the task began, E1 provided the child with the following instructions (with the appropriate term substituted for “mom” if it was not the mother who accompanied the child). “In this game your mom will turn around so that she cannot see the table. I will hide a toy in one of these cups and then we will have your mom turn back around. Then you will use your words and help your mom find the toy. In this game, you want to help your mom so she can find the toy.” During each trial, the caregiver turned away from the table and E1 hid the toy. Afterwards, the caregiver turned around and the child was instructed to “use your words and tell your mom where the toy is hidden.” E1 discouraged pointing by instructing children to put their hands behind their back. E1 helped the caregiver with prompting when needed to facilitate control. There were 5 trials, one per cup, presented in random order.

Recall task.

The Recall task examined the effects of language support on recall relative to spatial reference frames. Children were randomly assigned to the Language or Control Conditions. In the Language Condition, children were given descriptive cues on each trial specifying the hidden object’s location relative to nearby landmark(s) and then children repeated the cue (e.g., “I am hiding the toy by the pig”). In the Control Condition, children heard non-descriptive cues (“I am hiding the toy here”). See Table 1 for all cues. We had children repeat the descriptive cue in the Language condition to maximize the likelihood that children would encode spatial cues into language. Additionally, we did not have children repeat the cue in the Control condition as it was not meaningful to say the words “here” and may encourage children to adopt other strategies, such as gesture, which can influence spatial performance (Ehrlich et al., 2006; Miller, Andrews, et al., 2020). Prior research has not found differences in spatial performance when children repeat an experimenter’s language cue versus not (Farran & O’Leary, 2016).

Table 1.

Experimenter’s Cues during the Recall task

Condition	Cup Number	Experimenter’s Cue (“Look I am hiding the toy...”
Control	All	“here”
	1	“by the dog and cow”
	2	“by the cow”
Language	3	“by the pig”
	4	“by the frog”
	5	“by the dog”

There were four rotation types (Neither-, Child-, Both-, Table-move). The rotation types differed based on whether the child moved and/or table rotated 90° during the delay. The egocentric reference frame was misaligned when the child had a different viewpoint relative to the array (Child- and Table-Move trial types). The room-centered reference frame was misaligned when the array changed positions within the room (Both- and Table-move trial types). Across trials, the intrinsic reference frame was available as the positions of cups and landmarks on the array remained stable. Figure 2 displays a visualization of the procedures and specifies the reference frame alignment for each Rotation trial type.

Before the task began, children chose if their caregiver stayed with them in the testing space. When a caregiver stayed, the experimenter ensured that the caregiver stood in a corner behind the child to avoid being within the child’s view of the table. On trials when the child changed positions, the experimenters and caregiver both moved with the child to maintain the same relative positions, ensuring that they could not serve as landmarks.

At the start of the task, E1 provided instructions as follows: “In this game, I am going to hide the toy in one of these cups and then you will turn around, and we’ll count to 10 together; when we say 10, you will turn back around and find the toy.” The task began with a practice trial in which the experimenter hid the toy in cup 1, the cup closest to the child (Figure 1) while giving the descriptive or non-descriptive cue, depending on condition assignment. Then, E1 instructed the child to turn around and then E1 and the child counted to 10 aloud together. Afterward, the child turned back and searched for the toy. Children were encouraged to find the toy on their first try but could search until finding it. This trial was repeated if the child searched incorrectly on their first try.

Before each trial, E1 explained the directions, specifying whether the child moved and/or the table rotated during the delay. On all trials during the delay, children faced away from the table, facing the curtained walls. On Neither-move trials, the procedure was the same as the practice trial (Figure 2A). On Child-move trials, E1 guided the child to the second X-mark, 90° to the right around the table (Figure 2B). This trial type misaligned the egocentric reference frame, although children could use dead reckoning to identify their body’s position relative to the array (Newcombe et al., 1998). On Both-move trials, E1 rotated the table 90° counterclockwise and E2 helped the child walk 90° to their right (the same direction as in the Child-move trials, Figure 2C). The egocentric reference frame was available on this trial type; however, children needed to use an intrinsic reference frame to recognize that they were in the same position relative to the array, after they moved, and the table rotated. Thus, this trial type depends more on the intrinsic reference frame than the Child-move, and thus tends to be more difficult for children (Nardini et al., 2006). On Table-move trials, E2 rotated the table 90° degrees clockwise during the delay (the opposite direction of the Both-move trials, Figure 2D). This Rotation type depends most on the intrinsic reference frame as both the egocentric and room-centered frames were misaligned. On Both- and Table-move trials, E1 reminded the child that the table turned before allowing them to search. After search, E1 rotated the table to its original position in the child’s view and said, “Look, I am turning the table back.”

Children completed trials in one of two orders: Neither-, Table-, Child-, Both-move or Both-, Child-, Table-, Neither-move. This order differentiated levels of difficulty as found in Nardini et al. (2006). There were 12 trials, with 3 consecutive trials to each rotation type. The hiding locations were pseudo-randomized with the constraint that per rotation type there was a trial to: (1) cup 1, (2) cup 2 or cup 5, and (3) cup 3 or cup 4.

Scoring and Reliability.

For the Production task, children’s responses were transcribed during the experimental session by E2. Different research assistants later checked transcripts from video after the session and coded the transcripts for accurate mention of Color, Landmark, and Relational terms as well as Disambiguation. Children received one point per trial for each category; no extra credit was given for multiple mentions within the same category (e.g., mentioning two landmarks, “near the dog and cow”). Inaccurate descriptions were excluded from the analyses (21 trials, 3%). Children received a point for Disambiguation when their speech described the exact hiding location. Table 2 displays examples. We gave credit for both precise (e.g., “yellow cup by the cow” for cup 2) and imprecise but pragmatically differentiated locations (“by the cow” for cup 2 as it is the only cup with only the cow by it). We gave credit for pragmatically differentiated locations because we figured that representing locations that way would allow children to find the exact location. However, we also analyzed the Disambiguation score with giving credit for precise disambiguation only and found the same pattern of results. Data from 26 participants (21%) were double coded by a second research assistant; coders had 98% agreement. For the Recall task, E2 recorded search responses and indicated accuracy. Participants were excluded if they searched incorrectly on all Neither-move trials as this indicated that they did not understand the task or were not trying (see Participant section).

Table 2.

Example of Responses during Production Task and Disambiguation Coding

Cup	Disambiguated	Not Disambiguated
1	“Under a red cup. It’s close to two yellow cups next to a cow and a dog.”	“One of the red cups.”
	“It’s not under any yellow cups. And It’s not under the two red cups.”	“It’s by the cow.”
2	“In a yellow cup, It’s by the cow.”	“Behind the red cup, yellow cup.”
	“In the cow spot”	“Under the cup.”
3	“Piggy, red cup.”	“Under a red cup next to a yellow cup.”
	“In front of the piggy.”	“It’s under a red cup, by the pig and the frog.”
4	“It’s by the frog, not the yellow cup.”	“Somewhere on the side.”
	“Really near the frog.”	“It is under one of the red cups.”
5	“Behind the dog.”	“A blue plate. A yellow cup.”
	“It’s by the puppy, under a yellow cup.”	“On the side.”

Results

We extended research on the role of language in children’s spatial development by assessing effects of external language support and children’s own language production on their spatial reference frame selection during recall. We asked three primary questions: 1) are there age-related differences in children’s use of task-relevant language to describe spatial relations on the testing array as an indicator of their potential for verbal encoding; 2) are there age-related differences in the effectiveness of external language support on children’s reference frame selection during recall; and 3) does children’s task-relevant language production account for the effects of external language support on children’s spatial recall. In addressing these questions, we first report on performance in the Production task using descriptive and correlational analyses. We then analyze performance in the Recall task and then relations between performance in the Production and Recall task together. Analyses investigating the Production task included the smaller sample of participants who spoke (N = 117: 58 in Language and 59 in Control; see Participants section) and analyses investigating the Recall task only included the full sample (N = 121). Before testing our research questions, we conducted preliminary analyses to assess for effects of Gender. We found no significant gender differences in children’s Disambiguation (t_{115 =} 0.56, p = .289, d = .104) nor Recall task performance (t_{119 =} 0.44, p = .328, d = .080). We thus excluded gender from further analyses.

Production Task

For the Production task results, we first provide descriptions of the types of terms that children produced and how children disambiguated locations. Then to evaluate our first question on the relation between Age and children’s task-relevant term production, we report correlational analyses. Note that we have not separated results by Condition for these Production task analyses because this task occurred first in the session and was identical between Conditions. Table 3 displays the proportion of trials, for all trials as well as separately for trials when the utterance did versus did not disambiguate the locations, on which children: a) produced color, landmark, and relational terms; and b) produced specific combinations of terms. As shown in Table 3A, across all trials (first column), children produced color terms most frequently followed by landmark and relational terms. On average, children disambiguated locations on 42% of trials. In utterances that disambiguated locations (second column), children frequently produced landmark and/or relational terms. Production of color terms was similar between trials on which children did versus did not disambiguate the location (cf. top line of second and third columns).

Table 3.

Proportion of Trials on which Children Produced Each Type of Term

Types of Terms	All Trials (548 trials)	Disambiguation (229 trials)	No Disambiguation (319 trials)
A. Total Production
Color	.52 (287)	.50(115)	.54(172)
Landmark	.44 (241)	.94(215)	.08(26)
Relational	.45 (249)	.84(192)	.18(57)
B. Specific Combination of Terms
No Cues	.22 (123)	01 (2)	.38 (121)
Color Only	.26 (140)	00 (1)	.44 (139)
Landmark Only	.04 (24)	.10 (22)	01 (2)
Relational Only	.03 (14)	.02 (4)	.03 (10)
Color & Landmark	.02 (12)	.05 (12)	.00 (0)
Color & Relational	.05 (30)	.03 (7)	.07 (23)
Landmark & Relational	.18 (100)	.38 (86)	.04 (14)
Color, Landmark & Relational	.19 (105)	.41 (95)	.03 (10)

Notes. The decimals represent proportions of the column total with the number of trials in parentheses. Part A shows the proportion and number of trials on which children produced a color, landmark, or relational term. The proportions within these columns do not sum to 1 as children could produce multiple terms on a given trial. Part B shows the proportion and number of trials on which children produced each possible combination of terms; these rows sum to 1 in each column.

In Table 3B, we further detail children’s production to specify the combination of terms produced. When children disambiguated locations (second column), they most commonly produced a combination of landmark and relational terms, or a combination of color, landmark, and relational terms (79% together; bottom two rows). However, when children did not disambiguate locations (third column), they tended to produce no terms or only provided color terms (82% together; top two rows). In addition, we examined the specific types of relational terms that children produced. Most relational terms (77% of the 249 trials that included relational terms), described adjacent locations (e.g., “by”, “next to”), the same term type as the experimenter’s descriptive cues (i.e., “by”). Overall, these results suggest that children varied widely in the types of terms produced and the extent to which they disambiguated locations, with landmark and relational terms being most indicative of disambiguation.

To evaluate our first question, which was on whether there were age-related difference in children’s production of task-relevant terms, we conducted Pearson’s correlations between Age and each Production task measure. Age significantly correlated with each measure: color terms, r(115) = .40, p<.001; landmark terms, r(115) = .24, p = .009; relational terms, r(115) = .44, p<.001; and disambiguation, r(115) = .25, p = .006. Figure 3 plots the distributions underlying these correlations (see Appendix B for further statistics by Age Group and correlations with Condition). Overall, these results show the expected age-related differences in children’s production of task-relevant terms, suggesting that with increased age, children have greater potential for effectively using verbal encoding.

Figure 3.

Relations between Age and each Production task measure (Color terms, Landmark terms, Relational terms, and Disambiguation). Error bars reflect +/− 1 SE of the regression line. Individual data points from the sample of participants who spoke during this task (n = 117) are jittered for improved visibility. The data plotted between 4 and 5 years of age reflect participants from Miller et al. (2016).

Recall Task

Next we investigated our second question, which was whether there were age-related differences in the effectiveness of external language support on children’s selection of spatial reference frames during recall. To address our research question, in Model 1, we conducted an ANOVA with a continuous interaction term using the aov function in R (R Core Team, 2017), with predictor variables of Age (continuous, mean-centered), Rotation Type (Neither-, Child-, Both-, Table-move; within-subjects), and Condition (Language, Control; between-subjects), and interactions between all the variables. Results are displayed in Table 4. Consistent with past research, we found significant effects of Age, Rotation Type, and Condition, which were subsumed by Age × Rotation Type, Condition × Rotation Type, and Age × Condition interactions. No other effects were significant.

Table 4.

Model 1 ANOVA Results for effects of Age, Condition, and Rotation Type on Proportion Correct in Recall Task

	df	F	P	η2p
Between-subject effects
Age	1	47.42	< .001	.243
Condition	1	25.32	< .001	.146
Age × Condition	1	4.41	.038	.029
Error	117
Within-subject effects
Rotation Type	3	41.14	< .001	.260
Age × Rotation Type	3	4.65	.003	.038
Condition × Rotation Type	3	3.88	.009	.032
Age × Condition × Rotation Type	3	0.04	.988	.000
Error	351

Notes. Includes full sample (n = 121).

To understand the Age × Rotation Type interaction, we examined correlations with Age for each Rotation Type (shown in Table 5, All Conditions column). Age was significantly correlated in all Rotation Types, with strongest effects on trials misaligning reference frames (Child-, Both-, and Table-move). These findings conceptually replicate past research showing that there are age-related changes in children’s selection among spatial reference frames between 3- and 6-years of age (Nardini et al., 2006).

Table 5.

Correlations between Age, Disambiguation, and Recall Performance by Condition and Rotation Type

	All Conditions		Language		Control
Variable	Age	Disambiguation	Age	Disambiguation	Age	Disambiguation
Recall task total	34***	.19*	.26*	.16	44***	.23
Neither	.22*	.00	.07	−.03	.35*	.05
Child	45***	.18	.38**	.16	.55***	.22
Both	.40***	.32***	.31*	.17	.50***	.35**
Table	37***	.30**	.27*	.30**	.53***	.37**

Notes.

^*< .05

^**< .01

^***< .001. Includes the full sample (n = 121, Language n = 59, Control n = 63)

To investigate the Condition × Rotation Type interaction, we conducted two-tailed independent samples t-tests on Condition effects for each Rotation type (Table 6 displays means and standard deviations; see Appendix B for these means separated by Age Group). Proportion correct was higher in the Language than in the Control Condition for all Rotation types misaligning reference frames: Child-move (t_{119 =} 2.94, p = .004, d = .304), Both-move (t_{119 =} 2.80, p = .006, d = .515), and Table-move trial types (t_{119 =} 4.53, p < .001, d = .828), but not on the Neither-move trial type (t_{119 =} 1.61, p = .110, d = .304). These findings conceptually replicate past research and show that external language support facilitates children’s spatial reference frame selection (Miller et al., 2016; Miller, Kirkorian, et al., 2020).

Table 6.

Proportion Correct in Recall Task by Condition and Rotation Type

Measures	All Trials	Language	Control
All Trials	.62(.25)	.71(.23)	.54(.23)
Neither-Move	.82(.24)	.86(.22)	.79(.24)
Child-Move	.63(.36)	.72(.33)	.54(.36)
Both-Move	.55(.34)	.63(.32)	.47(.34)
Table-Move	.48(.38)	.63(.36)	.34(.34)

Notes. Includes full sample (n = 121)

To follow-up on the Age × Condition interaction, we examined correlations with Age on proportion correct for each Condition (Table 5, Recall Task total row). The correlation between Age and Recall task proportion correct was stronger in the Control than Language Condition. As seen in Figure 4, with increased Age, the difference in proportion correct between the Language and Control Conditions was lower, that is, the benefit of external language when using spatial reference frames was smaller with increased age, similar to prior to research using other spatial tasks (Dessalegn & Landau, 2013; Loewenstein & Gentner, 2005). Overall, results from the recall task show that external language supports children’s spatial reference frame selection during recall, especially on the trials that rely most on the intrinsic reference frame, and that differences across both Rotation Types and conditions were smaller with increased age.

Figure 4.

Proportion Correct by Age, Condition, and Rotation Type. The regression line is predicted from Model 1. Error bars are plotted +/− 1 standard error for point estimates from the model. Individual data points from the full sample (N = 121) are jittered to improve visibility. The data plotted between 4 and 5 years of age reflect participants from Miller et al. (2016).

Production and Recall Task

Lastly, for our third research question, we investigated whether and how children’s task-relevant language production related to the effects of external language support on children’s spatial performance. As described in the introduction, these analyses aimed to differentiate whether the effects of external language support on children’s recall performance would be lower in children who produced more task-relevant language or if external language support and children’s task-relevant language production would both uniquely predict children’s recall performance. Correlational statistics between children’s recall performance in each Condition and Disambiguation are displayed in Table 5. To investigate our question, we used the same ANOVA as in Table 4 with the addition of Disambiguation (continuous, mean-centered), which was calculated as the proportion of trials on which children disambiguated locations, see Table 3 middle column for details). Results are displayed in Table 7 and Figure 5. We found the same patterns among factors as in the previous model, except the Age × Condition interaction was marginal (though the effect size increased slightly; cf. Table 7). There was also a significant effect of Disambiguation and a significant Disambiguation × Rotation Type interaction. To follow-up on the Disambiguation × Rotation Type interaction, we compared correlations between Disambiguation and proportion correct in each Rotation Type (Table 5, All Conditions column). Disambiguation was significantly correlated in the Both- and Table-move trials, trials dependent most heavily on the intrinsic reference frame. No other effects were significant. Considering that the effect of Condition held when adding in Disambiguation, our results show that external language support and children’s task-relevant language production each uniquely contributed to children’s spatial performance. That is, children’s task-relevant language production did not significantly account for the effects of external language on children’s spatial performance. We also found a similar pattern as Model 2 when using Relational term production instead of Disambiguation, with the exception that the main effect of Relational term production failed to reach significance (see Appendix B, Table B2).

Table 7.

Model 2 ANOVA Results for Effects of Age, Condition, Disambiguation, and Rotation Type on Proportion Correct in Recall Task

	df	F	P	η2p
Between-subject effects
Age	1	46.19	< .001	.298
Condition	1	22.04	< .001	.168
Disambiguation	1	4.85	.030	.043
Age × Condition	1	3.43	.067	.030
Age × Disambiguation	1	1.63	.205	.015
Condition × Disambiguation	1	0.01	.929	.000
Age × Condition × Disambiguation	1	0.33	.565	.003
Error	109
Within-subject effects
Rotation Type	3	37.87	< .000	.258
Age × Rotation Type	3	4.34	.005	.038
Condition × Rotation Type	3	3.18	.024	.028
Disambiguation × Rotation Type	3	3.03	.030	.027
Age × Condition × Rotation Type	3	0.01	.999	.000
Age × Disambiguation × Rotation Type	3	1.86	.136	.017
Condition × Disambiguation × Rotation Type	3	0.25	.859	.002
Age × Condition × Disambiguation × Rotation Type	3	0.43	.731	.004
Error	327

Notes. Includes smaller sample who spoke during Recall task (n = 117).

Figure 5.

Proportion Correct by Rotation Type, Condition, and Disambiguation score. The regression line is predicted from Model 2, while controlling for Age. Error bars are plotted +/− 1 standard error for point estimates from model. Individual data points from the sample of participants who spoke during the Production task (N = 117) are jittered to improve visibility.

Of importance to our research question, the interaction between Condition and Disambiguation was not significant. As this finding is central to our research question, we chose to further explore this effect by separating participants based on whether they disambiguated on most trials (3 or more; points at or above .60 on the x-axis in Figure 5) and ran Model 1 again separately for the High and Low Disambiguation groups (n = 52 and 65, respectively). The goal of this analysis was to examine whether, despite the non-significant interaction, participants with high disambiguation benefited less from external language support than participants with low disambiguation. If so, this would suggest that children who can disambiguate in the production task do not need external language support for recall because they can use their own task-relevant language to support their recall performance. Alternatively, if both groups show similar benefits from external language, that would suggest that children who can disambiguate in the production task do not use their own language to support their spatial performance to the same extent as that of the external language. For participants in the High Disambiguation group, we found an effect of Condition (F_1,48 = 5.71, p = .021, η²p = .106), such that participants had higher recall performance in the Language (M = .74, SD = .31) than in the Control condition (M = .63, SD = .33). For participants in the Low Disambiguation group, we also found an effect of Condition (F_1,61 = 15.63, p < .001, η²p = .204), such that participants also performed higher in the Language (M = .68, SD = .33) than in the Control condition (M = .47, SD = .37). Overall, these results showed that children benefited from external language support even when they produced task-relevant language on most trials (i.e., effect of Condition in the High Disambiguation group), suggesting that they were not using their potential abilities to verbally encode as effectively as the experimenter’s language.

Discussion

The current research provides insights into the role that language plays in age-related differences in children’s spatial skill development by uncovering whether external language support affects children’s spatial performance in similar ways to how children’s spatial performance improves with age in the absence of such support. When testing for effects of children’s task-relevant language production and external language support separately, our findings were consistent with past research. We conceptually replicated several findings. First, we found that children’s Age correlated with their production of Relational terms (Figure 3 lower left panel), consistent with prior research showing that older children have higher spatial word production than younger children (Dessalegn & Landau, 2013; Kuperman et al., 2012). Second, we found that children in the Language Condition had higher recall compared to those in the Control Condition, specifically on Child-, Both-, and Table-move trials (Figure 4). This finding is consistent with research showing that external language support facilitated children’s spatial performance (Dessalegn & Landau, 2008; Loewenstein & Gentner, 2005; Shusterman & Spelke, 2005) on trials misaligning reference frames in recall (Miller et al., 2016; Miller, Kirkorian, et al., 2020). Third, we found effects of Age and interactions between Age and Condition on children’s recall performance (Figure 4). These finding are consistent with research showing age-related improvements in children’s reference frame selection during recall between 3 and 6 years (Nardini et al., 2006); and older children’s spatial performance benefitted from external language less than younger children’s (Dessalegn & Landau, 2013; Loewenstein & Gentner, 2005).

In addition, our finding investigating children’s task-relevant language production and their spatial performance separately extended research in showing age-related differences in children’s production of task-relevant terms within the context of the spatial task (Figure 3), and that reduced benefit of external language with age (Figure 4) extended to large-scale spatial tasks assessing spatial reference frame selection in recall. Overall, these findings show that, with age, children are more likely to have the language knowledge necessary for verbal encoding and less likely to benefit from external language support in their spatial recall performance.”

However, when investigating individual differences and our condition manipulation together, we found that task-relevant language production was not enough to account for age-related differences in the benefit of external language on children’s spatial performance. Specifically, we found additive effects such that both children’s task-relevant language production and external language independently predicted children’s selection among reference frames (Figure 5). Children who produced more task-relevant terms performed higher on the spatial task. However, irrespective of their task-relevant language production, children receiving external language support performed higher than those without. These patterns occurred when considering task-relevant language either in terms of disambiguation, which could include non-spatial language, or relational (spatial) term production.

Overall, these results identified limitations in the verbal encoding hypothesis, which posits that once children produce task-relevant language, they use such language within spatial tasks to support their performance (Dessalegn & Landau, 2013; Hermer-Vazquez et al., 2001; Pruden et al., 2011). Many children produced task-relevant language that was semantically similar to the external language support provided by the experimenter. Many even used the same relational terms. Yet, the external language support facilitated children’s spatial performance beyond their potential to effectively use verbal encoding. This work leaves open questions as to what children’s task-relevant language production affords their spatial performance if it does not lead them to effectively use verbal encoding, what other processes are involved leading to age-related improvements in spatial performance, and what processes change over development that contribute to children eventually using verbal encoding effectively.

Factors influencing early spatial development

As noted above, age-related increases in children’s task-relevant language production has been reliably shown to relate to spatial performance, but the theoretical interpretations of these relations have differed. We theorize that such production was not reflective of verbal encoding during the spatial task but instead reflective of how children attended to and represented task-relevant information. Miller and Simmering (2018) showed that 4-year-old’s task-relevant language production within a spatial scene description task correlated with their non-verbal attention in a synonymous memory task; and attention, not language, most strongly predicted spatial performance. Likewise, spatial performance can be supported without language by making spatial features salient or directing attention non-verbally (Learmonth et al., 2001, 2002; Miller, Kirkorian, et al., 2020). According to our theory, children who produced task-relevant language during the Production task encoded relevant features and used language. However, during the spatial task, without external language support, those children attended to task-relevant information but did not effectively use language. This led to weaker performance relative to children who both used more task-relevant language and who received language support.

In addition to changes in children’s production of task-relevant terms, which we theorize is indicative of their attention, what other factors are changing over development contributing to children’s reference frame selection and spatial performance more generally? Age was a strong predictor of spatial performance in all models, even beyond children’s task-relevant language production and external language support. We theorize that the effects of age may reflect changes in basic memory processes, executive function, and/or perspective taking. The current research was focused on the memory process of encoding, but developmental changes in maintenance and retrieval likely also contributed to developmental change. For instance, in using an intrinsic reference frame, the reference frame available on all trials, children need to encode and maintain the relative position of the hiding location relative to other landmarks on the array and then at retrieval select the intrinsic reference frame among both the egocentric and allocentric reference frames. Past research showed that 4-year-olds’ reference frame selection improved by facilitating maintenance through hiding the toy in the same location across all trials and by facilitating retrieval through having children view themselves and/or the array rotate (Miller & Simmering, 2016). In addition to developmental changes in general memory processes, Negen and Nardini (2015) suggested that developmental improvements in executive function contribute to children’s intrinsic reference frame selection. They proposed that children need to inhibit the egocentric and allocentric frames to select an intrinsic reference frame, and thus inhibitory control may facilitate this process. Lastly, perspective taking is improving between 4- and 8-years of age, where children begin to decrease egocentric responding and start to take on other perspectives (Frick et al., 2014). Children with better perspective-taking skills, may more flexibly retrieve information from room-centered and intrinsic reference frames because they can more readily adopt other perspectives. Basic memory processes, executive function, and perspective taking are not specific to the development of reference frame selection but have been found to contribute to spatial cognition more generally (Carr et al., 2018; Frick & Baumeler, 2017; Kozhevnikov et al., 2006).

Factors contributing to verbal encoding

Although we argue that our findings suggest that verbal encoding is not a central mechanism contributing to early developments in children’s spatial performance, we propose that verbal encoding is a strategy that children adopt for spatial tasks later in development. To effectively use verbal encoding, we propose that children need to realize the relevance of using language in service of spatial performance. In contrast to past research within spatial development emphasizing the spontaneous and implicit nature of verbal encoding (Dessalegn & Landau, 2013; Hermer-Vazquez et al., 2001), we advocate for verbal encoding as a strategy that children are explicitly aware of and apply when adaptive for a given task. In the current study, the communicative nature of our language manipulation enabled children to realize the relevance of language for facilitating their performance. Communication is ostensive and the act of communication conveys relevance (Wilson & Sperber, 2003). This is consistent with Shusterman et al. (2011) who found that using language to explicitly tell 4-year-olds that a spatial feature is relevant (“the red wall can help you”) facilitated the children’s spatial performance. However, we hypothesize that only later in development, past 6 years of age, do children regularly come to realize the relevance of language for performing spatial tasks and use verbal encoding more strategically in the absence of external support. This parallels findings within visuospatial working memory, showing improvements in verbal recoding strategies between 5- to 10- years of age (Elliott et al., 2021; Hitch et al., 1991; Pickering, 2001).

In understanding developmental changes in children’s strategic and explicit verbal encoding, we pull from decades of research on strategy use (Roebers, 2014 for review). This research found developmental increases during the elementary school years in deliberate and conscious strategies. Developments in strategy use arise, in part, from classroom instruction, such as the teaching of mnemonics, and changes in metacognition (Coffman et al., 2008; Crowley et al., 1997; Fabricius & Hagen, 1984; Krebs & Roebers, 2010). We theorize that children come to realize the relevance of language through increased experience using language, such as during scaffolded spatial activities (Hund et al., 2017; Plumert & Nichols-Whitehead, 1996) where they gain practice in using language to encode spatial information. Additionally, changes arise from developments in general strategy use, increasing children’s approaching task demands more deliberately and evaluating language as adaptive for the task.

Future Direction and Conclusions

In the future, it will be important to conduct longitudinal studies to investigate the role of language and non-verbal factors that contribute to spatial development. The current study was cross-sectional and cannot speak directly to causal mechanisms of developmental change. Past longitudinal studies investigated the role of language within spatial cognition but only tested children younger than 5-year-olds of age and focused primarily on exposure to spatial words and children’s spatial language (Casasola et al., 2020; Pruden et al., 2011). Future longitudinal studies should test children between 3- and 10- years of age and include, at multiple timepoints measures, of children’s spatial attention, spatial performance, executive function, mnemonic strategy use, as well as real-time measures of verbal and non-verbal spatial strategies. This will address unanswered questions related to the processes that children use in performing spatial tasks over development and the factors leading to such developmental change.

In conclusion, we showed that an experimental manipulation that facilitated children’s spatial performance by providing language support was not isomorphic with the ways that children’s own task-relevant language production related to their spatial performance. These findings also revealed limits that children’s production of task-relevant terms have in explaining developmental change. Future research should move beyond assessing whether children produce task-relevant language and identify additional mechanisms underlying verbal encoding development. Additionally, developmental researchers should be cautious in interpreting the age-related differences in the effects of experimental manipulations, as the manipulations may not parallel the processes involved in children’s spontaneous development.

Research Highlights.

• Tested language support and children’s relevant language use on spatial recall

• Older children’s spatial recall benefited less from external language support

• Relevant language use and language support each uniquely predicted spatial recall

• No evidence that children adopt verbal encoding once they produce relevant language

• Children benefit from language support even when producing task-relevant language

Appendix A

Analysis of Previously Published Data

To address the question of whether the inclusion of previously-published data from 4-year-olds could be driving the results reported here, we conducted additional analyses using a Bayesian repeated measures ANVOA in JASP (JASP Team, 2020) with a variable coded for whether the participant was included in the previous publication or not. We chose a Bayesian ANOVA for this analysis because it allows us to evaluate the evidence for versus against the inclusion of a variable when comparing models with and without that variable. In addition to this variable, we included the same predictors from Model 2 reported in text: Age and Disambiguation as continuous variables, Rotation as a four-level repeated measure, Condition as a between-subjects variables, and their interactions; we also added interactions between Previously-Published and the categorical predictors (Rotation, Condition).

To evaluate the evidence for whether a variable contributes explanatory power to the model, we used BF_incl across matched models in JASP (van den Bergh et al., 2020). This reflects whether models with the variable of interest account for the data better than corresponding models without that variable, when considering all other combinations of main effects and interactions. BF_incl values less than 1 are considered evidence against the variable’s inclusion, values between 1 and 3 are considered equivocal or weak evidence for inclusion of the variable, and values greater than 3 are considered substantial evidence for the variable’s inclusion (see Jarosz & Wiley, 2014, for discussion and further delineation of larger values). Table A1 shows the analysis of effects from the JASP output with the BF_incl for each variable. Of most relevance to the current question are the BF_incl values for the Previously-Published variable and related interactions: these values range from 0.035 to 0.236, providing evidence that models excluding these variables account for the data better than models including these variables.

Table A1.

Analysis of Effects from Bayesian Repeated Measures ANOVA

Effects	BF incl
Between-subjects effects
Age	543,851.117
Condition	1,345.677
Disambiguation	1.617
Previously-Published	0.177
Age × Condition	1.248
Age × Disambiguation	0.531
Condition × Disambiguation	0.288
Condition × Previously-Published	0.236
Age × Condition × Disambiguation	0.373
Within-subjects effects
Rotation	3.506 * 1017
Age × Rotation	1.191
Condition × Rotation	1.078
Disambiguation × Rotation	1.000
Rotation × Previously-Published	0.035
Condition × Rotation × Previously-Published	0.068
Age × Condition × Rotation	0.010
Condition × Disambiguation × Rotation	0.017
Age × Disambiguation × Rotation	0.253
Age × Condition × Disambiguation × Rotation	0.055

Note. Includes smaller sample who spoke during Recall task (n = 117). Text from JASP output (JASP Team, 2020): “Compares models that contain the effect to equivalent models stripped of the effect. Higher-order interactions are excluded. Analysis suggested by Sebastiaan Mathôt.”

Appendix B

Additional Statistics

In addition to the statistics reported in the main text, here we report additional statistics. In Table B1, we present descriptive statistics on performance in both Production and Recall Tasks separated by Age Groups and Condition. This table is included for ease of comparing the previously-published Age Group (4-year-olds) to those added in the current paper (3- and 5- to 6-year-olds).

In Figure B1, we present a figure of the correlation between Age and Disambiguation for each Condition. This is to facilitate visualization of the Age effects shown in Figure 3 of the main text. In Table B2, we include an additional regression model predicting Recall Task performance with Relational Terms as a variable instead of Disambiguation (Table B2; cf. Model 2 and Table 7 in the main text). The additional regression analysis is included for comparison with prior studies that analyzed the use of spatial relational terms without evaluating whether the utterances disambiguated the possible referents. As noted in the main text, the results in Table B2 qualitatively parallel those reported with Disambiguation as the measure from the Production Task, except that the main effect of Relation Term production is not significant.

Table B1.

Mean and Standard Deviation for each Production Task and Recall Task measure by Age Group

	Language				Control				All Condition
Measures	3yr	4yr	5–6yr	All Ages	3yr	4yr	5–6yr	All Ages	3yr	4yr	5–6yr	All Ages
	A. Production Task
Disambiguation	.26(.30)	.51(.38)	.47(.42)	.41(.39)	.27(.30)	.40(.43)	.56(.37)	.41(.39)	.26(.30)	.46(.41)	.52(.40)	.41(.39)
Color	.18(.26)	.80(.36)	.69(.42)	.56(.44)	.27(.29)	.56(.43)	.65(.35)	.49(.39)	.22(.28)	.68(.41)	.67(.38)	.52(.42)
Landmark	.28(.35)	.52(.41)	.46(.42)	.42(.41)	.27(.32)	.44(.46)	.65(.41)	.45(.43)	.27(.34)	.48(.43)	.56(.43)	.44(.42)
Relational	.26(.33)	.38(.35)	.61(.39)	.42(.39)	.22(.32)	.49(.45)	.71(.33)	.47(.42)	.24(.33)	.43(.40)	.66(.36)	.45(.40)
	B. Recall Task
All Trials	.60(.22)	.72(.22)	.80(.20)	.71(.23)	.36(.15)	.52(.19)	.73(.20)	.54(.23)	.47(.22)	.62(.23)	.76(.20)	.62(.25)
Neither-Move	.84(.26)	.85(.23)	.88(.20)	.86(.22)	.70(.31)	.77(.19)	.90(.15)	.79(.24)	.76(.29)	.81(.21)	.89(.17)	.82(.24)
Child-Move	.54(.37)	.77(.33)	.85(.23)	.72(.33)	.33(.37)	.53(.32)	.76(.24)	.54(.36)	.43(.38)	.65(.34)	.80(.24)	.63(.36)
Both-Move	.53(.28)	.62(.34)	.74(.33)	.63(.32)	.27(.24)	.48(.37)	.67(.30)	.47(.34)	.39(.29)	.55(.35)	.70(.31)	.55(.34)
Table-Move	.49(.37)	.66(.33)	.73(.36)	.63(.36)	.15(.20)	.30(.29)	.59(.36)	.34(.34)	.31(.34)	.48(.36)	.66(.36)	.48(.38)

Notes. Disambiguation was calculated as the number of trials on which the child’s speech fully differentiated the object’s location. Recall task measures includes full sample (n = 121) and the Disambiguation measure includes smaller sample who spoke during the Recall task (n = 117). Although we report all measures separately for the Language and Control condition, the Production task did not differ between conditions.

Figure B1.

Relations between Age and proportion Disambiguation by Condition. Error bars are plotted +/− 1 standard deviation from the regression line.

Table B2.

ANOVA Results for Effects of Age, Condition, Relational Terms, and Rotation on Proportion Correct in Recall Task

Model A	df	F	p	pη2
Between-subject effects
Age	1	45.58	<.000	.295
Condition	1	21.75	<.000	.166
Relational Terms	1	2.20	.141	.020
Age × Condition	1	3.57	.062	.032
Age × Relational Terms	1	0.68	.413	.006
Condition × Relational Terms	1	0.03	.854	.000
Age × Condition × Relational Terms	1	2.19	.142	.020
Error	109
Within-subject effects
Rotation	3	37.72	<.000	.257
Age × Rotation	3	4.32	.005	.038
Condition × Rotation	3	3.17	.025	.028
Relational Terms × Rotation	3	2.95	.033	.026
Age × Condition × Rotation	3	0.02	.998	.000
Age × Relational Terms × Rotation	3	1.50	.214	.014
Condition × Relational Terms × Rotation	3	0.61	.611	.006
Age × Condition × Relational Terms × Rotation	3	0.08	.970	.001
Error	327