Variability of input in preposition learning by preschoolers with developmental language disorder and typically-developing language

Abstract

The purpose of this study was to investigate the role of variability in teaching prepositions to preschoolers with typical development (TD) and developmental language disorder (DLD). Input variability during teaching can enhance learning, but is target dependent. We hypothesized that high variability of objects would improve preposition learning. We also examined other characteristics (e.g. vocabulary skills) of children who responded to treatment. We used a case series design, repeated across children (n = 18) to contrast how preschoolers learned prepositions in conditions that manipulated variability of objects and labels across three treatment sessions. We contrasted a high versus low variability condition for objects and labels for one group of typically-developing (TD) children (n = 6). In other groups (TD, n = 6; DLD, n = 6), we contrasted high versus low object variability only. Visual inspection and descriptive statistics were used to characterize gains. Half (n = 3) of TD participants showed a low variability advantage for the condition that combined object and label variability. In the condition that only contrasted object variability, the majority (n = 4) of TD participants showed a high variability advantage, compared to only two participants with DLD. In the high object variability condition, high receptive vocabulary scores were significantly correlated with high performance of learning prepositions (r_s = 0.71, p < 0.05). Combining high variability for objects and labels when teaching prepositions was not effective. However, high variability for objects can create a learning advantage for learning prepositions for children with typically developing language, but not all learners. Characteristics of different learners (e.g. receptive vocabulary scores) and language status (impaired or unimpaired) should be taken into consideration for future studies.

Keywords

children developmental language disorder input variability intervention prepositions preschool

I Introduction

Acquisition of prepositions is an important part of language development for young children. Prepositions include words such as up, over, and behind and are critical in making requests (‘Socks off, Mommy!’), following directions (‘Put your shoes on’), and locating objects (‘The one in the box’). Variability of input has been shown to be beneficial in language acquisition. However, the approach to providing learners with variable input differs depending upon the target (Gómez, 2002; Perry et al., 2010; Saffran, 2003; Suanda et al., 2014). Thus, the purpose of this study was to examine aspects of input variability that might increase the speed and robustness of preposition acquisition in typically-developing (TD) preschoolers and preschoolers with developmental language disorder (DLD; historically referred to as specific language impairment or SLI; Reilly et al., 2014), for whom the acquisition of prepositions is often a slow and tedious process (Leonard et al., 1997).

1 Typical development of prepositions

To know how to better teach prepositions to children with developmental language disorder (DLD), it is first necessary to know how they are typically learned. It is helpful to consider the problem space of the child acquiring a preposition. The preposition itself is a label with a mapping to the relation between the two objects (‘The chair is in the house’). Thus, in terms of cues for learning, a child has the potential to learn from the object cues (e.g. chairs generally have four legs and a back; houses generally have four walls, a roof, and a door; the preposition ‘in’ denotes containment in which the chair is surrounded by the house) and label cues (e.g. ‘chair’ and ‘house’ generally follow ‘a’ or ‘the’ or an adjective; the preposition ‘in’ links the two phrases of ‘the chair’ and ‘the house’). Thus, this raises the question: Which cues for learning – labels, objects, or both – could be targeted in treatment to enhance learning of prepositions?

Additionally, supplemental visual cues in the form of gestures that highlight label and referent mapping have been found to increase children’s novel word learning. Specifically, novel spoken words paired with referents that were accompanied by an adult gesture yielded higher comprehension scores (percentage of correctly matching referent to label) by young children with and without DLD than novel words that were not accompanied by gestures (Ellis Weismer and Hesketh, 1993). Thus, given the spatial nature of most prepositions, gestural cues given by adults may also play a role in a child’s learning of prepositions. (For a comprehensive review on the influence of gestural input by adults on language acquisition, see Ozcaliskan and Dimitrova, 2013.)

Typical development of prepositions follow a well-established trajectory according to categories of spatial relations they denote (Landau and Jackendoff, 1993). Prepositions that convey simple directional relations, up and down (Nelson, 1973) appear early in an English-learning child’s expressive vocabulary, and are often some of the first words of children who may have expressive inventories of fewer than ten words. These directional prepositions are followed by emergence of prepositions that denote contiguous relations, such as containment (in) and support (on) (Sinha et al., 1994). Later-developing prepositions include non-contiguous relations such as above and beside (Caselli et al., 1999). Although many times participants are preschool and kindergarten age, focus in the literature tends to be on refining their use of earlier developing prepositions of containment and support (e.g. Johannes et al., 2016; Landau et al., 2017; Nicholas, 2011), with less known about later-developing prepositions of non-contiguous relations. This lack of detail about young children’s initial acquisition of later-developing prepositions (e.g. above and beside) makes teaching approaches less informed.

2 Development of prepositions in preschoolers with developmental language disorder

Although the primary characteristic of preschool children with DLD is poor grammatical skills, this group also has documented deficits with vocabulary learning (e.g. Kan and Windsor, 2010) and weaker semantic representations (Alt et al., 2004, 2013; McGregor et al., 2002). Research on preposition use by preschool children with DLD is limited. There are more preposition errors in children with DLD’s utterances than is expected for their grammatical skill level. For example, in age-matched and language-matched comparison groups of typically-developing and language-impaired preschoolers, Watkins and Rice (1991) found that children with DLD, who all had low receptive vocabulary scores as part of their inclusion criteria, omitted prepositions more often than their language-matched younger counterparts omitted prepositions. The authors propose that children with DLD’s poorer performance when compared to their language-matched counterparts’ performance indicates that it is not due to stage in grammatical development, but likely weaker vocabulary development.

Grela et al. (2004) found that preschool-aged children with DLD both omitted and substituted prepositions, frequently using at, and less frequently using with and for. The authors concluded that the larger number of substitution errors relative to omission errors indicated that children with DLD understood the grammatical properties of the preposition (because they used a preposition in the correct syntactic frame) but did not fully understand the semantic properties to use the preposition in the correct semantic context. Other studies also discuss preposition deficits as difficulty with semantic representations. (Eisenbeiss et al., 2006; see also Puglisi et al., 2005). Overall, these findings suggest that children with DLD may have difficulty learning prepositions because of their weak formation of semantic representations.

3 The role of statistical learning

One way to promote rapid learning is through the use of statistical learning techniques. Statistical learning implies that statistically regular patterns, structures and co-occurrences found in language contribute to the learning process (Saffran, 2003). Thus, if input to the learner is structured in an optimal way using techniques like variability and complexity, learning is likely to happen implicitly (without trying) and quickly (e.g. Alt et al., 2012). It appears that different levels of variability and complexity are beneficial but, importantly, are target dependent. For example, non-adjacent dependent linguistic structures (e.g. he is running), may require more repetitions for learning than adjacent structures (e.g. the + cat), or noun classes may generalize better when learned in variable conditions whereas initial verb learning may require less variability (Gómez, 2002; Maguire et al., 2008; Perry et al., 2010).

Word learning studies show that object variability strengthens semantic representations, which in turn boosts word learning. For example, Perry et al., (2010) found that typically-developing 18-month-olds had improved learning of trained nouns, better generalization, and improved overall word learning when the features of objects, particularly shape, in the training set varied. Furthermore, the toddlers also generalized to learning novel label-object mappings as shown by substantial increases in their vocabulary. The authors concluded that the toddlers’ experience with learning to extract the relevant features to form a semantic representation of the trained nouns likely influenced how they proceeded in extracting patterns for semantic representations in other novel label-object mappings in their environment, making them better word-learners.

One way to strengthen semantic representations that support learning of prepositions may be to increase the exemplar variability of a teaching set in therapy. Use of variability of input to help children with language delay learn vocabulary have already shown promise (Alt et al., 2014). Alt et al. showed that target words said by a clinician in a variety of syntactic and situational contexts (i.e. with different objects) were more often acquired by the two-year-old late-talking children in the study when compared to control words. Like Perry et al. (2010), they also found that children’s overall vocabularies increased more quickly than would be expected by a maturational account. Although this study had a small number of participants, variability in input, both lexical and situational context, successfully facilitated word learning.

Variability can also be used to help children learn grammatical rules, which represent syntactic relations. Making use of an artificial language with non-adjacent dependencies comparable to the English correlate present progressive (e.g. a X b, is running), Gómez (2002) found that 18-month-olds with typically-developing language only learned when supplied with high variability (24 exemplars) in the X element. It appeared the surrounding linguistic variability allowed learners to circumvent rote memorization strategies, and learn the rule.

Recently, Plante et al. (2014) expanded this concept to natural language learning with preschool children with DLD. They found that target morphemes (e.g. past tense) presented with 24 unique verbs were produced correctly more often than morphemes modeled the same number of times, but with only 12 unique verbs. This shows that increased variability in input of verbs, and not mere repetition, successfully facilitates grammatical morpheme learning in children with DLD.

There is one caveat regarding the role of variability. Although high variability has been shown to be beneficial in most situations (like those reported above), there are scenarios in which high variability is actually a hindrance to learning. For example, Maguire et al. (2008) found that a significantly larger proportion of two- and three-year-old children who saw a single exemplar of a novel verb label learned it better compared to those in the multiple exemplar condition. Thus, even though high variability can create a learning advantage, it may not be helpful for some types of targets, particularly those that convey relations.

4 The current study

The current study aims to examine conditions that facilitate preposition learning. Levels of variability can be applied to two important components that surround prepositions: (1) the objects conveying spatial relations denoted by the target preposition and (2) the labels of those objects. If preposition learning is facilitated by object variability, we predict that target prepositions taught using a high variability of objects will be learned better than those taught with a low variability of objects.

Although variability has been helpful for language learning in children with DLD, there is the possibility that a context that facilitates learning for children with typical development may not be intense enough or provide enough exemplars to be beneficial for children with DLD. Thus, we need to investigate how children with different learning capabilities respond to these intervention approaches. Thus, intensity of the variability presented will be dialed up by providing various labels of objects (high variability of labels) and dialed down by using the same, generic labels to refer to those objects, ‘this’ and ‘that’ (low variability of labels).

Because this is the first study to our knowledge that tests the hypothesis that object variability affects preposition learning, an additional purpose of the present study is to initially evaluate the clinical feasibility of such treatment. Although this is a small study based on individual data, its contribution is not without merit to clinical practice (for discussion of case studies as informative to practitioners with regards to treatment effectiveness, see Vance and Clegg, 2012). In accordance with the feasibility stage of intervention as described by Fey and Finestack (2009), the present study aims to verify whether the hypothesis appears to yield the predicted outcome and to determine which outcome measures are the most useful from a clinical perspective. This study also aims to describe characteristics of children who respond (or do not respond) to treatment.

II Methods

1 Participants

Sixty-four children ages 3;0–5;11 with and without developmental language disorder were recruited to be potentially placed in one of three intervention groups: Group 1, which included typically developing children in a high variability for label and object input condition; Group 2, which included typically developing children in a high variability for object input condition; and Group 3, which included children with developmental language disorder in a high variability for object input condition. To be included, children needed to (1) be monolingual English-speaking, (2) have normal hearing, (3) have normal non-verbal cognitive skills and (4) during pre-treatment probes, demonstrate comprehension at least once of both in and on and have comprehension scores of zero for above and beside. This last criterion – knowing in and on but not knowing above and beside – was the ability that the three groups were consistently matched on. Eighteen monolingual English-speaking children were included in the treatment protocol in one of the three intervention groups (see Table 1). Additional descriptive measures were obtained.

Table 1.

Details of groups.

Participant number	SPELT-P2^a	PLS-4^b	K-ABC-II^c	Age (years; months)	Sex	Maternal education (years)	PPVT-IV^d
Group 1: Typically developing; high low variability for object and label:
1	108	117	119	3;3	male	16	102
2	101	109	104	3;6	female	12	97
3	129	118	99	3;3	female	17	119
4	112	124	109	3;2	female	11	119
5	93	114	99	3;0	female	12	105
6	106	111	95	3;4	female	13	105
Group 2: Typically developing; high low variability for object:
7	98	n/a	100	4;1	female	16	121
8	105	n/a	89	4;4	female	16	88
9	108	NC	104	3;1	female	12	111
10	131	126	104	3;3	male	17	103
11	89	111	89	3;6	female	14	78
12	102	114	85	3;3	female	12	102
Group 3: DLD; high low variability for object:
13	87	n/a	102	4;6	female	12	100
14	78	n/a	93	4;8	male	14	93
15	45	n/a	97	5;5	male	11	87
16	68	n/a	89	5;1	male	15	102
17	53	n/a	77	5;8	male	14	91
18	78	n/a	97	3;8	male	17	111

Notes. DLD = Developmental Language Disorder. NC = Not completed. ^a Standard scores from the Structured Photographic Expressive Language Test: Preschool: 2nd edition (Dawson et al., 2005). ^b Standard scores from Preschool Language Scales: 4th edition (Zimmerman et al., 2002). ^c Standard scores from nonverbal index of the Kaufman Assessment Battery for Children: 2nd edition (Kaufman and Kaufman, 2004). ^d Standard scores from Peabody Picture Vocabulary Test: 4th edition (Dunn and Dunn, 2007).

Monolingual English-speaking status was obtained through parental report. Hearing status was determined by passing a hearing screen. Cognitive status was determined by standard scores above 75 on the non-verbal assessment Kaufman Assessment Battery for Children: 2nd edition (KABC-II; Kaufman and Kaufman, 2004). From the children initially recruited the following were disqualified: nine due to bilingualism, one due to hearing issues, and one due to low non-verbal cognitive skills. Of the remaining 53 who underwent pre-treatment probes to test for baseline comprehension of prepositions, four children were disqualified after not demonstrating knowledge of on and 37 were disqualified due to already knowing above, beside, or both. All of the children demonstrated knowledge of in. Thus, of 49 children who knew in and on, 41% (n = 20) knew above and 63% (n = 31) knew beside. This suggests that the remaining 12 typically-developing children who qualified for the study may be on the lower end of typical for preposition acquisition.

Developmental language disorder status determined in this study is meant to be the same as the historical definition of specific language impairment, in which language disorder is primary, or not secondary to other developmental disorders (Bishop et al., 2016). Acceptable sensitivity and specificity scores per age group drove our use of particular tests to determine DLD status. Specifically, for three-year-old children, DLD status was determined by standard scores lower than 85 on the Preschool Language Scales: 4th edition (PLS-4; Zimmerman et al., 2002). which yields acceptable sensitivity (83.0%) and specificity (88.0%) for this age group. For four- and five-year-old children, DLD status was determined by standard scores lower than 87 on the Structured Photographic Expressive Language Test: 2nd edition (SPELT-P2; Dawson et al., 2005). which yields acceptable sensitivity (90.6%) and specificity (100%) for this age group (Greenslade et al., 2009). Although there are no known sensitivity and specificity cut-offs for three-year-olds for the SPELT-P2, this test was also given to the three-year-olds in the study using the same criteria for four- and five-year-olds for the purpose of using uniform scores among all ages for correlations. Twelve children (ages 3;0 to 4;4) were determined to have typically-developing language and six children (ages 3;8 to 5;8) were determined to have developmental language disorder. Receptive vocabulary scores were obtained using the Peabody Picture Vocabulary Test: 4th edition (PPVT-4; Dunn and Dunn, 2007) as a descriptive measure. Maternal education level, an indicator of socioeconomic status that is predictive of language ability (Dollaghan et al., 1999) was also obtained through parent report. It should also be noted that there were sex differences across groups, with the group of children with typical development consisting mostly of girls (ten out of twelve) and the group of children with DLD consisting mostly of boys (five out of six).

2 Stimuli

a Prepositions

The non-contiguous spatial relations conveying above and beside were selected to be taught because 9.6% and 6.9%, respectively, of typically developing children between the ages of 1;6 to 2;6 produce these prepositions (Caselli et al., 1999). It was unknown prior to this study what percentage of typically-developing three- and four-year-olds produce them, but the low percentage of children using them at 2;6 suggested there would be room to grow.

b Objects

Selection of the objects was prioritized primarily for their ability to represent the prepositional relations, to be amenable to variability, and to have two labels. Following Perry et al. (2010), we created sets of items that had high or low object variability. In the high variability object condition, objects varied in terms of color, texture, and size, along with different pairings of the objects. For example, a high variability set for beside included a small checker placed in close proximity to a larger basket and a long straw placed in close proximity to a small vase. In the low variability object condition, objects varied in terms of color and texture but were consistent in size; for examples of high object variability, see Figure 1.

Figure 1.

Example of diversity of objects for high object variability.

c Labels

In the high variability label condition we used 24 unique noun pairs in order to mirror the effect of high variability for form found in Plante et al. (2014). Objects were referred to by their names, using synonyms for the same exemplar (e.g. ‘pail’ and ‘bucket’). In contrast, for the low variability condition, all objects were referred to as ‘this’ or ‘that.’

3 Design, procedures and data analysis

Three intervention groups were created. The group of children with typical-development were divided into two groups: (1) high/low variability for object and label; (2) high/low variability for object. All the children with DLD were in the high/low variability for object condition. Utilizing a within-participant, case series design, every child was taught both prepositions. We counterbalanced which preposition was taught in each condition across participants.

Pre-treatment procedures included hearing screenings, assessments for language, cognition, and receptive vocabulary, and preposition probes. These occurred across three to four sessions by pairs of trained undergraduate and graduate research clinicians. See the assessment and treatment schedule in Table 2. The graduate research clinicians were enrolled in a clinical certification program and advised by a certified speech language pathologist. All clinicians underwent fidelity checks for protocol administration.

Table 2.

Assessment and treatment schedule.

Day	1	2	3	4	5	6	7
Assessment	Hearing screening;	PPVT-4				K-ABC-II	PLS-IV
	SPELT-P2						(< 4;0)
Probe	Pre-Tx		Tx 1	Tx 2	Tx 3
Treatment		1	2	3

Notes. Tx = Treatment. SPELT-P2 = Structured Photographic Expressive Language Test: Preschool: 2nd edition (Dawson et al., 2005); PPVT-4 = Peabody Picture Vocabulary Test: 4th edition (Dunn and Dunn, 2007); K-ABC-II = Kaufmann Assessment Battery for Children: 2nd edition (Kaufman and Kaufman, 2004); PLS-IV = Preschool Language Scale: 4th edition (Zimmerman et al., 2002).

In order to assess knowledge of the prepositions, we used pre-treatment probes for each target preposition (above and beside) along with foil prepositions (in and on). We assessed comprehension by trying to elicit both verbal and nonverbal productions. Specifically, children could demonstrate their understanding by either saying the name of the preposition, or by correctly demonstrating the preposition using objects. For the probe sessions, we always used the same two objects, a comb and a coozie (insulated beverage holder), and tested for prepositions in pseudorandom order. To probe verbal production, one researcher sat across from the child while the other researcher sat next to the child while covering her eyes. The first researcher then asked the child while the other researcher’s eyes were covered, ‘See this? Where is it? Tell her.’ while shaking the comb. Eye-covering of a second clinician was meant to provide a pragmatically appropriate context to minimize responses using deictics. The clinician was also sure to sit next to the child so that her point of view was the same to ensure clarity of the intended response. If a child responded with ‘there’ or pointed to the object, he was gently reminded that the second clinician could not see and was probed once more. To probe nonverbal production, the clinician handed the child one object and placed the other object on the table, giving instructions for changing the placement of the object such as ‘take this and put it beside this’ or ‘take this and put it above this.’ When the clinician referred to the second object, she was careful to not point to the object but to lift it so as not to inadvertently or incorrectly cue the child with a spatial relation. Children were offered stickers when the probe session was completed.

Treatment was administered across three 20 minute sessions (see Table 2). Twelve object pairings showing spatial relations were provided for one preposition, above, then twelve for the other preposition, beside, in the same session. The combination of type of object in each of the twelve pairings were different for each session and pseudorandomized across sessions. The order of which session each combination was presented in was counterbalanced across children in each group. In other words, objects were carefully cycled through combinations to counteract potentially unforeseen problematic items or item combinations. Prepositions were also counterbalanced for order across sessions.

In line with the child’s point of view, each demonstration of an object pairing conveying a spatial relation was accompanied by one clinician saying a preposition in phrases. For high label variability, the first clinician said ‘The ladle is above/beside the pail’ and for low label variability the first clinician said ‘This is beside/above this.’ The clinician placed each object as she referred to them. Then, the second clinician pointed and said a phrase with synonyms ‘Oh, the spoon is above/beside the bucket’ for high label variability or ‘Oh, this is beside/above this!’ for low label variability. The terms ‘this’ and ‘that’ were used for the low variability condition for Group 1 (e.g. compared variability for objects and labels) and were used for all examples for Groups 2 and 3. No additional support or feedback (e.g. recasting or correction) was provided to the child. Any response of the child was met with a cheerful ‘oh’ from the clinicians.

Post treatment, comprehension of each target and foil preposition was probed through elicited verbal and non-verbal production probes as described above. Ten probes (using the exact set of materials that was trained the prior day) were for items used in the training sets in the high variability condition (three for above, three for beside, two for in and two for on) and a separate ten probes were for items in the low variability condition (three for above, three for beside, two for in and two for on). Thus, there were 6 available data points per condition (high or low) per session. The inclusion of the foils (in and on) decreased the chance that children would have a 50% chance of choosing the correct preposition in the verbal production modality. For the nonverbal production component, there was an infinite number of potential locations in which to place the object (and indeed, we saw quite a few interesting choices) in order to demonstrate understanding of a preposition. Stickers were offered at the end of each probe session.

To analyse data for this within-participant case series design, each probe type was graphed for each child, plotting treatment probe session and number of correct probes for the high and low variability conditions. Using traditional single-case methodology (e.g. Horner and Spaulding, 2010), a visual analysis of gains was completed to determine if there was a learning advantage in either condition. We looked for accuracy in terms of initial rate of acquisition and total overall performance. Correlations were calculated to determine associations between standardized test scores and accuracy on preposition tasks.

III Results

To rule out confounding variability contributed by the specific prepositions, t-tests were calculated for learning scores for above versus beside. Results across all children (n = 18) indicated that neither above nor beside was learned better: t = 0.97, df = 16, p = 0.34, so they were collapsed for subsequent analyses. Only nonverbal production data are reported because verbal production data yielded very few responses other than deictics (e.g. ‘Right here!’). The children’s nonverbal responses were interpreted as correct or incorrect, with correct defined as a position reasonably corresponding to the targeted preposition and incorrect defined as not being a position reasonably corresponding to the targeted preposition. Thus, incorrect positions could have corresponded to multiple prepositions (e.g. in, on, under, in front of, beside), and not just the alternative preposition of the pair taught meaning that correct and incorrect responses were not subject to 50/50 chance, as there was one correct answer, and an infinite number of incorrect answers. Children’s responses were scored independently by a different research assistant for 11% of the data sample and yielded an inter-rater reliability of 97.2%.

1 Gains examined as rate of accuracy and total accuracy

For Figures 2 –4, data illustrate within-child preposition learning of above and beside. The blue line indicates the high variability for objects and labels and the red line indicates the low variability for objects and labels. Recall that the highest score a child could achieve for a single condition within an individual probe session was six.

Figure 2.

Intervention Group 1: Prepositions correct for each probe session for high and low variability label and object conditions for each child with typically developing language.

Figure 3.

Intervention Group 2: Prepositions correct for each probe session for high and low variability object conditions for each child with typically developing language.

Figure 4.

Intervention Group 3: Prepositions correct for each probe session for high and low variability object conditions for each child with developmental language disorder.

If high variability facilitates learning, we should see gains when high variability is presented for both objects and labels. Again, high variability for objects was implemented as using many different objects to convey the spatial relations of above or beside, while low variability of objects was implements with using similar objects that differed only in color. High variability of labels was implemented as using synonyms for labels for the objects, such ‘The ladle is above/beside the pail’ and ‘The spoon is above/beside the bucket’ in the same condition. Low variability of labels was implemented as non-specific determiner phrases for objects, such as ‘This is above/beside this’ and ‘That is above/beside that’ in the same condition. Gains are defined as initial rate of correct responses and total number of correct responses. An advantage of initial rate of correct responses was defined as an initially steeper slope of correct responses for one condition over another. Total number of correct responses is self-explanatory.

To examine the hypothesis that high variability for objects and labels helps preposition learning, visual analyses of data were conducted for typically-developing children presented prepositions in high variability for objects and labels and low variability for objects and labels. Figure 2 shows that two children (Children 5 and 6) demonstrated no preposition learning, one demonstrated mixed learning, and three demonstrated a low variability advantage (Children 2, 3, 4). For these children, the advantage is based on both initial rate and total number of correct responses. Child 1’s profile showed an unstable learning pattern based on condition.

If high variability objects (only) facilitates learning, we should see larger gains in the high variability condition. Again, high variability for objects was implemented as using many different objects to convey the spatial relations of above or beside, while low variability of objects was implements with using similar objects that differed only in color. Low variability of labels was implemented for all conditions as non-specific determiner phrases for all objects, such as ‘This is above/beside this’ and ‘That is above/beside that’ in the same teaching session. As seen in Figure 3, visual analysis reveals that, for the typically-developing children, one child (Child 11) demonstrated no preposition learning, four demonstrated a high variability advantage (Children 7, 9, 10, 12), and one child (Child 8) showed a low variability learning advantage. Child 9 showed an improved total number of correct responses in the high variability condition. Children 7, 10 and 12 demonstrated improved initial rate and total number of correct responses. Child 8 showed a low variability learning advantage in terms of both initial rate and total number correct.

If high variability of objects alone provides learning advantages for typically-developing children, we hypothesized that this condition might also provide learning advantages for children with DLD. Identical to the conditions in Group 2, high variability for objects was implemented as using many different objects to convey the spatial relations of above or beside, while low variability of objects was implements with using similar objects that differed only in color. Low variability of labels was implemented for all conditions as non-specific determiner phrases for all objects, such as ‘This is above/beside this’ and ‘That is above/beside that’ in the same teaching session. As seen in Figure 4, visual analysis shows that two children with DLD (Children 17, 18) demonstrated a learning advantage in the high variability condition, with an advantage for both initial rate and total. Four children with DLD (Children 13, 14, 15 and 16) showed a learning advantage in the low variability condition. All four children show an advantage for both initial rate and total number correct.

2 Size of gains

The number of correct responses were converted to percentages, in which the advantage of high versus low variability conditions were based on the proportion of prepositions correct out of the total number possible (18) across sessions. Table 3 illustrates the size of the gain when an advantage for a particular condition was found, independent of initial rate of learning. An advantage in this situation is a purely mathematical decision: which condition yielded a higher percentage of correct responses?

Table 3.

Number and percent prepositions correct by condition for treatment probes and percent learning advantage for all participants.

Participant number	High variability number (percentage)	Low variability number and percentage	Advantage: percent difference
Group 1: Typically developing; high low variability for object and label:
1	5 (27.8)	4 (22.2)	high 5.56
2	0 (0)	15 (83.3)	low 83.3
3	12 (66.7)	16 (88.9)	low 22.2
4	11 (61.1)	14 (77.8)	low 16.7
5	0 (0)	0 (0)	no learning
6	0 (0)	0 (0)	no learning
Group 2: Typically developing; high low variability for object:
7	18 (100)	13 (72.2)	high 27.8
8	0 (0)	16 (88.9)	low 88.9
9	3 (16.7)	1 (5.6)	high 11.1
10	17 (94.4)	0 (0)	high 94.4
11	0 (0)	0 (0)	no learning
12	5 (27.8)	0 (0)	high 27.8
Group 3: DLD; high low variability for object:
13	1 (5.6)	12 (66.7)	low 61.1
14	12 (66.7)	17 (94.4)	low 27.8
15	2 (11.1)	6 (33.3)	low 22.2
16	3 (16.7)	4 (22.2)	low 5.6
17	7 (38.9)	5 (27.8)	high 11.1
18	15 (83.3)	1 (5.6)	high 77.8

Notes. The maximum number correct was six for each probe. There were three probes, thus children could have a maximum of 18 probes correct. DLD = Developmental Language Disorder.

In Table 3, Group 1, which was comprised of typically-developing children exposed to the high variability of object and label and low variability of object and label, showed a slight advantage in favor of the high variability of object and label condition for one child (Child 1), a strong advantage in favor of the low variability of object and label for one child (Child 2), and moderate advantages in favor of the low variability of object and label for two children (Children 3 and 4). The remaining two children did not show learning (Children 5 and 6).

Group 2, which was comprised of typically-developing children exposed to the high or low variability of object only, showed moderate advantages in favor of the high variability condition for three children (Children 7, 9, 12), a strong advantage in favor of the high variability condition for one child (Child 10), and a strong advantage in favor of the low variability condition for one child (Child 8). The remaining child did not demonstrate learning (Child 11).

Group 3, which was comprised of children with developmental language disorder exposed to the high/low variability of object only conditions showed a slight advantage in favor of the low variability condition for one child (Child 16), a moderate advantage in favor of the low variability condition for two children (Children 14 and 15), a strong advantage in favor of the low variability condition for one child (Child 13), and a strong advantage in favor of the high variability condition for one child (Child 18).

3 Correlations

To examine factors that might contribute to individual differences in performance, we examined general language measures (SPELT-P2), vocabulary (PPVT-4), nonverbal cognitive scores (K-ABC-II) and maternal level of education (MLE) and correlated them with total prepositions correct in the high and low variability conditions. We examined this separately for the high variability for object only versus high variability for object and label conditions, as seen in Tables 4 and 5. Good performance on the high object and high label variability condition was strongly positively correlated with general language scores (r_s = 0.94, p < 0.05) shown in Table 4. None of the measures of interest were correlated with good performance in the low object and low label variability condition.

Table 4.

Spearman rank order correlations of test scores and demographic data with prepositions learned in high and low variability conditions for the high variability for object and label condition.

	SPELT-P2^a	PPVT-IV^b	K-ABC-II^c	Maternal education (years)	Prepositions learned: High	Prepositions learned: Low
SPELT-P2^a	–	0.64	0.23	0.40	0.94*	0.60
PPVT-IV^b	0.64	–	−0.23	0.01	0.68	0.17
K-ABC-II^c	0.23	−0.23	–	−0.19	0.33	0.27
Maternal education (years)	0.40	0.01	−0.19	–	0.33	0.22
Prepositions learned: High	0.94*	0.68	0.33	0.33	–	0.64
Prepositions learned: Low	0.60	0.17	0.27	0.22	0.64	–

Notes. * Correlation was significant at p < .05. ^a Standard scores from the Structured Photographic Expressive Language Test: Preschool: 2nd edition (Dawson et al., 2005). ^b Standard scores from Peabody Picture Vocabulary Test: 4th edition (Dunn and Dunn, 2007). ^c Standard scores from nonverbal index of the Kaufman Assessment Battery for Children: 2nd edition (Kaufman and Kaufman, 2004).

Table 5.

Spearman rank order correlations of test scores and demographic data with prepositions learned in high and low variability conditions for the high variability for object groups.

	SPELT-P2^a	PPVT-IV^b	K-ABC-II^c	Maternal education (years)	Prepositions learned: High	Prepositions learned: Low
SPELT-P2^a	–	0.38	0.40	0.26	0.07	−0.30
PPVT-IV^b	0.38	–	0.52	0.38	0.71*	−0.15
K-ABC-II^c	0.40	0.52	–	0.07	0.25	0.00
Maternal education (years)	0.26	0.38	0.07	–	0.48	−0.03
Prepositions learned: High	0.07	0.71*	0.25	0.48	–	−0.04
Prepositions learned: Low	−0.30	−0.15	0.00	−0.03	−0.04	–

Good performance on the high object only variability condition was strongly positively correlated with receptive vocabulary scores (r_s = 0.71, p < 0.05) shown in Table 5. None of the measures of interest were correlated with good performance in the low object only variability condition.

IV Discussion

We explored how different levels of input variability affected learning of prepositions. Given that this was an early feasibility study, we followed Fey and Finestack’s (2009) guidelines of evaluating whether our predictions were supported and which characteristics of the treatment might be clinically relevant. We also considered limitations of the study.

We predicted that high variability in the label and object presentation would be most helpful if variability facilitates learning. However, our results do not support that conclusion. No child in Group 1 showed a moderate or strong advantage in for the high variability for both object and label condition. It is possible that this high amount of variability, as implemented through objects and labels, was simply overwhelming to the learners. As mentioned earlier, Maguire et al.’s (2008) findings on verb learning also showed a low variability advantage. Hsu and Bishop (2014) found that older children with DLD (6–11 years of age) benefitted from lower variability on a sentence training task. Both these tasks are more grammatically complex compared to single word learning. It may be the case that the more complex the task, the more potential there is for variability to overwhelm, rather than help the learner.

Also, the fact that the SPELT-P2 scores were strongly correlated with performance on the high variability for object and label condition may suggest that children were not simply treating the highly varying labels as statistical noise. Children with better language skills (and thus more established grammatical rules) learned more in the high object and high label condition, even though there was no clear overall high variability advantage. Our intention was for the labels to serve as unstable, unpredictable stimuli in order to highlight prepositions as the constant, stable stimuli, or the pattern to be detected, per classic statistical learning principles. However, it is possible that the children may have been attending to the labels, especially those children with weaker general language skills (who still have emerging grammatical rules), as indicated by lower SPELT-P2 scores. The conclusion that the wider range of label stimuli may have hindered learning serves as a warning when selecting what to vary, considering the way children respond to real words.

Our second prediction was that children would benefit from high variability for objects if they learned prepositions as semantic representations primarily through object cues, particularly spatial relations among the objects. This seems to be supported by our data for children with typically-developing language. Four out of six typically-developing preschoolers showed a moderate to strong learning advantage in the high variability for object condition when measured both via initial rate of learning and percentage of prepositions correctly learned. Overall, in line with Perry et al. (2010), this suggests that high object variability may support learning of prepositions in a majority of children with typically-developing language.

Our third question examined how using high variability for objects would affect children with DLD learning prepositions. Recall that research suggests difficulty with vocabulary acquisition for DLD is related to weaknesses in semantic representations (Alt et al., 2004, 2013; McGregor et al., 2002). In contrast to the TD group, only two out of six of the preschoolers with DLD in our study showed a learning advantage in the high variability for object condition. However, four out of six of the preschoolers with DLD showed a learning advantage in the low variability for object condition. It appears that certain circumstances provide cues that benefit children with TD but not children with DLD. This difference may be related to degree of variability, number of exposures, or other factors that have yet to be explored.

V Clinical relevance of using variability in treatment

Given that there are few studies on how prepositions are learned by either typically-developing children or children with developmental language disorder, we now explore the clinical relevance of our design within the thorough framework of Fey and Finestack (2009). We address the following questions posed by Fey and Finestack in the hopes that this additional information will better guide future work in this area.

1 Were there outstanding characteristics to the children who were particularly successful or unsuccessful learners?

Although these findings for Groups 1 and 2 are for typically-developing children who demonstrated overall language scores that were clearly within normal limits, it is worth considering that our eligibility data showed that 63% of seemingly typically-developing children between the ages of 3;0 and 4;4 already knew either one or both of the prepositions, above and beside. The implication, then, is that the participants of Groups 1 and 2 who were included in this study were on the lower end of the typical preposition acquisition trajectory. There may have been other characteristics not detected by our tests, such as attention or memory issues, that affected the children’s learning.

In the high variability of objects only condition (Groups 2 and 3), TD and DLD children’s higher receptive vocabulary scores were significantly correlated with better performance of learning prepositions. Thus, children with stronger receptive vocabulary may be better equipped to learn prepositions in high variability for object conditions. This is especially important for children with DLD, who were far less likely to learn in the high object variability condition than their typically-developing peers. Although the current study concerns learning prepositions, it is in line with work from Nicholas (2011) that vocabulary scores are predictive of expressive use of prepositions in young children.

Although receptive vocabulary scores were only significantly correlated with better learning of prepositions in the high variability for objects condition, it is also worth noting that the three typically-developing children who learned best, regardless of condition, had the three highest vocabulary scores. Children 3, 4, and 7 learned between 61% and 100% of prepositions in both high and low variability conditions.

It is worth noting that there were some non-responders in our study in which 0% of prepositions were learned in any conditions. It is common in early treatment studies to have children who are non-responders. All three of these non-responders were considered typically-developing according to our tests, but there were no clear patterns or indicators that we could find to suggest that they would not respond to learning the prepositions.

2 How frequently and over how long a period is intervention likely to be necessary to achieve a measurable effect?

Out of the total of 12 typically-developing children tasked with learning the prepositions, five accurately responded to at least one probe in each condition (42%), four learned in one condition and not the other (33%), and three did not learn in either condition (25%). Thus, a total of 75% of participants demonstrated some learning of prepositions, and 25% of participants demonstrated no learning. This may indicate that further studies may choose to increase the teaching time from three sessions to four or five sessions in order to promote a higher percentage of prepositions learned across participants. This also informs clinicians and educators that three sessions of teaching for approximately 10 minutes each is likely sufficient for the majority of learners to make some gains; however full mastery may take longer.

All the children with DLD learned at least some prepositions in both conditions. While there was an advantage in the low variability condition for more children in this group, it would be important not to dismiss the potential of the high variability condition. It is possible that children with DLD may use the low variability condition to bootstrap learning in the high variability condition. Thus, it may be beneficial to increase the number of treatment sessions to allow for the learning advantages from exposure to the high variability condition.

3 Do the children enjoy and/or will they tolerate the approach and the intensity of treatment anticipated to be needed? Do the children stay engaged in the activities planned?

Perhaps the most clinically relevant portion of the experiment was the teaching portion, which by design was kept short—no more than 5 minutes for each preposition and all the children demonstrated no to minimal difficulty attending to this portion of the study. Notably, the high object conditions kept the children more entertained, as they were excited to see which new object would be presented next. For example, one child said, ‘I want to play with those ones next’ while pointing to the next high variability object pair. Many children also indicated their enjoyment by running up excitedly to the researchers upon their entrance into a classroom.

4 Do the activities lend themselves to the frequent administration of intervention procedures?

Luckily, preposition learning can be done with any object at hand, and intervention times were relatively short (approximately 5 minutes per preposition), thus we conclude that the procedures (especially the high object only condition techniques) readily lend themselves to easy, frequent administration of intervention procedures.

5 How much time is required to train intervention agents to administer the intervention?

In keeping with the overall perceived ease of administration of this intervention by these researchers, we can identify that training intervention agents to perform the preposition teaching as described for treatment Groups 2 and 3 (high/low object variability) takes little time. All that is required is to teach administrators to state ‘this is above/beside that’ while pointing to the indicated object. Group 1 with high label variability combined with high object variability was more challenging to train and prepare for; careful selection of objects was necessary given that many objects do not have obvious label synonyms. Careful coordination between administrators was also necessary so that one always used one specific label and the other administrator the other label. A list of labels was helpful to alleviate the memory load, as was it helpful to organize the objects in the order written on the label list. Given that the high variability for object and label yielded the least favorable results, it is unlikely that these procedures and constraints for using label variability would need to be encountered.

6 What outcome measures are most useful clinically?

Of the verbal and nonverbal responses elicited, the most useful outcome measures was clearly the nonverbal production measure. Children were able to demonstrate their understanding of prepositions this way, but almost no children would verbally produce a preposition in the probe condition. This floor effect suggests that different measures to examine verbal production of prepositions would need to be used for preschoolers in this age range. It also dovetails with the general observation that children demonstrate mastery of receptive language before mastery of expressive language.

VI Summary and conclusions

Incorporating variability, a principle of learning theory, into preposition instruction has the potential to increase learning for preschool-aged children. However, it also has the potential to make learning more difficult. In this experiment, high variability seems to work best for teaching prepositions to preschoolers when only objects, and not object and labels, are varied and when the learners do not have developmental language disorder. However, clinicians should be careful not to interpret this as a blanket statement. This was a first pass at examining variability in teaching prepositions, which have unique grammatical and semantic properties. We were not able to account for all the potential factors that might influence learning. Clearly, there are still many details to uncover about how principles of learning theory apply to different situations and different learners.

In terms of characteristics of the learners, there was a suggestion that children with higher receptive vocabulary scores are likely to do better in a high variability condition. Although most of the children with DLD did not show an advantage for the high object variability condition, this does not suggest an inability to learn using principles of learning theory. The literature has shown that people with language disorders can learn with treatments infused with these principles (e.g. Alt et al., 2014; Plante et al., 2014). However, it may be the case that children with DLD will need a different application of these principles in terms of number of examples, length of the intervention, or other manipulations. Although there are more details to hone in on in terms of the best way to use variability when teaching prepositions to preschoolers, the initial promise of the technique makes it a worthwhile line of inquiry to pursue.

Footnotes

Acknowledgements

Thank you to: the children, families, directors, and staff of the preschools and daycares that participated in this study; members of the L4 lab for help with data collection; Becky Vance, Barbara Kiernan, Lea Cuzner, and Amy Ashcraft for help with recruitment of children with language impairment; Shawnesha Wallace and Emma Woods for assistance with assessment; Jennifer Oleson for operations support, transcription, and scoring; and Elena Plante and Rebecca Gómez for useful discussion. The work presented in this paper is based on work by the first and third author as partial fulfillment of the requirements for an advanced clinical degree.

Author’s Note

Katrina E Nicholas is now affiliated with California State University, East Bay, USA.

Declaration of conflicting interest

The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.

Funding

The author(s) received no financial support for the research, authorship, and/or publication of this article.

References

Alt

Meyers

Alt

(2013) Using ratings to gain insight into conceptual development. Journal of Speech, Language, and Hearing Research 56: 1650–61.

Alt

Meyers

Ancharski

(2012) Using principles of learning to inform language therapy design for children with specific language impairment. International Journal of Language and Communication Disorders 47: 487–98.

Alt

Meyers

Oglivie

Nicholas

Arizmendi

(2014) Cross-situational statistically based word learning intervention for late-talking toddlers. Journal of Communication Disorders 52: 207–20.

Alt

Plante

Creusere

(2004) Semantic features in fast-mapping: Performance of preschoolers with specific language impairment versus preschoolers with normal language. Journal of Speech, Language, and Hearing Research 47: 407–20.

Bishop

DVM

Snowling

Thompson

Greenhalgh

, CATALISE consortium (2016) CATALISE: A multinational and multidisciplinary delphi consensus study: Identifying language impairments in children. PLOS ONE 11: e0158753

Caselli

Casadio

Bates

(1999) A comparison of the transition from first words to grammar in English and Italian. Journal of Child Language 26: 69–111.

Dawson

Eyer

Fonkalsrud

(2005) Structured photographic expressive language test: Preschool. 2nd edition. DeKalb, IL: Janelle.

Dollaghan

Campbell

Paradise

et al . (1999) Maternal education and measures of early speech and language. Journal of Speech, Language, and Hearing Research 42: 1432–43.

Dunn

(2007) Peabody picture vocabulary test. 4th edition. Circle Pines, MN: American Guidance Service.

10.

Eisenbeiss

Bartke

Clahsen

(2006) Structural and lexical case in child German: Evidence from language-impaired and typically-developing children. Language Acquisition 13: 3–32.

11.

Ellis Weismer

Hesketh

(1993) The influence of prosodic and gestural cues on novel word acquisition by children with specific language impairment. Journal of Speech and Hearing Research 36: 1013–25.

12.

Fey

Finestack

(2009) Research and development in child language intervention: A five-phase model. In: Schwartz

(ed.) Handbook of child language disorders (pp. 513–531). New York: Psychology Press.

13.

Gómez

(2002) Variability and detection of invariant structure. Psychological Science 13: 431–35.

14.

Greenslade

Plante

Vance

(2009) The diagnostic accuracy and construct validity of the structured photographic expressive language test: Preschool: 2nd edition. Language, Speech, and Hearing Services in Schools 40: 150–60.

15.

Grela

Rashiti

Soares

(2004) Dative prepositions in children with specific language impairment. Applied Psycholinguistics 25: 467–80.

16.

Horner

Spaulding

(2010) Single-case research designs. In: Salkind

(ed.) Encyclopedia of research design (pp. 1386–94). Thousand Oaks, CA: Sage.

17.

Hsu

Bishop

DVM

(2014) Training understanding of reversible sentences: A study comparing language-impaired children with age-matched and grammar-matched controls. PeerJ 2: e565.

18.

Johannes

Wilson

Landau

(2016) The importance of lexical verbs in the acquisition of spatial prepositions: The case of in and on. Cognition 157: 174–89.

19.

Kan

Windsor

(2010) Word learning in children with primary language impairment: A meta-analysis. Journal of Speech, Language, and Hearing Research 53: 739–56.

20.

Kaufman

(2004) Kaufman assessment battery for children. 2nd edition. Circle Pines, MN: AGS.

21.

Landau

Jackendoff

(1993) ‘What’ and ‘where’ in spatial language and spatial cognition. Behavioral and Brain Sciences 16: 217–38.

22.

Landau

Johannes

Skordos

Papafragou

(2017) Containment and support: Core and complexity in spatial language learning. Cognitive Science 41: 748–79.

23.

Leonard

Eyer

Bedore

Grela

(1997) Three accounts of the grammatical morpheme difficulties of English-speaking children with specific language impairment. Journal of Speech and Hearing Research 40: 741–53.

24.

Maguire

Hirsh-Pasek

Golinkoff

Brandone

(2008) Focusing on the relation: Fewer exemplars facilitate children’s initial verb learning and extension. Developmental Science 11: 628–34.

25.

McGregor

Newman

Reilly

Capone

(2002) Semantic representation and naming in children with specific language impairment. Journal of Speech, Language, and Hearing Research 45: 998–1014.

26.

Nelson

(1973) Structure and strategy in learning to talk. Monographs of the Society for Research in Child Development 38: 1 and 2, 1–135.

27.

Nicholas

(2011) Children’s omission of prepositions in English and Icelandic. PhD Thesis, University of Arizona, Tucson, AZ, USA. Available at: https://eric.ed.gov/?id=ED528700 (accessed January 2019).

28.

Ozcaliskan

Dimitrova

(2013) How gesture input provides a helping hand to language development. Seminars in Speech and Language 34: 227–36.

29.

Perry

Samuelson

Malloy

Schiffer

(2010) Learn locally, think globally: Exemplar variability supports higher-order generalization and word learning. Psychological Science 21: 1894–1902.

30.

Plante

Oglivie

Vance

et al . (2014) Variability in the language input to children enhances learning in a treatment context. American Journal of Speech Language Pathology 23: 530–45.

31.

Puglisi

Befi-Lopes

Takiuchi

(2005) Use and comprehension of prepositions by children with specific language impairment. Pró-Fono Revista de Atualização Científica 17: 331–44.

32.

Reilly

Bishop

DVM

Tomblin

(2014) Terminological debate over language impairment in children: Forward movement and sticking points. International Journal of Language and Communication Disorders 49: 452–62.

33.

Saffran

(2003) Statistical language learning mechanisms and constraints. Current Directions in Psychological Science 12: 110–14.

34.

Sinha

Thorseng

Hayashi

Plunkett

(1994) Comparative spatial semantics and language acquisition: Evidence from Danish, English, and Japanese. Journal of Semantics 11: 253–87.

35.

Suanda

Mugwanya

Namy

(2014) Cross-situational statistical word learning in young children. Journal of Experimental Child Psychology 126: 395–411.

36.

Vance

Clegg

(2012) Use of single case study research in child speech, language and communication interventions. Child Language Teaching and Therapy 28: 255–58.

37.

Watkins

Rice

(1991) Verb particle and preposition acquisition in language impaired preschoolers. Journal of Speech and Hearing Research 34: 1130–41.

38.

Zimmerman

Steiner

Pond

(2002) Preschool language scale. 4th edition. San Antonio, TX: The Psychological Corporation.