METHOD

Materials

We chose five familiar objects to be used as distracters to test children's knowledge of the names for the novel objects: a toy car, a ball, a book, a teddy bear and a spoon. For the experimental items, we used eighteen novel objects (see Figure 1 and Appendix A). We randomly assigned eighteen novel names to the objects, which were all one or two syllables long, easily pronounceable and resembled English words (e.g. wug, donka or kig, see Figure 1 and Appendix A). The names were judged by five native British adult speakers to be nonce words. Objects were paired so that each pair could be combined in two ways, one of which was always a FOR relation and the other a HAS or LOCATED relation (six HAS relations and three LOCATED relations). In addition, both combinations could be referred to by the same compound. For example, for the compound kig donka, a kig was either glued to a donka (HAS) or a donka was used as a container for a kig (FOR). For the compound bindle fep, a fep was located inside a bindle (LOCATED) or a fep was used to open a bindle (FOR). To avoid any effects of name–object pairings, we swapped the names for the two objects within a pair for half of the participants in each age group. There were always three exemplars of each object pair, one to introduce the objects individually, the other two to present the two relations.

Fig. 1.

Pairs of novel objects used in experiments.

RESULTS AND DISCUSSION

Apart from one occasion, when one three-year-old refused to choose an object in the compound test, all children and all adults made a decision and chose one of the combined objects. As expected, combinations with either HAS or LOCATED relations showed very similar results (see Table 1). We therefore collapsed these two relations for further analyses. Figure 2 shows the number of choices for HAS/LOCATED relations for the five age groups. When averaging responses over items, an analysis of variance with age group (adults, five-, four-, three- and two-year-olds) as independent variable and number of HAS/LOCATED responses as dependent variable showed an effect of age group (F(4, 138)=7·4, p<0·001, η_p²=0·177). Tukey HSD post hoc tests revealed that the responses of two-year-olds differed significantly from those of four-year-olds (p=0·003), five-year-olds (p<0·001) and adults (p=0·001). No other differences were significant. Furthermore, children's performance across compounds was very similar. The number of children choosing the HAS/LOCATED interpretation was not significantly different among the nine items in any of the age groups: two-year-olds (χ ² (8, N=196)=2·83, p=0·945), three-year-olds (χ ² (8, N=167)=1·18, p=0·162), four-year-olds (χ² (8, N=147)=5·88, p=0·661), five-year-olds (χ² (8, N=152)=5·38, p=0·716) and adults (χ ² (8, N=102)=3·18, p=0·923).

Fig. 2.

Mean number of HAS/LOCATED choices (out of nine) for the different age groups in Experiment 1. Horizontal line represents chance level and numbers different from chance are marked by *.

TABLE 1.

Mean proportions and standard deviations of HAS and LOCATED responses in Experiment 1 for all age groups

Comparing numbers of HAS/LOCATED responses against chance (4·5) showed that two-year-olds (t(29)=9·4, p<0·001, d=1·71) and three-year-olds (t(29)=4·7, p<0·001, d=0·87) chose HAS/LOCATED relations more often than predicted by chance. Five-year-olds' and adults' choices, on the other hand, did not differ from chance (all ts <1) and four-year-olds were at the borderline (t(27)=2·0, p=0·053, d=0·38). Thus, two- and three-year-olds clearly preferred HAS/LOCATED relations over FOR relations when asked to identify the referent of a compound. Because five-year-olds and adults did not exhibit a preference for either relation, they perceived both interpretations as being equally plausible. Four-year-olds appear to be at a transition stage. Importantly, they do not show a clear preference for HAS/LOCATED relations.

In this experiment, participants were exposed to quite a number of novel objects together with their functions. Young children might have been overloaded with so much new information, especially function information. It might therefore be possible that children revealed a bias towards HAS/LOCATED interpretations because they were overloaded with function information. To rule out such an interpretation, we compared participants' responses for the first four and the last four compounds for each age group (Bonferroni-corrected α=0·05/5=0·01). We found that, for most age groups, responses did not significantly differ for the first and last four compounds (two-year-olds: 3·13 vs. 2·97, t(29)=0·84, p=0·407; four-year-olds: 2·36 vs. 2·11, t(27)=0·94, p=0·355; five-year-olds: 1·88 vs. 2·12, t(32)=−0·98, p=0·332; adults: 2·18 vs. 1·86, t(21)=1·20, p=0·246) except for three-year-olds, who chose HAS/LOCATED interpretations more often for the first four compounds than for the last four compounds (2·80 vs. 2·23, t(29)=2·89, p=0·007, d=0·64). This rather suggests that three-year-olds started off with a preference for HAS/LOCATED interpretations, but considered FOR relations as plausible interpretations more often in the latter half of the experiment. This might have been caused by an increasing number of function relations presented to them as candidates for compound interpretations. Importantly, it is very unlikely that the overall preference for HAS/LOCATED interpretations in two- and three-year-olds is due to them being overloaded with function information.

The linguistic descriptions of the relations in this experiment were chosen to clearly capture the specific relations of the objects in order to enhance children's understanding of the relations. However, the young children in our experiment might not have fully understood all the verbs that we used to describe the relations (e.g. ‘lifts’ and ‘stores’) and therefore might have chosen the easier understandable HAS/LOCATED relations. In addition, Appendix B shows that the use of different verbs for the FOR relations led to unequal distributions of verbal expressions. FOR relations were expressed by a variety of verbs (e.g. pushes, holds, opens), while HAS/LOCATED relations were more consistently expressed by using either has (six times) or is in (three times). Adult studies have shown that the interpretation of a novel noun–noun combination (e.g. student vote) is affected by whether the preceding phrase uses the same relation (e.g. Gagné, Reference Gagné2001). If children are affected in the same way, the dominance of HAS/LOCATED relations in our material might have primed children's preference for these relations. While we know of no a priori reason to expect that younger children might have been more strongly affected by priming, it is a logical possibility. Because of these reasons we conducted a second experiment, in which we kept the verbal expressions for the two choices equally frequent over all trials, replacing the verbal expressions for the FOR relation with a simpler ‘is … for’ and only using HAS relations and not LOCATED relations.

METHOD

GENERAL DISCUSSION

Our findings indicate that when young children interpret a novel noun–noun compound, they have a bias towards interpreting the referents of the compound's constituents as being related by a HAS or LOCATED relation as compared to being related by a FOR relation. This preference was not affected by the verbal expressions used to describe the compound's relations. In the current set of experiments, the preference was present for two- and three-year-olds, but not for five-year-olds or adults.

The findings for five-year-olds contrast with those of an earlier compound explanation task, in which they preferred HAS/LOCATED interpretations (Krott et al., Reference Krott, Gagné and Nicoladis2009). However, the present study used a forced choice task; FOR and HAS/LOCATED interpretations were presented as equal alternatives, and the FOR interpretations were made concrete, which made them strong competitors. In addition, responses did not require many metalinguistic skills, which means there was no reason to avoid the more difficult to explain FOR relations. Finally, we used compounds composed of novel words rather than compounds composed of existing words (as in Krott et al., Reference Krott, Gagné and Nicoladis2009). Our finding therefore shows that in these circumstances five-year-olds do not exhibit a HAS/LOCATED bias. Because adults had not shown this bias in an explanation task, they were not expected to show it in the present paradigm either. Because our results revealed a gradual decrease of the HAS/LOCATED bias from age two on, it is in line with the assumption that we are dealing with a word learning bias that gradually disappears when children become older. This bias might be the reason that children's compound interpretations become adult-like very late, so that even six- and nine-year-olds occasionally reveal it (Krott et al., Reference Krott, Gagné and Nicoladis2009; Nicoladis, Reference Nicoladis2003; Parault et al., Reference Parault, Schwanenflugel and Haverback2005).

What remains unanswered is what exactly is the constraint on interpretations of novel compounds that makes young children prefer HAS/LOCATED interpretations over FOR relations. It is unlikely that children prefer HAS and LOCATED relations because they do not understand or remember the presented FOR relations. Even infants aged 1 ; 6 and younger are sensitive to the function of objects (e.g. Barnat, Klein & Meltzoff, Reference Barnat, Klein and Meltzoff1996; Booth & Waxman, Reference Booth and Waxman2002; Madole, Oakes & Cohen, Reference Madole, Oakes and Cohen1993) and it has been shown that at least three-year-olds understand and remember simple functions of novel objects, similar to the ones used in our experiment (e.g. Diesendruck, Markson & Bloom, Reference Diesendruck, Markson and Bloom2003; Smith et al., Reference Smith, Jones and Landau1996). What distinguishes HAS and LOCATED relations from FOR relations is that they are concrete and tend to be perceptually stable because they link referents that tend to be permanently physically contiguous, while FOR relations are more abstract, not always perceptually available because they link referents that are not continuously physically contiguous. For example, an apple pie always HAS apples and zoo animals are typically LOCATED in a zoo. A baby bottle (bottle FOR babies), however, is not continuously used to feed a baby, nor does a teacup (cup FOR tea) continuously hold tea. It is therefore more difficult to associate baby with baby bottle or tea with teacup and it might be more difficult to create a conceptual representation that denotes the relation (FOR) between the constituents baby and bottle. In other words, these compounds, although transparent to adults, might initially be seen as opaque from the child's perspective, similar to the opaque contribution of straw to a strawberry. We chose a paradigm that made the relations more equal in that both HAS/LOCATED relations and FOR relations were visually perceivable and in both cases the constituents were permanently visible. Acting out the FOR relation might have drawn more attention to the FOR relations. However, the latter did not seem to be the case because no age group preferred FOR interpretations. An important difference that remained between the two object combinations was that objects related by HAS and LOCATED relations were permanently physically contiguous and therefore perceptually stable, while the FOR relations were not perceptually stable because the function was presented for a short time and the physical relationship, for example the physical distance, between the two objects changed. Because young children preferred HAS/LOCATED interpretations, they might have preferred interpretations with permanent physical contiguity. Older children had revealed a bias towards HAS and LOCATED interpretations when asked to explain the meaning of unknown compounds (Krott et al., Reference Krott, Gagné and Nicoladis2009). In the present study, they might have benefited from viewing a concrete depiction of the two interpretations for the compound, which might have made a FOR interpretation more plausible. Alternatively, not needing to explain the FOR relations might have made it easier to respond with a FOR interpretation.

The bias towards HAS and LOCATED relations and against FOR relations resembles another word learning bias that concerns perceptual features, namely children's focus on the shape of artefacts (e.g. Gentner, Reference Gentner and Gardner1982a; Landau, Smith & Jones, Reference Landau, Smith and Jones1998; Merriman et al., Reference Merriman, Scott and Marazita1993; Smith et al., Reference Smith, Jones and Landau1996; Tomikawa & Dodd, Reference Tomikawa and Dodd1980). Both biases can overrule function information, both rely more heavily on perceptually stable features of an object than does the function (or FOR relation) and both are more prominent in two-year-olds than five-year-olds. As mentioned, it can be ruled out that their common source is the problem with understanding functions as such because much younger children have been shown to be sensitive to the function of objects (e.g. Barnat et al., Reference Barnat, Klein and Meltzoff1996; Booth & Waxman, Reference Booth and Waxman2002; Madole et al., Reference Madole, Oakes and Cohen1993). The difficulty with function seems to arise in the context of word learning (e.g. Deák, Ray & Pick, Reference Deák, Ray and Pick2002; Landau et al., Reference Landau, Smith and Jones1998). Both biases might therefore have a common source, namely children's attention to and reliance on stable perceptual features of objects.

Note that it is not the stability itself that distinguishes perceptual features such as shape or HAS/LOCATED relations from function. The intended function of an object or the function relation within a compound is stable as well. This is in accordance with the observation that compounds as a whole tend to be used to express stable (i.e. habitual) relations rather than accidental relations (Downing, Reference Downing1977; Gleitman & Gleitman, Reference Gleitman and Gleitman1970; Levi, Reference Levi1978). But in contrast to features such as shape or HAS relations, which tend to be both perceptually and conceptually stable, functional stability lies at a pure conceptual level. For example, while the shape of a key remains both perceptually and conceptually stable, the function of a key is not always perceivable, but it remains stable at a conceptual level. Similarly, the HAS relation in clock tower (tower that HAS a clock) is perceptually and conceptually stable, because the tower always has a clock. In contrast, the function of a baby bottle is not perceptually stable because one cannot always see that a baby bottle is used to feed a baby, but it remains stable at a conceptual level. It therefore appears to be a perceptual stability rather than a pure conceptual stability that young children initially focus on when interpreting the meaning of artefact nouns and compound nouns. As children gain more experience with words and their referents, they appear to be able to infer a more abstract stability that underlies word meaning, such as function.

But what leads young children to focus on perceptual stability? One account that tries to explain children's biases toward perceptual features is Smith and colleagues' ‘attentional learning account’ (ALA). The ALA argues that such biases are attentional biases that arise from the learning of statistical regularities in children's early noun categories (e.g. Gershkoff-Stowe & Smith, Reference Gershkoff-Stowe and Smith2004; Graham, Williams & Huber, Reference Graham, Williams and Huber1999; Jones, Reference Jones2003; Smith et al., Reference Smith, Jones and Landau1996; Smith & Samuelson, Reference Smith and Samuelson2006). For example, children are drawn to the shape of objects, while they should focus on their function instead, because they have learned from artefact nouns they know that artefacts refer to ‘things in categories organized by shape’ (Smith & Samuelson, Reference Smith and Samuelson2006). The emergence of the shape bias is also understood as being closely linked to vocabulary growth. Smith (Reference Smith and MacWhinney1999) showed that children's shape bias emerged after the number of count nouns in their vocabulary reached fifty or more. Similarly, Gershkoff-Stowe & Smith (Reference Gershkoff-Stowe and Smith2004) showed in a longitudinal study of eight children that the emergence of a shape bias coincided with an acceleration of the children's noun vocabulary. Furthermore, the vocabulary of children aged 1 ; 3 to 1 ; 7 has been shown to accelerate due to training of categories that are organized by shape (Samuelson, Reference Samuelson2002; Smith, Jones, Landau, Gershkoff-Stowe & Samuelson, Reference Smith, Jones, Landau, Gershkoff-Stowe and Samuelson2002).

Extending the ALA to compounds would mean that children learn from the compounds they know that compounds refer to two (or more) objects that are related by HAS or LOCATED relations, and that this bias develops with an early increase in compound vocabulary size. For that, HAS and LOCATED relations would need to be the preferred interpretation within their vocabulary. It is not clear whether this is the case, though. As mentioned in the introduction, adults' interpretations of compounds occurring in the CHILDES database suggest that FOR is the dominant relation in children's compound vocabulary, which would appear to rule out such an explanation.

The question arises whether the FOR relation is less dominant in the vocabulary of younger children and increases with age. Similar to the survey in Krott et al. (Reference Krott, Gagné and Nicoladis2009), we investigated this question by gathering noun–noun compounds in the British-English transcripts in the CHILDES database (MacWhinney, Reference MacWhinney2000), focusing on the child and child-directed speech of two-, three-, four- and five-year-olds (911 transcripts of 66 two-year-olds, 162 transcripts of 68 three-year-olds, 37 transcripts of 23 four-year-olds and 23 transcripts of 18 five-year-olds). We then added the codings of modifier–head relations of these compounds gathered for the study by Krott et al. (Reference Krott, Gagné and Nicoladis2009). Those were based on the majority coding of five British native speakers using the relations listed in Appendix C, which are inspired by Levi's categories (Levi, Reference Levi1978). Table 2 provides an overview of the relations occurring in compounds produced by children and adults, together with their percentage within the compound vocabularies at different ages. The distributions of relations are very similar to that of all British-English transcripts in CHILDES, as presented in Krott et al. (Reference Krott, Gagné and Nicoladis2009), and do not differ much between the different age groups. The distributions are very skewed, with a few relations making up the majority of compounds, among which are the relations of interest, i.e. FOR, HAVE1 (which is equivalent to the HAS relation in our study), and LOCATED. FOR relations are the most dominant relations, equally prevailing in both child speech and child-directed speech (about 30–40%), and importantly this dominance is very similar for all age groups, including adults (χ ² (7, N=1404)=2·83, p=0·900). HAVE1 and LOCATED relations occur each in less than 10% of the compounds. Similar to the FOR relations, the occurrence of LOCATED and HAVE1 relations does not differ across age groups, including adults (LOCATED: χ ² (7, N=1404)=3·49, p=0·836; HAVE1: χ ² (7, N=1404)=2·50, p=0·927).

TABLE 2.

Percentage of modifier–head relations in CHILDES compounds in children's and adults' speech at ages 2, 3, 4 and 5 (n=number of compounds produced)

Given these distributions, if children's preference in compound interpretations was based solely on patterns in their compound vocabulary, then one would expect a preference for FOR interpretations rather than HAS or LOCATED interpretations. However, young children might not understand all relations the way adults do. For example, they might understand a teapot as a pot that HAS tea in it instead of a pot that is used FOR tea. Similarly, they might think a handbag is a bag LOCATED in a hand, not a bag that is carried by hand. These interpretations are not completely incorrect, but they are not how adults would define them. If young children ‘misunderstand’ a lot of function relations, then the bias for HAS/LOCATED interpretations might indeed originate in statistical regularities within children's vocabulary. Older children, i.e. five-year-olds, might have a better understanding of functional relations in compounds and therefore see functional interpretations and HAS/LOCATED interpretations as equally plausible when both interpretations are presented to them, as in the present study. It is possible that a critical mass of compounds with a FOR relation in children's vocabulary is necessary for children to consider the FOR relation as a possible relation when interpreting novel compounds. In sum, the ALA could in principle account for a HAS/LOCATED bias, but one would need to know more about young children's understanding of relations in familiar compounds. In addition, one would need to investigate whether the HAS/LOCATED bias is closely linked to children's growing compound vocabularies.

To conclude, we have shown that young children prefer HAS/LOCATED interpretations over FOR interpretations for novel compounds. We have suggested that this preference arises because HAS/LOCATED relations are more concrete and perceptually stable than the FOR relation. In the current studies, which provided a concrete depiction of the objects, older children (i.e. five-year olds) did not exhibit a preference for the HAS/LOCATED interpretation. Together with studies on children's interpretations of simple objects, our findings suggest that children might be biased to focus on stable perceptual features when interpreting unfamiliar words. The particular reason for such a bias remains unclear, although statistical regularities in the vocabulary, as suggested by the ALA, is a viable possibility.