Abstract
Despite the common intuition that labeling may be the best way to teach a new word to a child, systematic testing is needed of the prediction that children learn words better from labeling utterances than from directive utterances. Two experiments compared toddlers’ label learning in the context of hearing words used in directive versus labeling utterances. In Study 1 (N = 64) 24-month-olds learned a novel label equally well from directive and labeling utterances, whereas 18-month-olds only learned in the labeling context. When the novel label was placed at the end of a directive utterance in Study 2 (N = 16), even 18-month-olds were able to learn it. These findings highlight young children’s flexibility in interpreting words in a variety of contexts, and the importance of considering the various linguistic and non-linguistic settings where words are encountered.
Young children learn many new things through exploration and discovery, yet there are crucial aspects of the world that they must learn from others (Harris & Koenig, 2006). Words are a good example; because they are arbitrary conventional communicative symbols, their meanings must be learned from others. This need not imply explicit teaching, however; a great deal of young children’s learning happens in spontaneous non-teaching situations (Callanan & Valle, 2008; Rogoff, 2003). In the current studies, we compare young children’s learning of words in didactic and non-didactic contexts.
Much of the language learning literature, beginning with Brown’s (1958) description of the ‘original word game,’ has emphasized didactic naming of objects, ostensive definition, as an important way for children to learn word meanings (Callanan, 1991; Gelman, Coley, Rosengren, Hartman, & Pappas, 1998). In this early work, ostension was defined by an adult in joint attention with a child, pointing to an object and labeling it, as in ‘This is a mango’ (Ninio & Bruner, 1978). More recently, the word ‘ostension’ has been used by Csibra and Gergely (2006, 2009) to describe a broader range of situations, where any intent to communicate new information is counted as ostension. Because of the multiple meanings of the word, we will refer to ostensive definition utterances as ‘labeling utterances.’
Didactic labeling seems an intuitively straightforward way to connect an object to its name. In many experimental studies of word learning, therefore, children are presented with novel words in just such statements. Further, Tan and Schafer (2005) found a positive correlation between parents’ self-reported ostensive naming and toddlers’ word learning (measured by looking time). To test the hypothesis that labeling utterances (‘This is an X’) are more effective word learning contexts for young children than non-labeling utterances, however, a systematic comparison of these two types of utterances is needed. For a variety of reasons, it is possible that labeling utterances are easier to learn from, especially early in development. Labeling may make it easy for the child to establish joint attention with the adult, in that the adult’s referential intention may be made clear by redundant non-linguistic (pointing, eye gaze) and linguistic cues (the labeling sentence frame).
However, several studies have demonstrated that young children can learn new labels even when the object being labeled is not physically present at the time the label is uttered. For example, Tomasello, Strosberg, and Akhtar (1996) showed that 18-month-olds were able to hold a novel label in mind and attend to non-verbal cues to an experimenter’s referential intent. The experimenter said, ‘Let’s find the gazzer!’ and in one condition showed a dissatisfied expression at the first object found, then a joyful expression at the second object. Children subsequently showed that they paired the word with the object that the experimenter seemed happy to find. Other studies have found that two-and-a-half and three-year-olds learn common and proper nouns as well in non-labeling as in labeling contexts (Jaswal & Markman, 2001, 2003), and that two-year-olds learn object labels and verbs through overhearing (Akhtar, 2005a; Akhtar, Jipson, & Callanan, 2001) and actually learn verbs better in non-labeling than in labeling contexts (Tomasello & Kruger, 1992).
One non-labeling frame in which young children often hear names for objects is the imperative or directive, where an adult asks the child to do something with an object, e.g., ‘Pick up the doll.’ Early studies suggested that directives may not support children’s word learning, and may even be negatively related to vocabulary development (Della Corte, Benedict, & Klein, 1983; Nelson, 1973; but see Barnes, Gutfreund, Satterly, & Wells, 1983). For example, Hoff (2006) reported negative correlations between parents’ use of directives and children’s vocabulary development, arguing that this is perhaps because directive speech may be focused on controlling the child and redirecting his/her attention.
Some researchers have argued, then, that learning from labeling utterances may be easier than learning from directive utterances because labeling utterances more often follow the child’s attention, whereas directives often redirect the child’s focus. Clearly, considerable evidence has supported the view that children learn words more easily when a speaker follows their attention rather than redirects it (Tomasello & Farrar, 1986). However, directive speech does not always involve directing attention; Akhtar, Dunham, and Dunham (1991) found that mothers’ use of directives that followed their children’s focus positively predicted children’s subsequent vocabularies (see Masur, Flynn, & Eichorst, 2005; Masur, Flynn, & Lloyd, 2013, on ‘supportive’ versus ‘intrusive’ directiveness; Pine, 1992). Moreover, even directives that lead a child’s attention can be beneficial if they succeed in redirecting focus; Shimpi and Huttenlocher (2007) found that maternal utterances that attempted to but did not succeed in redirecting one-year-old children’s focus negatively predicted vocabulary, whereas those that were successful positively predicted vocabulary.
Thus, we know that toddlers can learn new words in various non-labeling contexts, but no study has experimentally compared the effectiveness of labeling utterances to directive utterances that do not redirect attention. Correlational studies suggest that directives that follow children’s attention may be just as effective for word learning (Akhtar et al., 1991; Shimpi & Huttenlocher, 2007). It is important to investigate whether, as is commonly assumed, ostensive definition is a particularly effective way of introducing a new object label in the very early phases of word learning. In two experiments, we systematically compared 18- and 24-month-old children’s learning of object labels in labeling and directive contexts that both involved joint attention, testing the hypothesis that children’s word learning is more robust with labeling than with directive speech.
Study 1
In Study 1, 18- and 24-month-old children were taught novel labels for novel objects in two between-subjects conditions. In the Labeling condition, novel names were introduced in statements such as ‘It’s a wuggie!’ In the Directive condition, new labels were embedded in requests for children to drop objects down a plastic chute: ‘Put the wuggie down there.’ Subsequently, each child was asked both a comprehension question (‘Which one is the wuggie?’) and a preference-control question (‘Which one do you like?’), in counter-balanced order.
Method
Participants
Participants were recruited from a pool of mostly middle-class European-American families who had volunteered to participate in studies of child development in typically developing children. The final sample consisted of 32 toddlers (17 girls) aged 16–21 months (M = 18.28 months, SD = 1.53), and 32 (15 girls) aged 22–26 months (M = 24.5 months, SD = 1.44). Seven additional 18-month-olds (3 males, 4 females) and two additional 24-month-olds (1 male, 1 female) were dropped due to uncooperativeness, experimenter error, or parent over-involvement. To ensure that the two experimental groups did not differ in productive vocabulary, we used the short version of the CDI words checklist (Reznick & Goldsmith, 1989). The mean productive vocabulary score of 18-month-olds in the Labeling condition was 24.0 words (SD = 19.66); in the Directive condition, the mean was 16.38 words (SD = 20.31). For 24-month-olds in the Labeling condition the mean vocabulary score was 54.12 words (SD = 25.24); in the Directive condition the mean was 61.38 words (SD = 35.98). Neither of the group differences were significant based on independent sample t-tests.
Design
Children were randomly assigned to the Labeling or Directive condition, with approximately equal numbers of males and females in each condition and age group. Each child saw two sets of four objects in counter-balanced order, with one object serving as the named ‘target’ in each set. Each object served as the target an equal number of times in each condition. After one set of objects, children experienced a comprehension test trial, and after the other, they experienced a preference control trial. In each condition, half of the children experienced the comprehension trial first.
Materials
One set of novel objects included a rolling toy that wobbled, a noisemaker, a plastic chain, and a rubber beehive. The other included a spinning top, colorful linked blocks, a wide yo-yo, and a wallpaper roller. Other novel objects were available as replacements if a given child had a name for any of these eight objects. Each participant saw one set of objects for the comprehension trial, and the other for the preference trial. Each set was used equally often in each type of trial. A set of familiar objects (Barney figure, ball, doll, spoon) were used in a warm-up task.
The labeling training apparatus consisted of four opaque buckets with lids mounted in a row on a wooden plank. The directive training apparatus consisted of a set of four opaque buckets with lids, and a chute embedded in a large plastic cube. Each bucket contained a single hidden object, objects were always placed in the same order, and two objects (one from each set) were designated as the target objects for each participant. Trays were used to present the objects during the testing phases.
Procedure
Participants came to a laboratory playroom for one half-hour visit. Parents were shown the novel objects and asked whether their child produced or comprehended names for any of them. If a parent reported that the child knew an object, it was replaced with another from a reserve set. Substitute objects were used for 11 children, but none served as targets. Parents were asked to complete a productive vocabulary checklist (Form A of the CDI words checklist, short version, Reznick & Goldsmith, 1989).
After a short warm-up period, the experimental session began. An observer behind a one-way mirror videotaped all sessions. An assistant moved objects to buckets as prescribed by the experimental conditions. The child sat on the floor or on his/her parent’s lap facing the experimenter with the hiding apparatus between them.
Labeling condition
The experimenter began with a warm-up round, opening each bucket one at a time, stating, ‘Let’s see what’s in here!’ and showing the child each familiar object. The child was allowed to play with each object for a few seconds and was asked to replace it before being shown the next one.
The experimenter then proceeded to two blocks of training and testing (one block followed by comprehension testing and one by preference testing). In each block, the child played three rounds of a finding game with the four novel objects. Each round consisted of the experimenter extracting each object from its respective bucket once. Objects were extracted one at a time, with each being replaced before another was found. The experimenter talked about the target and non-target objects equally and behaved in a predetermined way: she held up the object with wide eyes and an ‘Ah!’ gasp, saying, ‘It’s a toma! This is a toma! Here’s the toma!’ when presenting the target, and saying, ‘Look at this one! Here’s this one! See this one!’ when presenting the non-targets. The target and non-target objects were then placed in random positions on a tray for either the comprehension or preference test.
A warm-up comprehension task, designed to elicit giving and showing responses from the child, preceded the first test trial. The four familiar objects from the warm-up round were presented on a tray, and after the child had correctly chosen at least two objects, indicating that he or she had understood and complied with the requests, the experimenter proceeded to either the comprehension or preference trial. In the comprehension test trial, the experimenter said ‘Show me the toma’ while establishing eye contact with the child and not looking at the objects. After the child chose an object, the experimenter said, ‘Okay’ in a neutral tone of voice and replaced the toy in the tray. If the child did not respond, the request was repeated until a toy was chosen. For four children, the parent was asked to ask the child for the target object, or the experimenter asked the child to ‘Give Mommy the wuggie.’ Two of these children were 18-month-olds in the Labeling condition, one was an 18-month-old in the Directive condition, and one was a 24-month-old in the Labeling condition. Care was taken to be sure that the parent did not direct the child’s choice in these cases (if so, they would have been dropped from the study for parent over-involvement).
The preference control trial was identical except that the experimenter asked the child to ‘Show me your favorite toy’ or ‘Pick the one you like the best.’
The finding game was played twice (with different sets of objects), once followed by a comprehension test trial, and once by a preference control trial; thus each child served as his/her own control. The word ‘toma’ was used to label the target object for one set, while ‘wuggie’ or ‘modi’ were used for the other target. The label was used three times in each round of the finding game for a total of nine models.
Directive condition
The experimental set-up was identical in the Directive condition except that both the buckets and the chute were placed between the child and the experimenter. The experimenter began with a warm-up round, opening each bucket one at a time, stating ‘Let’s put this one down the chute,’ and guiding the child in dropping the familiar objects down the chute. After the child dropped an object, the experimenter replaced it in its bucket and closed the lid before extracting the next object.
In the training trials that followed, each round of the finding game consisted of the experimenter extracting each novel object, one at a time, and encouraging the child to put it down the chute. The experimenter held up each object with wide eyes and a surprised gasp, saying, ‘Put the toma down the chute! Put the toma in here! Put the toma down here!’ when presenting the target object, and saying ‘Put this one down the chute! Put it in here! Put this one down here!’ when presenting the non-target objects. The action, rather than the label, was emphasized. Each child heard a total of nine tokens of each novel word – three tokens in each round. The experimenter always searched the buckets in the same order, and knew which object was the target and what word to use by referring to a cue card on the wall. As in the Labeling condition, children in the Directive condition completed two training plus testing blocks with different sets of objects, one ending with a comprehension test and one with a preference test (in counter-balanced order). The first trial (either comprehension or preference) was preceded by a warm-up comprehension trial with the familiar objects from the warm-up round.
Two independent observers reviewed 25% of the videotapes and achieved 100% agreement on which object was chosen on the comprehension and preference trials.
Results and discussion
The number of children (out of 16) in each condition who chose the target object on the comprehension and preference trials is shown in the first two rows of Table 1. Above-chance performance, p < .05, measured by a binomial test (with chance probability = .25), required at least 9 of the 16 children to choose the correct item. As seen in the table, the number of children choosing the target object on the comprehension trial was greater than chance for 24-month-olds in both conditions, but above chance for 18-month-olds only in the Labeling condition. Performance on the preference control trials was not different from chance for either age group in either condition. Both age groups performed equally well on the preference trials in the Labeling and Directive conditions, Fisher exact probability test, p > .05.
Number of children (out of 16) choosing target object on comprehension and preference trials.
p < .05 (binomial test).
While these data support the claim that 24-month-olds are as likely to learn new words from directive utterances as labeling utterances, further comparisons are needed of individual patterns of response. In particular, taking advantage of the within-subject component of the study, we considered how many children chose the target item for both comprehension and preference questions. Sign tests that take into account all three possible combinations of scores (ties, choosing target for preference but not comprehension, choosing target for comprehension but not preference) revealed that a significant number of 24-month-old children chose the target object for the comprehension but not the preference trial in both the Labeling and the Directive conditions, ps < .05. Sign tests for the 18-month-old children’s responses revealed a non-significant trend in the Labeling condition, p < .07, one-tailed, but chance performance in the Directive condition, p > .10, one-tailed. These tests showed that the predicted pattern occurred more frequently than expected by chance for the 24-month-olds in both conditions, thus, individual patterns are consistent with the results of the binomial tests, at least for the 24-month-old children.
By the age of two years for this middle-class US sample, then, these results suggest that new words can be learned quickly, whether introduced in deliberate labeling utterances or embedded in directives. Younger children, however, may learn better from labeling utterances. It is striking that by 24 months novel words were learned equally easily from labeling and directive utterances. For the 18-month-olds, follow-in directives may be less helpful than labeling for learning new words. Alternatively, the results could be explained by a confound between conditions. In the Labeling condition, children heard the novel word at the end of the utterance (‘It’s a modi.’). In the Directive condition, however, the novel label occurred in the middle of the utterance (‘Can you put the modi down here?’). Given evidence that children pay close attention to the ends of utterances (Echols & Newport, 1992; Shady & Gerken, 1999), it may be difficult, especially for 18-month-olds, to parse the new word from the middle of the experimenter’s sentence. In Study 2 we tested this alternative explanation.
Study 2
Method
Participants
Sixteen children (8 male), between 16 and 21 months (M = 18.7 months, SD = 1.62) participated in a modified Directive condition. The families came from the same population as the previous study; no child participated in both studies.
Materials
The materials were identical to Study 1.
Procedure
The procedure was similar to Study 1, with the exception that all children participated in a Directive condition with a slightly reworded introduction of the novel label for the target object. Rather than saying, ‘Put the modi down here,’ the experimenter pointed to the chute and said, ‘Drop the modi.’ Stress was on the verb rather than on the novel noun, so that pragmatically the utterance sounded like a directive rather than a labeling utterance. For each of the non-target objects the experimenter said, ‘Drop this one.’ In the testing phase, as in Study 1, each child experienced a comprehension test trial and a preference control trial in counter-balanced order.
Results and discussion
The number of children choosing the target object on the comprehension and preference trials is shown in Table 1 (bottom row). Above-chance performance, p < .05, measured by binomial test (with chance probability = .25) required at least 9 of 16 children to choose the correct item. As the table shows, the number of children choosing the target item on the comprehension trial was greater than expected by chance, but the number choosing the target item on the preference trial did not differ from chance.
A sign test revealed that the likelihood of the predicted pattern (choosing the target object on the comprehension but not the preference trial) occurred more often than expected by chance, p < .05, one-tailed.
Because this was a modification of the Directive condition used in Study 1, we compared the comprehension performance of the children in this study with that of the 18-month-old children in Study 1. A Fisher exact probability test comparing the 18-month-old children in Study 2 to those in the Labeling condition of Study 1 revealed no difference between the two groups in comprehension, p > .05. A second Fisher exact test uncovered a significant difference between the comprehension of the 18-month-olds in the two Directive conditions, with children in Study 2 performing better than those in Study 1, p = .031, one-tailed.
General discussion
The hypothesis addressed by these studies was that didactic labeling is a particularly effective way to teach a new word to a young child and would therefore lead to more learning than hearing a novel word in a directive. Our data lead us to reject this hypothesis. In Study 1, 24-month-olds were just as likely to learn a new word, as evidenced by a comprehension test, when they heard it in a directive as in a labeling statement. The younger children in Study 1 showed a different pattern, with less learning in the Directive condition, but when a confound was corrected in Study 2, even 18-month-olds were above chance in learning new words from directive utterances, and performed as well as 18-month-olds in the Labeling condition of Study 1.
The current findings suggest that young children are quite flexible in learning new words in different conversational settings. This seems consistent with studies of children learning words through ‘fast mapping,’ such as in Carey and Bartlett’s (1978) classic study where children learned a new color word through a single exposure in the midst of activity and conversation. Horst and Samuelson (2008) point out the importance of distinguishing between children’s ability to identify a referent in the fast mapping paradigm versus their ability to retain the new word meaning over a longer time period. Indeed, Sabbagh and Shafman (2009) showed that children who learned a word from an uncertain speaker remembered that the person used the word, but did not retain the word meaning after a delay. And yet Jaswal and Markman (2003) found that three-year-olds retained the meaning of a new word for several days, even when it was learned through inference rather than labeling. Future research should extend our understanding of the time course of word learning, for example by exploring younger children’s retention of word-object pairings as a function of the conversational setting in which they originally learned the word.
It is interesting to consider children’s word learning in the directive context in light of Csibra and Gergely’s (2006, 2009) pedagogy theory, which hypothesizes that humans are evolutionarily adapted to learn from others through ostension. Given Csibra and Gergely’s more inclusive definition for ostension one might argue that both conditions in our study fit this definition of an adult intentionally communicating something to a child. Whereas the intention of a labeling utterance is to teach a word, directive utterances are intended to engage a child in action. Our findings suggest that 18-month-old toddlers can learn new words even when the explicit goal of the adult’s communication is not to teach a word, but to direct the child’s behavior.
These findings leave open the question of how important joint attention is in early word learning. To equate the two conditions as much as possible, joint attention was present in both the Labeling and Directive conditions, and some have argued that joint attention is a key component of early word learning (Tomasello & Farrar, 1986). However, children as young as 18 months can learn words through overhearing others’ conversations (Floor & Akhtar, 2006; Gampe, Liebal, & Tomasello, 2012), shedding doubt on joint attention as a requirement for early learning (Akhtar, 2005b; Akhtar & Gernsbacher, 2007; see also Scofield & Behrend, 2011).
Further research on parents’ labeling in everyday settings and children’s word learning is needed, given evidence that the kind of ostensive ‘naming lessons’ that occur in middle-income western families may not be as common in other communities (Heath, 1982; Lieven, 1994; Rogoff, 1990; Schieffelin & Ochs, 1986). Perhaps it is especially impressive that the middle-class US children in our studies were equally able to learn from directive statements as labeling statements, given that they are growing up in a cultural setting where labeling utterances are so common. Yet, even in middle-class European-American families, explicit labeling is just one among many contexts in which children hear object names (Callanan & Sabbagh, 2004; Gelman et al., 1998; Luce & Callanan, 2010).
Clearly there is much work to do to better understand how children’s word learning develops in light of diverse and constantly changing situations. Further research is needed on variations in language spoken to children (Camaioni, Longobardi, Venuti, & Bornstein, 1998; Laosa, 1980; Leaper, Anderson, & Sanders, 1998) and on how these variations are linked to children’s early word learning (Akhtar et al., 1991; Shimpi, Fedewa, & Hans, 2012; Vigil, Tyler, & Ross, 2006). As children get older it is also important to consider changes in children’s word learning contexts as they hear novel words used in a wider range of social settings (Akhtar & Montague, 1999), and develop different strategies for learning words (Kalashnikova, Mattock, & Monaghan, 2014).
In short, these findings question existing assumptions that explicit labeling is the most effective way for very young children to learn new words. Future research is needed to explore more fully the range of communicative settings in which children are exposed to words, how this may vary across cultural groups, and to consider subtle questions about the depth of learning that can result from brief exposure to a new word.
Footnotes
Funding
This research was partly funded by a Faculty Research Grant from the University of California, Santa Cruz.