How do children with developmental language disorder extend novel nouns?

Highlights

• Children with DLD generalized nouns for solid objects, animates and roles correctly.

• They generalized substance and spatial nouns more poorly than children without DLD.

• They relied more easily on shape than on relations when extending novel nouns.

• Increasing the number of training exemplars is beneficial for novel noun extension. Vocabulary level is significantly correlated with novel noun extension.

Abstract

In this study, we investigated the ability of children with developmental language disorder (DLD) to extend nouns referring to different categories of novel objects. In a word extension task, we used several types of object entities (solid, animate, nonsolid, functional, and spatial relations) for which children needed to attend to diverse properties (shape, texture, role, or spatial relation) to decide category membership. We compared 15 school-aged children with DLD with typically developing (TD) children matched on either age or vocabulary. Our results indicate that children with DLD were impaired in extending novel words for nonsolid substances and relational objects, whereas age-matched TD children performed well for all object classes. Similar to children with DLD, TD children matched on language had difficulty in extending spatial relation categories. We also show that children with DLD needed more learning exemplars and relied more on shape-based information than TD children, especially for spatial configuration objects. Overall, our findings suggest that children are able to learn regularities between object properties and category organization and to focus on diverse features according to the object presented when extending novel nouns. They also provide clear evidence linking DLD to deficits in novel name generalization and word learning.

Introduction

Vocabulary is the set of words of a language known by an individual. Whereas vocabulary size grows quickly in typically developing (TD) children (Bloom, 2002), children with developmental language disorder (DLD) seem to struggle to expand their lexicon. DLD is a neurodevelopmental disorder characterized by impairments in language development and learning, including difficulties in novel word learning, and it is not due to a biomedical condition linked to genetic or neurological causes (Bishop, Snowling, Thompson, Greenhalgh, & CATALISE‐2 Consortium, 2017). These language impairments persist until middle school and beyond, and they have an arguably deleterious impact on children’s everyday life.

Previous research has shown that children with DLD produce their first words later than their peers (Trauner, Wulfeck, Tallal, & Hesselink, 2000). Furthermore, as reported by McGregor, Oleson, Bahnsen, and Duff (2013), their vocabulary is characterized by not only limited breadth (i.e., the number of words they know) but also limited depth (i.e., how well they comprehend the meanings of the words they know). In a word definition task, McGregor et al. showed that children with DLD produce significantly fewer informative definitions than TD children even when the analyses included only children who manifested a real ability to provide definitions. These findings suggest that children with DLD may learn novel words but that the meanings of words remain less detailed than in TD children.

Word learning is a complex multidimensional process that encompasses several components such as phonological encoding, form–meaning mapping, lexical access, word retention, and semantic knowledge and organization as well as working memory (Kan & Windsor, 2010, p. 740). Several studies have reported that children with DLD perform worse than their age-matched peers but similar to their language-matched peers in learning novel words (for a review, see Kan & Windsor, 2010). In addition, they are slow to learn new words, have difficulty in retaining new word labels, encode fewer semantic features of newly learned items, and require more exposure to novel words in order to learn them (Alt et al., 2004, Jackson et al., 2016, Nation, 2013). It should be noted that these studies have focused mainly on the initial stage of lexical learning (form–meaning mapping) and not on all aspects of lexical development (Kan & Windsor, 2010). The authors considered broad variables such as the number of novel word exposures, types of words (e.g., nouns or verbs), types of tasks or output formats (e.g., identification or production of words), and presentation format (e.g., fast or slow). These factors tell us what children learned easily (e.g., nouns better than verbs) and/or the best way in which to discriminate children with or without DLD (e.g., testing after high exposure or with comprehension and recognition tasks), but they tell us nothing about the underlying causes of these difficulties.

A promising line of research aiming at understanding word learning difficulties in children with DLD has been to investigate the cognitive processes known to support typical lexical acquisition such as novel word generalization (i.e., word extension). When young children learn a novel word, they need to understand the set (category) of instances to which this word refers to be able to extend or generalize it to new instances in various contexts (Gentner and Namy, 1999, Kan and Windsor, 2010). For example, if children encounter an animal and are told that it is called poodle, they must understand the features characterizing poodles in order to extend it accurately later while restricting the word poodle to poodles only. Furthermore, novel name extensions are not constrained in the same way across noun types. It has been argued that children learn very early that names for solid objects apply to objects that share the same shape. To illustrate, when children learn the word cup, they need to specifically attend to the object shape because shape is a relevant property when deciding whether or not an object can be included in the cup category (Smith, Jones, Landau, Gershkoff-Stowe, & Samuelson, 2002). This shape bias favors shape over other properties such as size and color (Gershkoff-Stowe and Smith, 2004, Jones and Smith, 1998). The shape bias has been consistently observed in 2- and 3-year-old children (Gershkoff-Stowe and Smith, 2004, Jones and Smith, 1998, Samuelson and Smith, 2000). Smith et al. (2002) showed that teaching 17-month-old infants nouns for categories well organized by shape resulted in these children developing a shape bias not only for these categories but also for other novel solid object categories. Interestingly, infants who developed the shape bias increased their vocabulary for solid objects outside the laboratory during the course of the study. Therefore, this bias is an important tool for learning in young children.

Other categories are organized by perceptual features other than shape, and other word extension biases are at play. For example, when objects are identified as animates, the shape bias is weakened and texture becomes a critical feature to extend new words. Jones, Smith, and Landau (1991) showed that 2- and 3-year-old children extended new words by shape when objects had no eyes but did so by shape and texture when eyes were added to the same objects. Jones and Smith (1998) found that 3-year-olds extended novel nouns by shape for objects without shoes and by texture for objects with shoes. In this context, children over 2 years of age extend nonsolid object names by texture and not by shape (Soja, Carey, & Spelke, 1991). However, such a texture bias seems to be more fragile (Samuelson & Horst, 2007) or more context dependent (Perry, Samuelson, & Burdinie, 2014) than the shape bias.

Another category of nouns, relational nouns, has also received much attention. One feature of relational nouns is that they are defined not by perceptual features but rather by extrinsic properties. “Their meanings include relations between other concepts” (Gentner, 2005, p. 248). Nouns such as predator, robbery, and neighbor are relational because they refer to a relational structure between entities (e.g., between two animals or between an individual and an object), whereas their referents are often visually dissimilar or have no common intrinsic property (Gentner, 2005). It is worth noting that relational nouns are acquired later than solid object names. In fact, there are few relational nouns in the MacArthur Communicative Developmental Inventory until 30 months of age (Gentner, 2005). Moreover, when relational nouns are acquired, they are often understood as object names defined by concrete features. Contrary to school-aged children, preschoolers tend to define relational nouns in terms of objects’ perceptual properties. For example, children find it hard to believe that a 2-year-old infant could be an uncle because uncles need to be about 24 or 25 years old (Keil & Batterman, 1984).

Regarding novel word learning, it has been argued that children are able to use different kinds of properties to extend novel nouns referring to various object classes, depending on the object with which they are confronted (Jones and Smith, 1998, Jones et al., 1991, Snape and Krott, 2018). Thus, children seem to be sensitive to regularities between object properties and category organization. This sensitivity is important because of its impact on the size of the lexicon. For example, Thom and Sandhofer (2009) found that 20-month-old infants who were trained with more color words were better able to extend new color words than children who were trained with fewer words, suggesting that word extension is facilitated in domains for which children have a labeling experience. Thus, as children learn more words, they develop attentional biases toward properties that are relevant to category membership (Jones & Smith, 1998). Because object names are the most frequent in young children’s daily life, it is not surprising that the shape bias is their first and most robust attentional bias (Samuelson, Horst, Schutte, & Dobbertin, 2008; but see Bloom, 2002, for a different interpretation).

Most of the evidence regarding the above attentional biases has been obtained with a single training exemplar design in which participants are shown a training stimulus and then are asked to select, among a set of options, the one that has the same name. However, in many learning situations, two or more stimuli are introduced simultaneously with the same word, giving children the opportunity to compare these stimuli before generalizing their names to novel instances. There is now ample evidence that a comparison format leads to more conceptually based generalizations in novel noun learning tasks than single stimulus formats (see Augier and Thibaut, 2013, Namy and Gentner, 2002). A general explanation is that comparisons lead participants to first focus on salient similarities between training exemplars (e.g., shape) and to later search for deeper commonalities, even though less salient (Namy & Gentner, 2002). This is especially the case for relational nouns for which comparisons support the identification of relational features over perceptual features in young children (Christie and Gentner, 2010, Gentner et al., 2011, Thibaut and Witt, 2015, Thibaut and Witt, 2017). For example, Christie and Gentner (2010) found that 3- and 4-year-old were able to extend a novel noun for a spatial relation when two learning exemplars were provided, but not when only one learning exemplar was presented. Multiple object presentation favors generalization in word extension tasks, although it can give rise to generalizations based on salient but irrelevant properties (see Augier & Thibaut, 2013).

Only a few studies have systematically investigated novel noun extension in children with DLD. Schwartz, Leonard, Messick, and Chapman (1987) found that these children apply a new noun to unnamed exemplars less than their language-matched peers. Gray (1998) also found that children with DLD do not perform as well as age-matched peers when they need to extend words to new contexts (i.e., color photographs or black line drawings). More recently, Collisson, Grela, Spaulding, Rueckl, and Magnuson (2015) found that, contrary to their age-matched peers, preschool-aged children with DLD did not exhibit a shape bias. They did not preferentially extend solid object names by shape given that they relied equally on shape, color, and texture. Therefore, DLD seems to be associated with word extension impairment in addition to deficits in other aspects of word learning such as semantic representation and form–meaning mapping (Kan and Windsor, 2010, McGregor et al., 2013).

However, although other classes of nouns have received much attention in TD children, to our knowledge no studies have investigated word extension for words other than solid object names in children with DLD. Pandolfe, Wittke, and Spaulding (2016) showed that adolescents with DLD were impaired in their knowledge of words related to driving, which are abstract words and, for many of them, relational words. Moreover, previous studies with children with DLD have shown that learning other types of relational words, such as verbs, is problematic for this population (Windfuhr, Faragher, & Conti-Ramsden, 2002). This is interesting because the difficulties of children with DLD in learning verbs suggest that learning other types of words such as relational nouns may also be hampered in this population.

Although being exposed to two learning instances facilitates relational noun extension in TD children (Christie and Gentner, 2010, Gentner et al., 2011), such an experimental setup might not benefit children with DLD because DLD is associated with difficulty in using comparisons and structural alignment (Leroy et al., 2014, Leroy et al., 2012). These children also have difficulty in identifying relational similarities when they are not supported by perceptual features (Leroy et al., 2012, Leroy et al., 2014). In analogical reasoning tasks, Leroy et al. (2014) showed that children with DLD had poorer performance in a linguistic task (composed of syllables) and a nonlinguistic analogical reasoning task (composed of pictures) in which they needed to complete a sequence sharing the same relational structure as previously presented sequences. Interestingly, in the linguistic task, the performance of children with DLD was poorer for relational items that were not backed up by perceptual cues (e.g., /na-ba-ba/ and /ty-sy-sy/) than for relational items that were backed up by perceptual cues (e.g., /my-ly-ly/ and /me-le-le/). This suggests that relational terms themselves may be difficult to learn and generalize by children with DLD when different instances of a relation do not share any perceptual cue (e.g., different classes of objects can be connected by the “neighbor of” relational noun).

In sum, children with DLD have difficulty in learning and extending new words (Gray, 1998, Kan and Windsor, 2010, Schwartz et al., 1987). They do not develop a shape bias at the same age as their TD peers (Collisson et al., 2015), and they have difficulty in learning abstract and relational words (Pandolfe et al., 2016, Windfuhr et al., 2002). Moreover, children with DLD are impaired in using comparison, and they have difficulty in focusing on relational features rather than perceptual features (Leroy et al., 2012, Leroy et al., 2014). In contrast, TD children can identify regularities across objects in order to extend novel names for different classes of entities. TD children also develop attentional biases toward features that are relevant to category membership (Jones and Smith, 1998, Jones et al., 1991, Soja et al., 1991). In this context, within-category comparisons seem to contribute to the abstraction of conceptually relevant properties, especially for words that are not defined by perceptually based features such as relational nouns (Christie and Gentner, 2010, Gentner et al., 2011, Thibaut and Witt, 2015).

The main purpose of this study was to examine generalization abilities in school-aged children with DLD using a word extension task. During the experiment, children first were introduced to an unfamiliar object together with its name (a nonword), and then they were asked to indicate the other stimulus to which they would generalize the novel noun.

This study had four objectives. First, we evaluated the ability of children with DLD to extend nouns associated with different kinds of entities and to identify the relevant properties to category membership according to the object that has been introduced. Indeed, novel noun generalization abilities have not been compared for nouns referring to various types of entities. Thus, comparing how novel words for various types of entities are acquired remains an open question. We compared extensions of novel nouns for five object classes: solid objects, animate objects, nonsolid substances, functional role categories, and spatial configuration categories. These object classes have already been examined separately (Christie and Gentner, 2010, Gentner et al., 2011, Jones and Smith, 1998, Smith et al., 2002, Soja et al., 1991) or in comparison (e.g., solid and nonsolid objects; see Kucker et al., 2019, for a recent review) in various studies in TD children, but to our knowledge these five classes have never been compared in the same study in children with DLD or in TD children.

Second, we have seen that these abilities are related to the size of the lexicon (Smith et al., 2002, Thom and Sandhofer, 2009). Therefore, it is possible that children with DLD are impaired relative to age-matched peers but perform similarly to children matched on a vocabulary measure. Thus, we included two TD matched groups: one younger (the language-matched group) and one older (the age-matched group). This double-matching procedure would allow us to disentangle language-level issues from age-related issues.

Third, we contrasted single-object and multiple-object presentations. Indeed, it has been repeatedly shown that the opportunity to compare multiple instances of the same category leads to more accurate generalizations (Christie and Gentner, 2010, Gentner and Namy, 1999). Given the limited size of our available sample of participants with DLD, we could not run the comparison and single-object conditions as a between-participants factor. Thus, we decided to first start with a single occurrence and, in case of failure, multiple exemplars of the same category were presented. Note that this new methodological within-participants approach corresponds to what may happen in real-world situations. Children sometimes encounter one stimulus of a category and later encounter two (or more) specimens of the same category.

Fourth, we analyzed a possible generalization bias of children with DLD to focus on salient perceptual features even in relational categories. Indeed, as mentioned above, younger children sometimes generalize novel relational nouns based on perceptual similarities (Christie and Gentner, 2010, Keil and Batterman, 1984). This may also happen with children with language deficits given that these children have already been found to favor perceptual similarities over relational features (Leroy et al., 2012, Leroy et al., 2014).