Get 20M+ Full-Text Papers For Less Than $1.50/day. Start a 7-Day Trial for You or Your Team.

Learn More →

Constraint on word‐learning constraints

Constraint on word‐learning constraints Abstract: Many researchers have agreed that word learning in young children is guided by so-called “word-learning principles.” However, these principles may make it difficult to learn a substantial part of the lexicon unless they are appropriately controlled. To learn proper names, the taxonomic assumption and/or the shape bias must be overridden; to learn names for substances, withdrawal of the whole-object assumption and/or the shape bias is required; and to learn lexical hierarchies, the mutual exclusivity assumption must be suspended. In certain languages, syntax can provide useful information in this situation. For example, if a novel noun is given to an object for which the name is known in the syntactic frame “This is X,” Englishspeaking children may assume the noun to be a proper noun, and this will help them override the taxonomic assumption. However, this information is not available to Japanese children, since the Japanese language does not have grammatical markers to flag the distinction between count nouns and mass nouns, or the distinction between proper nouns and common nouns. In this paper, I discuss how Japanese children get around this problem. Key words: noun acquisition, word-learning principles, constraint, syntactic cues, crosslinguistic comparison. It has been agreed among researchers that children do not face word learning unprepared. In other words, they are not led astray by the well known induction problems in inferring a meaning of a word (Quine, 1960). Instead , they constrain the potentially infinite possibilities in mapping a word onto its meaning (for detailed discussion of this, see Imai, 1997; Markman, 1989) by using internal principles which help them to map a word onto its meaning even at first exposure to it (e.g., Markman, 1989; Markman & Hutchinson, 1984). Among the proposed principles or constraints, the wholeobject assumption, the taxonomic assumption, and the mutual exclusivity assumption (e.g., Markman, 1989; Markman & Hutchinson, 1984; Markman & Wachtel, 1988) have attracted much attention (e.g., Golinkoff, Mervis, & HirshPasek, 1994; Hall, 1991, 1994; Imai, Gentner, & Uchida, 1994; Landau, Smith, & Jones, 1988; Merriman & Bowman, 1989; Waxman & Gelman, 1986). Although a majority of studies agree that these principles are in fact used by children from a very early age, there has been much debate with respect to the specific nature of each of these principles. For example, are they available prior to the onset of word learning and applied from the first word (Waxman & Markow, 1995)? Are they universally applied irrespective of any specific linguistic properties of the input language (Imai & Gentner, 1997)? A more serious problem that has remained unanswered, however, arises from the fact that all of these principles must be suspended in some circumstances and overcome in others. That is, although the word-learning principles such as the whole-object assumption, the taxonomic assumption/shape bias, and the mutual exclusivity assumption can greatly constrain © 1999 Japanese Psychological Association. Published by Blackwell Publishers Ltd, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA. M. Imai the possible search space in mapping words on to concepts, they could also block the learning of a substantial portion of a vocabulary unless their applications are somehow controlled. For example, the whole-object assumption should not be applied when a child learns words for substances, such as water, sand, and sugar (Imai & Gentner, 1997; Soja, Carey, & Spelke, 1991). Learning names for specific individuals, that is, proper nouns, requires suspension of the taxonomic assumption (Hall, 1991; see also Woodward & Markman, 1997). The mutual exclusivity assumption must be overcome in order for a child to learn category names at different levels of the taxonomic hierarchy as well as names for particular individuals (Gelman & Taylor, 1984; Hall, 1991; Haryu & Imai, 1999; Taylor & Gelman, 1989; Waxman, Shipley, & Shepperson, 1991). Thus, for successful word learning, children must not only possess the word-learning principles but they also need heuristics that enable them to determine when to apply, suspend, or overcome the principles. In fact, the literature suggests that 2-year-olds’ vocabulary includes nonobject words such as names for substances and events as well as proper names (e.g., Nelson, Hampson, & Shaw, 1993). Children are able to learn basic-level object labels very rapidly by applying word-learning principles such as the whole-object assumption and the shape bias, but at the same time are also able to learn nonobject words and proper names by suspending them. How, then, do children control and constrain the application of the word-learning principles? It has been suggested in the literature that two sources of information are particularly important in this respect, namely, semantic (including ontological) knowledge, and syntactic cues for different form classes. In the case of English, it so happens that there is a high correlation between semantic (ontological) classes and syntactic classes of nouns. That is, individuated entities, typically solid objects, are mapped onto count nouns, while nonindividuated entities, typically substances, are mapped onto mass nouns. Furthermore, among the names for individuated entities, © Japanese Psychological Association 1999. names for particular individuals, that is, proper nouns, are syntactically distinguished from names for object kinds, that is, count nouns, in that count nouns, but not proper nouns, occur with determiners (cf., Bloom, 1994). Numerous studies have investigated the relative role of semantics and syntax in word learning (e.g., Bloom & Kelemen, 1995; Gelman & Taylor, 1984; Imai & Mazuka, 1997; Soja, 1992; Subramanyam & Landau, 1995). These studies suggest that both semantic knowledge and syntax influence word learning, and that there is an intricate interplay between the two. For example, Gelman and Taylor (1984) manipulated the linguistic form class (proper noun vs. common noun) and conceptual type of the referent (animal vs. inanimate object) in examining how 2-year-old English-speaking children interpret novel nouns. They found that when a noun was given to a stuffed toy in the proper-noun syntax, children restricted the word to the named animal, interpreting the noun as a name for a particular individual. In contrast, when a noun was presented in the count-noun syntax, they were willing to extend it to another animal of the same kind, suggesting that they interpreted the noun as a name for a particular kind of animal rather than as a name for a particular individual. However, the 2-year-olds did not show this pattern when the referent was an inanimate object. When a novel noun was introduced in the proper-noun syntax, they did not restrict the word to the named toy. Rather their performance became random. This suggests that 2-year-old English-speaking children know that it does not make sense for inanimate objects to have proper names. Thus, Gelman and Taylor’s study showed that both syntax and semantics influence children as young as 2 years of age in how they assign meanings to novel nouns. Given this intricate relationship between semantics and syntax, some researchers, P. Bloom in particular (e.g., Bloom, 1994), have gone so far as to claim that children do not need word-learning principles such as Constraint on word-learning constraints the whole-object assumption or the taxonomic assumption. All they need is knowledge of distinct ontological classes with respect to individuation (individuals, kinds of individuals, kinds of portions), and of how these conceptual classes map on to syntax. But what happens when such syntactic cues are completely lacking in a particular language? In Japanese, all nouns, including count nouns, mass nouns, and proper nouns, are treated the same in syntax. That is, there is no syntactic marker distinguishing the names of objects, substances, and proper nouns. Furthermore, the syntax does not mark the singular/plural distinction, either. Thus, the following English expressions, “This is a dax,” “Those are daxes,” “This is some dax,” “This is dax,” “This is Dax” are all translated into a single expression, “Kore [This] wa [topic/subject marker] dax desu [is].” In other words, when one hears “Kore wa dax desu” without seeing the named entity, there is absolutely no way of inferring whether dax refers to a single object, multiple objects, a substance/portion, a property (such as color), or a particular individual. The question then is how Japanese children determine whether a novel noun is a name for an object category, or a name for a substance, or a name for a particular individual. Are they totally at a loss? Or are they somehow able to constrain and coordinate the application of multiple word-learning principles even without cues from syntax? In this paper, I will argue that this is in fact the case. Two lines of my research are described, and I explore how young Japanese children control and coordinate wordlearning principles for efficient word learning. I will then discuss the implications of my work on Japanese children’s word learning for the nature of children’s word learning across languages. assumption and the shape bias.1 Dedre Gentner and I asked whether Japanese children are able to extend novel nouns in an ontologically correct fashion, extending a noun associated with an object on the basis of shape but extending a noun associated with a substance on the basis of material identity, without blindly applying the whole-object assumption and the shape bias (Imai & Gentner, 1997). Inspired by the seminal work of Soja et al. (1991) on the role of conceptual understanding of the ontological difference between objects and substances in early word learning, we wished to know whether Japanese children would demonstrate the same appreciation of the ontological distinction in their word learning, even though their native language lacks syntactic cues that flag the two distinct ontological classes. This question is important since there has been a debate on the origin of the ontological distinction between objects and substances. The philosopher Quine conjectured that children come to know the ontological differences between objects and substances only after learning the count/mass syntax (Quine, 1969). Challenging Quine, Soja et al. (1991), who are developmental psychologists, argued that children possess this knowledge perhaps at birth, and certainly before the acquisition of count/ mass syntax. The Japanese language is an ideal medium to test which of the two positions are correct, Constraints on the application of the whole-object assumption and the shape bias by ontological knowledge I will first discuss how Japanese children constrain the application of the whole-object In both of the studies described in this paper (Imai & Gentner, 1997; Imai & Haryu, 1998), the shape bias and the taxonomic assumption are indistinguishable. The shape bias and the taxonomic assumption yield similar consequences for words denoting basic-level objects, since at the basic level, category members are highly similar in shape (Rosch, 1978). Thus we consider the shape bias without considering the taxonomic assumption separately. The taxonomic assumption and the shape bias yield different results when category membership and shape similarity conflict. At early stages of word learning, children’s assumption about the possible extension of a word meaning is better characterized as a shape bias rather than a taxonomic assumption, but this early shape bias gradually shifts into a taxonomic assumption (Imai et al., 1994). © Japanese Psychological Association 1999. M. Imai since Japanese grammar does not make a distinction between these two ontological classes. A strong interpretation of Quine yields a prediction that it is not possible for Japanese children to learn the ontological distinction between objects and substances. In this case, Japanese children may not possess the wholeobject assumption and the shape bias at all, or, even if they do, they may not be able to exert the ontologically appropriate application of these principles. In contrast, if Soja et al.’s universal ontology view is correct, Japanese children should possess the ontological distinction between objects and substances, and be able to utilize this knowledge in inferring word meanings. We thus extended Soja et al.’s paradigm cross-linguistically, and compared monolingual Japanese-speaking and English-speaking children and adults. A cross-linguistic study of word-meaning projections for objects and substances Our subjects were monolingual Japanesespeaking and English-speaking 2-year-olds, 21⁄ 2-year-olds, 4-year-olds, and adults. The paradigm was a forced-choice triad task, where a novel label was given to the standard entity, and the subject was asked to determine which of the two alternatives could share the same label as the originally named entity. One of the alternatives, the shape alternative, had the same shape, but was made out of a different material. The other alternative, the material alternative, consisted of a portion of the standard entity. The stimulus materials included three types of entity as the referents of novel labels: complex objects, simple objects, and substances. The first type, which we characterized as complex objects, were real objects (artifacts) that had fairly complex shapes and distinct functions, although the functions were unknown to the children (e.g., a T-shaped plumbing fixture). The second type, which were called simple objects, were made of a solid substance such as clay or wax, and were formed into a very simple shape such as the shape of a kidney. These entities were somewhat between © Japanese Psychological Association 1999. the complex objects and substances, allowing a double construal. They could be construed as individuated objects, since they met the physical properties of so-called “Spelke objects.” If you picked up a part of the entity and moved it, the whole thing moved. At the same time, it could be construed as a lump of some substance as well. You could continue breaking the entity into pieces until each piece was no longer visible to the naked eye, and a micropiece of it was still the “same” substance as the original. The third type of entity, the substances, were nonsolid substances such as sand or hair gel. However, unlike typical substances in the real world, they were arranged in somewhat distinct, interesting shapes when presented (see Figure 1). For both language groups, novel labels were embedded in a syntactic frame that would not reveal the noun’s count/mass status. As mentioned earlier, in the case of Japanese, this exactly reflects the language’s structure. The label was uttered in a sentence, such as “Kore wa dax desu,” and dax could be interpreted as a name for a type of object, a name for a substance, a name for a particular individual, or even a name for a property. In English, we had to ensure that the noun’s count/mass status would not be revealed. The experimenter said, “Look at this dax! Can you find the tray that also has the dax on it?” In this task, if subjects interpreted the novel noun of ambiguous syntactic status as a name for a type of object, they would extend the label on the basis of shape identity; if they interpreted it as a substance name, they would extend it on the basis of material identity, suspending the whole-object assumption and the shape bias. To repeat, whether the noun was an object name or a substance name could not be detected from the syntax in this task. Thus we should be able to assume that the pattern of word-meaning projection for a given entity, that is, whether the child extends the noun on the basis of shape or material, reflects the child’s construal of that entity. Although the experimental results had several important aspects, in this paper I will focus on the application of the whole-object Constraint on word-learning constraints Complex objects Substances Porcelain lemon juicer Hand cream Wooden lemon juicer Pieces of porcelain Simple objects Hair gel Blobs of hand cream Kidney-shaped wax Kidney-shaped plaster Figure 1. Pieces of wax Sample materials used in the Imai and Gentner (1997) study. © Japanese Psychological Association 1999. M. Imai Complex objects 100 assumption and the shape bias (for other aspects of the study, see Imai & Gentner, 1997; Imai & Mazuka, 1997). Are the whole-object assumption and the shape bias universally used in young children’s word learning? If so, are Japanese children able to restrict the application of these word-learning principles to the appropriate kind of entity, that is, preindividuated objects? Further, is there any influence of the input language in the application of these principles, given that Japanese children could “constrain” the application of the whole-object assumption and the shape bias by ontological knowledge about objects and substances? Universal application of the whole-object constraint and the shape bias. The answers to the first two questions raised above are affirmative. Japanese children, just as Englishspeaking American children, extended a novel label on the basis of shape when the label was associated with an object of complex shape. In both language groups, there was a large difference in the proportion of shape responses between the complex-object trials and the substance trials, and this difference held across all age groups (see Figure 2). These results suggest that children as young as 2 years apply the whole-object assumption and the shape bias independent of the structure of the input language. Furthermore, they also suggest that children universally know from the very early stages of word learning that objects and substances belong to fundamentally distinct ontological classes. Application of the whole-object assumption and the shape bias is constrained by this universal ontological knowledge, as suggested by Soja et al. (1991) and Soja (1992) (see also Hall, 1996). Influence of the language in applying the principles. However, as to the third question, the answer is also affirmative but in a restricted sense. That is, we found a substantial crosslinguistic difference for simple objects and substances. When a label was associated with a simple object such as a kidney-shaped lump of wax, English-speaking children and adults unanimously interpreted the label as an object name, while Japanese-speaking children © Japanese Psychological Association 1999. 80 % shape response 40 American 20 Japanese 21/2 Age Adults Simple objects 80 % shape response 40 American 20 Japanese 2 21/2 Age 100 Substances American 80 Japanese 4 Adults % shape response 21/2 Age Adults Figure 2. The results of the Imai and Gentner study: Proportion of shape responses on (a) complex-object trials, (b) simpleobject trials and (c) substance trials. Constraint on word-learning constraints showed a random response. That is, Englishspeaking children did not hesitate to apply the whole-object assumption for this type of entity while Japanese children did not know whether it was appropriate to do so. Interestingly, when Japanese children heard a label associated with a nonsolid substance configured into a shape, they were not at all affected by the interesting shape, extending the label exclusively on the basis of material. This time, it was Englishspeaking children who had trouble determining whether the whole-object assumption should be applied. Universality, linguistic relativity, and the perceptual salience of entities What conclusions can we draw from these results? First, Quine’s view was bluntly refuted, since children as young as 2 years were able to appreciate the ontological distinction between objects and substances, and projected word meanings differently for the two kinds of entity, regardless of whether or not the grammatical categorization in their native language corresponded to the ontological class. This ontological knowledge successfully constrained the application of the whole-object assumption and the shape bias even though their native language did not provide them with syntactic cues. However, the cross-linguistic data also suggested that the linguistic structure of the speaker’s native language influenced the construal of individuation, especially when the perceptual “affordance” of the entity was weak and ambiguous. Perceptual properties are often very powerful indicators for categorization. Often, people can determine what category a given entity X belongs to just by looking at it, without knowing details of its nonapparent properties. However, the degree to which perceptual affordance on the part of entities invites humans to classify a given entity into a particular category (in this case, either an object class or a substance class) varies across different entities. With some entities, their perceptual properties are highly indicative of only one particular category, while with others, their perceptual properties do not point toward a particular categorization. In our case, the perceptual properties of complex objects (such as those used in our experiments) strongly suggest to us their preindividuated nature, while the perceptual affordance of simple objects is much lower, allowing us to construe them either as preindividuated objects or nonindividuated substances. English-speakers uniformly construed such simple-shape solid entities as individuated objects, while Japanese children originally had no conception of their status of individuation. Furthermore, Japanese speakers had no trouble construing the nonsolid substances as nonindividuals even though they were formed into complex, interesting shapes. However, this manipulation – making substances appear in complex shapes – made it difficult for English speakers to determine whether these entities should be construed as individuals or nonindividuals. As noted by the linguistic anthropologist Lucy (1992), shape is perhaps the best indicator for individualization. The distinct, complex shapes in which the nonsolid substances appeared may have led the English-speaking subjects to think that the label referred to a particular shape rather than the substance per se, even though they were fully aware that the standard entity could not be an object. These results tell us that there is an intricate interaction between internal constraints (i.e., biases we possess internally) and external constraints (i.e., constraints the world places on us). The internal constraints involve two types of bias, one of which is language-independent cognitive disposition, and the other a bias built up by language-specific grammatical categories. At the same time, our construals are also constrained by a third factor, which lies in the outside world: namely, the perceptual affordance of each entity by which humans are invited to make a particular categorization (see Malt, 1995). When a given entity’s perceptual properties strongly suggest object-ness or substance-ness, language-specific grammatical categorization plays a relatively small role in our classification behavior. It is when the entity’s perceptual affordance is weak that © Japanese Psychological Association 1999. M. Imai language-specific categorization influences our construal more directly.2 To summarize, the cross-linguistic research demonstrated that young Japanese children, just as their English-speaking counterparts (Hall, 1996; Soja et al., 1991), use wordlearning principles such as the whole-object assumption and the shape bias at the first stages of word learning and, furthermore, that they are able to control the application of these principles by using their ontological knowledge about objects and substances as well as by using the perceptual properties of the named entities. Constraints on the application of the shape bias/taxonomic assumption: Learning of proper names and nonbasic category terms We now know that Japanese children, from 2 years old onward, can appropriately restrict the application of the whole-object assumption and the shape bias even when their native language does not tell them whether a given noun should be interpreted as a count noun or mass noun. The next question is whether Japanese children successfully control the disposition to Interestingly, it is these types of entity whose construal was most greatly affected by syntax among English speakers. Mazuka and I (Imai & Mazuka, 1997; in preparation) investigated to what extent English count/ mass syntax would influence the construal of the three types of entity. In this study, the English-speaking 4-year-olds and adults were assigned either to the count-noun condition or to the mass-noun condition. In the count-noun condition, novel labels were presented as count nouns (e.g., “This is a dax.”) for all 12 trials. In the mass-noun condition, labels were presented as mass nouns (e.g., “This is some dax.”). We found that, when the label’s syntactic class severely conflicted with the target entity’s perceptual affordance (i.e., a label in the count-noun syntactic frame was associated with a nonsolid substance or noun in the mass-noun syntax was associated with a complex object), the subject’s performance became random. That is, the subjects could not replace the default construal with a different construal suggested by syntax. In contrast, for the simple objects, mass-noun syntax easily shifted the English speakers (both 4-year-olds and adults) from their default object construal to a substance construal. © Japanese Psychological Association 1999. extend labels to similar objects (i.e., the taxonomic assumption/the shape bias) in order to learn proper names and common names when a label is given to a discrete object. In learning proper names, this disposition must be suspended. In a sense, this problem seems to be more challenging for children than learning object names and substance names. The object/ substance distinction is largely detectable by the entity’s perceptual properties. Typical objects (such as the complex objects in Imai and Gentner’s study) are solid and have a distinct shape that is not easily changed just by touching. In contrast, nonsolid substances (such as sand) are not rigid. As mentioned in the earlier section, one can tell, just by its appearance, that the shape of a particular portion of sand is only transient and can be easily changed when someone touches it. In determining whether a novel noun is a proper noun or a common noun, no such perceptual information is available, since the referent of the word is identical. For Englishspeaking children, syntax can resolve this problem, and children can use this information from a very early age (Gelman & Taylor, 1984; Katz, Baker, & Macnamara, 1974; see also Bloom, 1994). Importantly, however, the use of syntactic cues interacts with children’s internal bias to interpret novel nouns as basic-level category names. On the one hand, it has been reported that the taxonomic assumption overrides syntactic information for unfamiliar objects. Hall (1991) introduced a novel word in a proper-name syntactic frame (“This is X”) for an unfamiliar, monster-like toy animal whose name had been unknown to the children. He found that the children were willing to extend the word to similar toy monsters in spite of the proper-noun syntax. On the other hand, for familiar objects, (English-speaking) children even as young as 2 years can sometimes use syntax to override the taxonomic assumption (and the mutual exclusivity assumption as well), interpreting the second label as a proper name (Gelman & Taylor, 1984). However, importantly, Gelman and Taylor also demonstrated that the proper-noun interpretation was Constraint on word-learning constraints constrained by semantic appropriateness: The children made a proper-noun interpretation when the labeled object was an animal but showed random performance when the label was given to an inanimate object such as a ball (see also Hall, 1994). From these findings, the following pictures emerge. Children are universally biased to assume that novel nouns are category names (in other words, the taxonomic assumption/ shape bias is universally used), and this bias tends to override syntactic information when the latter is in conflict with the former (Hall, 1991; Markman & Wachtel, 1988); but children can make use of cues from syntax when the newly labeled object already has a name, provided that the interpretation is semantically plausible. But how can Japanese children select the single most appropriate interpretation out of so many competing alternative interpretations? If a Japanese child hears a new noun in a typical sentence frame “Kore wa X desu,” it could be interpreted like any of the following English sentences: “This is an X” (object category name); “This is X” (property name); “This is X” (proper name); or “This is some portion of X” (substance name). However, we know that Japanese children would not make the last interpretation for complex objects (at least when an unfamiliar novel object is named). It is true that English syntax does not disambiguate all possible interpretations. For example, hearing a noun in the count-noun syntactic frame (such as “This is an X”), children may assume that the noun refers to a class of individuals, but the syntax does not tell children which level of lexical hierarchy the noun should be mapped on to (Hall & Waxman, 1993). In hearing a noun in a syntactic frame such as “This is X,” X could refer to either a particular individual or a property of the named object (Hall, 1994). Despite this, English count/mass syntax does significantly reduce the original vast search space: Englishspeaking children at least do not have to consider the possibility of noun X referring to a category in hearing “This is X,” or in the case of hearing “X” in the count-noun frame they can abandon the possibility that “X” may be a proper name or a property name. No such help is available for Japanese children from syntax. How do they get around this problem? To examine this question, Etsuko Haryu and I recently studied how Japanese 2-year-olds interpret novel labels that are associated with familiar and unfamiliar animals and inanimate objects (Imai & Haryu, 1998). Study 1: Interpretation of novel nouns for unfamiliar objects We first examined how Japanese children interpret novel nouns associated with unfamiliar objects. Two-year-old monolingual Japanesespeaking children were randomly assigned to the animal condition or to the inanimate condition. The structure of the stimuli and the procedure were identical across the two conditions. An unfamiliar object was named in a sentence frame something like “Kore wa neke desu,” where neke is the target noun. As mentioned above, the noun neke could be an object category name, proper name, property name, or substance name. It is impossible to infer whether the noun is a proper noun or a common noun from the structure of the sentence, although we know that, based on the results of Imai and Gentner (1997), the child would not think that the noun refers to a portion of the named object or to the material it is made of, since the labeled objects in this study were all “complex objects.” The named object was taken out of the child’s view after the naming session, and was then presented again with four other objects. The four objects included a subordinate-level item, a basic-level item, a superordinate-level item, and a distractor (see Figure 3).3 The subordinate item was identical to the original in shape, size, and material. When the original was a toy animal, the subordinate item was Note that, in this setup, the item allowing the interpretation of the word as a property, which is equivalent to Hall’s (1994) property-match item, was not included because selecting appropriate item(s) out of five objects was already very demanding, and this interpretation seemed unlikely given the results of Hall (1994) as well as Imai and Gentner (1997). © Japanese Psychological Association 1999. M. Imai Figure 3. (a) Sample stimulus set used in the animal condition and (b) sample stimulus set used in the inanimate condition in Imai and Haryu (1998). distinguishable from the original object by clothes and/or accessories (e.g., a hat, a ribbon, or hair band). For the inanimate object sets, the original and the subordinate item differed only in color. The basic-level item was very similar (but not identical) to the original in shape, but © Japanese Psychological Association 1999. was different from it in material, color, and/or size. The superordinate item had a very different appearance (both in shape and color) from the original but it came from the same superordinate category. A distractor item was drawn from a different ontological category (i.e., when Constraint on word-learning constraints the named object was a toy animal, then the distractor object was an inanimate object, and vice versa). The distractor was included so that we could detect whether the child was responding randomly. The five objects (the original and the four variations) were all placed in front of the child, and the location of each object relative to the child was counterbalanced. The child was asked to “neke o sagashite,” which could mean “find a neke/nekes/Neke/some neke.” The subject could select either a single object or multiple objects at one time. Since Japanese does not mark the singular/plural distinction, the instruction would not bias the child toward selecting only one or more than one object. The selected object(s) were put into a box, leaving the nonselected objects in front of the child, who was then asked whether there was any more neke there. This procedure was repeated until the child said “no” to the prompt. If children interpreted the noun as a proper name, they would be expected to select only the named object. If they interpreted the noun as a common noun, they should select multiple objects. In the latter case the selected objects should always include the target and the object(s) that were more similar to the original than the least similar object in the set. The shape bias predicts that children will extend the label up to the basic-level item (i.e., selecting the original, the subordinate item, and the basic-level item) but not to the superordinate item. There were three stimulus sets in each condition. When the child went through all three sets, the procedure was repeated. Each child thus received six trials. What pattern is predicted? One possible pattern is that Japanese children face difficulty in determining whether the noun refers to a particular individual or to a class of individuals. In this case, the difficulty may be greater in the animal condition than in the inanimate condition since proper-noun and common-noun interpretations are both possible for animals while the proper-noun interpretation is implausible for inanimate objects (Gelman & Taylor, 1984; Hall, 1994). Alternatively, Japanese children may interpret the noun as a common noun in either case, guided by the bias to assume that a label refers to a class of objects (i.e., the taxonomic assumption/shape bias). This second possibility would be consistent with Hall’s (1991) finding with English-speaking 2-year-olds. The child’s response in each trial was classified as one of five mutually exclusive response categories: proper-noun response, subordinate response, basic-level response, superordinate response, and unclassifiable response. For example, a child’s response was counted as a proper-noun response when only the original object was selected. Likewise, the response was counted as subordinate when the child selected both the original and the subordinate item, but no other objects. For each child, we counted the number of trials (a maximum of six) that were classified into each of the five response categories. The results were clear. Whether the novel label was given to an animal or an inanimate object, the 2-year-old Japanese children interpreted it as a common name. The children made common-noun interpretations (at the subordinate, basic, or superordinate level) more than 85% of the time for both conditions (animal, 87.5%; inanimate, 94.2%; see Table 1). Among the possible common-noun interpretations (i.e., subordinate, basic-level, superordinate responses), the basic-level interpretation was made most frequently (47.2% in the inanimate condition and 54.2% in the animal condition). This suggests that when children hear a novel noun associated with an unfamiliar object, either animate or inanimate, they assume by default that the noun refers to a kind of object rather than to a particular instance of the object, relying on shape similarity in determining the extension of the category. But what do children do when they hear a novel label associated with a familiar object, that is, one they already know the name of? This question was examined next. Study 2: Interpretation of novel nouns for familiar objects The structure of the stimulus sets was identical to that used in the above study except that the © Japanese Psychological Association 1999. M. Imai Table 1. Results of Studies 1 and 2 in Imai and Haryu (1998): Percentage of responses in each condition in 2-year-olds in the animal condition and inanimate condition Proper Subordinate Basic Superordinate Unclassifiable Study 1 (unfamiliar) Animal condition Inanimate condition Study 2 (familiar) Animal condition Inanimate condition 11.1 1.4 59.7 (0) 9.7 (0) 23.6 25 27.7 (0) 48.6 (2.6) 54.2 47.2 8.3 (77.8) 16.7 (83.3) 9.7 22 0 (22) 22.2 (13.9) 2.8 4.2 4.2 (0) 2.8 (0) Note. The percentages in parentheses indicate the proportion of response falling into each response type when familiar words were provided. named objects were now familiar. Novel labels were given to instances of bear, penguin and monkey in the animal condition. In the inanimate object condition, novel labels were given to instances of ball, cup, and spoon. As in Study 1, Japanese 2-year-olds were examined; half of them were assigned to the animal condition, and the other half were assigned to the inanimate object condition. To make sure that our stimuli were valid, we had another group of Japanese 2-year-olds classify the objects in each set using the familiar label – bouru (ball), koppu (cup), supuun (spoon) for the sets in the inanimate object condition; pengin (penguin), kuma (bear), saru (monkey) for the sets in the animal condition. For example, the experimenter said to the child, “kuma [bear] o [accusative participle] sagashite [find],” meaning “Find a bear/bears.” We expected the children to select the original, the subordinate item, and also the basic-level item, but not the remaining two items. The children were highly successful in this task (see the percentages given in parentheses in Table 1). What pattern is predicted in this study? Because children already know the basic-level names, the mutual exclusivity assumption may make it difficult for them to accept the novel labels. If this is the case, we might then expect the children to exhibit inconsistent behavior, often showing the “unclassifiable” response or failing to show a distinct preference in response patterns. © Japanese Psychological Association 1999. Note that the principle of contrast (Clark, 1987) does not predict difficulty in learning novel names for objects that already have a name. Rather, it predicts that children should think that the extension of the novel label does not exactly overlap with that of the extension of the basic-level category that they already know. In this case, children will face three possibilities: (a) the extension of the novel noun may be restricted to the named object; (b) the extension may be broader than the basic-level category; (c) the extension may be narrower than the basic-level category, but not restricted to the named object. For animals, all three are possible. For inanimate objects, the first possibility is implausible. The question now is whether young Japanese children have a systematic bias toward any particular solution. The results showed that even 2-year-old Japanese children did not have much difficulty in accepting novel labels given to the familiar objects: the proportion of responses that could not be categorized as any of the four consistent response types (i.e., “unclassifiable”) was very low in spite of the fact that the base probability for an “unclassifiable” response is much higher (28/32) than that for each of the four consistent response types, for which the base probability was only 1/32 each. Consistent with the principle of contrast, the children rarely interpreted a novel label as a complete synonym of a word they already knew. That is, they rarely showed a basic-level interpretation in either Constraint on word-learning constraints condition. Furthermore, the interpretation of novel nouns was greatly influenced by the named object’s animacy status. The children who heard a novel label associated with familiar animals interpreted the noun as a proper name 59.7% of the time (i.e., they chose the original object and no other), whereas those who heard a novel label for inanimate familiar objects made a proper-name response only 9.7% of the time (see the bottom half of Table 1). The children in the inanimate condition showed a subordinate interpretation (choosing the original and that most closely resembling it) most often (48.6%), and the proportions of other response types were all low. Constraints on the taxonomic assumption/ the shape bias and the mutual exclusivity assumption in learning proper names and nonbasic category terms The above research presents a fairly comprehensive picture of how Japanese children assign meanings to novel nouns. Japanese children as young as 2 years showed very reasonable, orderly behavior in word learning even though the grammar in their native language does not provide any cues for narrowing down multiple, competing possibilities for the meaning of a given noun. When a novel label is given to an object that does not yet have a label, children assume that the label is a name for an object category, whether the referent is an animal or an inanimate object. If the named object already has an established name, and if the object is an animal, children tend to interpret the label as a name for the particular individual rather than interpreting it as a name for a narrower or a broader category.4 However, when a novel label was given to an inanimate object, they did not interpret it as a name for a particular instance any more. Instead, they mapped the noun onto a narrower category, that is, a subordinate category. It is interesting that, for the novel labels for the familiar inanimate objects, the children rarely interpreted the novel noun as a superordinate category term, although the superordinate and subordinate interpretations were equally plausible. How should we account for this preference for the subordinate over the superordinate interpretation? Does this mean that the subordinate level is privileged relative to the superordinate level? I conjecture that this is not the case; rather, the answer may lie in children’s bias to extend nouns to objects of similar shape. In Imai and Haryu’s research, the subordinate item had identical shape to the original, whereas the superordinate item had a very different shape. In contrast, both the American children in Imai et al. (1994) and the Japanese children in Imai and Uchida (1995) willingly extended novel labels to objects of similar shape which were drawn from a different ontological class from the original (e.g., a banana and a feather). This suggests that the preference for the subordinate interpretation does not necessarily mean that children always assume that the second label for an inanimate object refers to the subordinate category rather than the superordinate category. Conclusions and implications: The mechanism of word learning In this paper, I have considered the question of how Japanese children deal with the hard problems of induction in word learning without much aid from syntax, aid which is available to children learning English. I have argued that, based on the empirical evidence from two series of studies, Japanese children do not go astray in the maze of induction problems in word learning, contrary to Quine’s concern. At a very early stage of word learning, Japanese children are able to get around the Quinian induction problems quite easily and learn new words very efficiently. But how? © Japanese Psychological Association 1999. However, this conclusion warrants some caution. It may not be animacy alone that induces the proper-noun interpretation. The fact that the toy animals wore clothes may have led the children to personify these animals (see Hall, 1994). If they hear a novel label for a familiar animal in the wild (e.g., neke for a real bear in a forest), it is unclear whether Japanese children would interpret the word to be the proper name of the animal or a nonbasic category name. In this respect, children are likely to utilize general knowledge of the world and pragmatic knowledge in addition to semantic knowledge in disambiguating the meaning of novel words. M. Imai I would like to propose that children are not only equipped with word-learning principles which guide them in mapping new words onto their meanings, but are also equipped with a system – we may call it a “meta-mechanism” – that enables them to control the application of the principles. I propose that it is this system as a whole, rather than each of the word-learning principles alone, that makes such efficient word learning possible in young children. In this system, multiple factors, including semantic/ ontological knowledge (e.g., Carey, 1997; Gelman & Taylor, 1984; Hall, 1994, 1996; Imai & Gentner, 1997; Jones, Smith, & Landau, 1991; Soja et al., 1991), perceptual salience on the part of labeled entities (e.g., Imai & Gentner, 1997), knowledge of how syntactic classes map onto conceptual classes (e.g., Bloom & Kelemen, 1995; Imai & Mazuka, 1997; Soja, 1992), knowledge of functions and other properties about the labeled entities (e.g., Kemler-Nelson, 1995; Kobayashi, 1997), and pragmatic knowledge such as ability to infer other people’s communicative intentions (e.g., Nelson, 1985; Tomasello, 1997), all contribute to constrain word meanings. More importantly, however, these different sources of constraint are appropriately coordinated so that children will not go astray when two or more factors are in conflict. For example, semantic knowledge of animals and inanimate objects and the syntactic cue (if available) do control the application of the taxonomic assumption so that children can learn proper names. On the other hand, results in the literature (Hall, 1991; Imai & Haryu, 1998; Markman & Wachtel, 1988) suggest that the suspension of the taxonomic assumption is difficult unless the labeled object is unfamiliar: when the syntactic information is in conflict with the taxonomic assumption for an unfamiliar object (i.e., a novel label for an unfamiliar object appears in the proper-noun syntactic frame), the taxonomic assumption seems to take precedence over the syntactic information. Thus, syntax, semantic knowledge, and the mutual exclusivity assumption control the application of the taxonomic assumption in an interactive fashion. © Japanese Psychological Association 1999. Furthermore, even when one of the important sources of constraint is unavailable in a particular language, children seem to find a way to compensate for the missing information. Although a very important source of information for constraining word meanings, that is, cues from syntax, is not available to Japanese children, they are able to learn word meanings as efficiently as children learning English and other languages with count/mass syntax by utilizing word-learning principles and other available sources of information. However, the fact that Japanese children can deal with the induction problems in word learning without the aid of syntax does not mean that syntactic information is not useful for word learning in children whose native language does provide cues from syntax. In fact, a number of studies suggest that syntax influences the construal of the named entity in English-speaking children (e.g., Bloom & Kelemen, 1995; Hall, 1994; Gelman & Taylor, 1984; Imai & Mazuka, 1997; Katz, Baker, & Macnamara, 1974; Soja, 1992; also see footnote 2). I believe that the implication from the Japanese data is that children are very flexible and are good at making use of whatever helpful information is available to them. If the syntax of their language provides useful information, they will become attentive to this information at a very early age, together with other types of linguistic and nonlinguistic constraint. If the syntax does not provide useful information, they then find a way to constrain and determine word meanings by other means, by establishing heuristics that efficiently coordinate available resources such as semantic knowledge, pragmatic/contextual knowledge, and perceptual cues from the world. Importantly, this also suggests that syntactic cues might be of a relatively weak, secondary status as a constraint, which can assist the process of constraining word meanings but which may not be indispensable (see also Haryu & Imai, 1999, for a relevant discussion). In closing, I would like to propose that future research in word learning needs to explore exactly how multiple sources of constraints, including word-learning principles and other linguistic and nonlinguistic factors, are Constraint on word-learning constraints coordinated, and how this coordinated system interacts with the linguistic structure of different languages. http://www.deepdyve.com/assets/images/DeepDyve-Logo-lg.png Japanese Psychological Research Wiley

Constraint on word‐learning constraints

Imai, Mutsumi
Japanese Psychological Research , Volume 41 (1) – Mar 1, 1999
Download PDF
16 pages

Loading next page...
 
/lp/wiley/constraint-on-word-learning-constraints-lRJEZqNMD8

References (0)

References for this paper are not available at this time. We will be adding them shortly, thank you for your patience.

Publisher
Wiley
Copyright
Japanese Psychological Association 1999
ISSN
0021-5368
eISSN
1468-5884
DOI
10.1111/1468-5884.00101
Publisher site
See Article on Publisher Site

Abstract

Abstract: Many researchers have agreed that word learning in young children is guided by so-called “word-learning principles.” However, these principles may make it difficult to learn a substantial part of the lexicon unless they are appropriately controlled. To learn proper names, the taxonomic assumption and/or the shape bias must be overridden; to learn names for substances, withdrawal of the whole-object assumption and/or the shape bias is required; and to learn lexical hierarchies, the mutual exclusivity assumption must be suspended. In certain languages, syntax can provide useful information in this situation. For example, if a novel noun is given to an object for which the name is known in the syntactic frame “This is X,” Englishspeaking children may assume the noun to be a proper noun, and this will help them override the taxonomic assumption. However, this information is not available to Japanese children, since the Japanese language does not have grammatical markers to flag the distinction between count nouns and mass nouns, or the distinction between proper nouns and common nouns. In this paper, I discuss how Japanese children get around this problem. Key words: noun acquisition, word-learning principles, constraint, syntactic cues, crosslinguistic comparison. It has been agreed among researchers that children do not face word learning unprepared. In other words, they are not led astray by the well known induction problems in inferring a meaning of a word (Quine, 1960). Instead , they constrain the potentially infinite possibilities in mapping a word onto its meaning (for detailed discussion of this, see Imai, 1997; Markman, 1989) by using internal principles which help them to map a word onto its meaning even at first exposure to it (e.g., Markman, 1989; Markman & Hutchinson, 1984). Among the proposed principles or constraints, the wholeobject assumption, the taxonomic assumption, and the mutual exclusivity assumption (e.g., Markman, 1989; Markman & Hutchinson, 1984; Markman & Wachtel, 1988) have attracted much attention (e.g., Golinkoff, Mervis, & HirshPasek, 1994; Hall, 1991, 1994; Imai, Gentner, & Uchida, 1994; Landau, Smith, & Jones, 1988; Merriman & Bowman, 1989; Waxman & Gelman, 1986). Although a majority of studies agree that these principles are in fact used by children from a very early age, there has been much debate with respect to the specific nature of each of these principles. For example, are they available prior to the onset of word learning and applied from the first word (Waxman & Markow, 1995)? Are they universally applied irrespective of any specific linguistic properties of the input language (Imai & Gentner, 1997)? A more serious problem that has remained unanswered, however, arises from the fact that all of these principles must be suspended in some circumstances and overcome in others. That is, although the word-learning principles such as the whole-object assumption, the taxonomic assumption/shape bias, and the mutual exclusivity assumption can greatly constrain © 1999 Japanese Psychological Association. Published by Blackwell Publishers Ltd, 108 Cowley Road, Oxford OX4 1JF, UK and 350 Main Street, Malden, MA 02148, USA. M. Imai the possible search space in mapping words on to concepts, they could also block the learning of a substantial portion of a vocabulary unless their applications are somehow controlled. For example, the whole-object assumption should not be applied when a child learns words for substances, such as water, sand, and sugar (Imai & Gentner, 1997; Soja, Carey, & Spelke, 1991). Learning names for specific individuals, that is, proper nouns, requires suspension of the taxonomic assumption (Hall, 1991; see also Woodward & Markman, 1997). The mutual exclusivity assumption must be overcome in order for a child to learn category names at different levels of the taxonomic hierarchy as well as names for particular individuals (Gelman & Taylor, 1984; Hall, 1991; Haryu & Imai, 1999; Taylor & Gelman, 1989; Waxman, Shipley, & Shepperson, 1991). Thus, for successful word learning, children must not only possess the word-learning principles but they also need heuristics that enable them to determine when to apply, suspend, or overcome the principles. In fact, the literature suggests that 2-year-olds’ vocabulary includes nonobject words such as names for substances and events as well as proper names (e.g., Nelson, Hampson, & Shaw, 1993). Children are able to learn basic-level object labels very rapidly by applying word-learning principles such as the whole-object assumption and the shape bias, but at the same time are also able to learn nonobject words and proper names by suspending them. How, then, do children control and constrain the application of the word-learning principles? It has been suggested in the literature that two sources of information are particularly important in this respect, namely, semantic (including ontological) knowledge, and syntactic cues for different form classes. In the case of English, it so happens that there is a high correlation between semantic (ontological) classes and syntactic classes of nouns. That is, individuated entities, typically solid objects, are mapped onto count nouns, while nonindividuated entities, typically substances, are mapped onto mass nouns. Furthermore, among the names for individuated entities, © Japanese Psychological Association 1999. names for particular individuals, that is, proper nouns, are syntactically distinguished from names for object kinds, that is, count nouns, in that count nouns, but not proper nouns, occur with determiners (cf., Bloom, 1994). Numerous studies have investigated the relative role of semantics and syntax in word learning (e.g., Bloom & Kelemen, 1995; Gelman & Taylor, 1984; Imai & Mazuka, 1997; Soja, 1992; Subramanyam & Landau, 1995). These studies suggest that both semantic knowledge and syntax influence word learning, and that there is an intricate interplay between the two. For example, Gelman and Taylor (1984) manipulated the linguistic form class (proper noun vs. common noun) and conceptual type of the referent (animal vs. inanimate object) in examining how 2-year-old English-speaking children interpret novel nouns. They found that when a noun was given to a stuffed toy in the proper-noun syntax, children restricted the word to the named animal, interpreting the noun as a name for a particular individual. In contrast, when a noun was presented in the count-noun syntax, they were willing to extend it to another animal of the same kind, suggesting that they interpreted the noun as a name for a particular kind of animal rather than as a name for a particular individual. However, the 2-year-olds did not show this pattern when the referent was an inanimate object. When a novel noun was introduced in the proper-noun syntax, they did not restrict the word to the named toy. Rather their performance became random. This suggests that 2-year-old English-speaking children know that it does not make sense for inanimate objects to have proper names. Thus, Gelman and Taylor’s study showed that both syntax and semantics influence children as young as 2 years of age in how they assign meanings to novel nouns. Given this intricate relationship between semantics and syntax, some researchers, P. Bloom in particular (e.g., Bloom, 1994), have gone so far as to claim that children do not need word-learning principles such as Constraint on word-learning constraints the whole-object assumption or the taxonomic assumption. All they need is knowledge of distinct ontological classes with respect to individuation (individuals, kinds of individuals, kinds of portions), and of how these conceptual classes map on to syntax. But what happens when such syntactic cues are completely lacking in a particular language? In Japanese, all nouns, including count nouns, mass nouns, and proper nouns, are treated the same in syntax. That is, there is no syntactic marker distinguishing the names of objects, substances, and proper nouns. Furthermore, the syntax does not mark the singular/plural distinction, either. Thus, the following English expressions, “This is a dax,” “Those are daxes,” “This is some dax,” “This is dax,” “This is Dax” are all translated into a single expression, “Kore [This] wa [topic/subject marker] dax desu [is].” In other words, when one hears “Kore wa dax desu” without seeing the named entity, there is absolutely no way of inferring whether dax refers to a single object, multiple objects, a substance/portion, a property (such as color), or a particular individual. The question then is how Japanese children determine whether a novel noun is a name for an object category, or a name for a substance, or a name for a particular individual. Are they totally at a loss? Or are they somehow able to constrain and coordinate the application of multiple word-learning principles even without cues from syntax? In this paper, I will argue that this is in fact the case. Two lines of my research are described, and I explore how young Japanese children control and coordinate wordlearning principles for efficient word learning. I will then discuss the implications of my work on Japanese children’s word learning for the nature of children’s word learning across languages. assumption and the shape bias.1 Dedre Gentner and I asked whether Japanese children are able to extend novel nouns in an ontologically correct fashion, extending a noun associated with an object on the basis of shape but extending a noun associated with a substance on the basis of material identity, without blindly applying the whole-object assumption and the shape bias (Imai & Gentner, 1997). Inspired by the seminal work of Soja et al. (1991) on the role of conceptual understanding of the ontological difference between objects and substances in early word learning, we wished to know whether Japanese children would demonstrate the same appreciation of the ontological distinction in their word learning, even though their native language lacks syntactic cues that flag the two distinct ontological classes. This question is important since there has been a debate on the origin of the ontological distinction between objects and substances. The philosopher Quine conjectured that children come to know the ontological differences between objects and substances only after learning the count/mass syntax (Quine, 1969). Challenging Quine, Soja et al. (1991), who are developmental psychologists, argued that children possess this knowledge perhaps at birth, and certainly before the acquisition of count/ mass syntax. The Japanese language is an ideal medium to test which of the two positions are correct, Constraints on the application of the whole-object assumption and the shape bias by ontological knowledge I will first discuss how Japanese children constrain the application of the whole-object In both of the studies described in this paper (Imai & Gentner, 1997; Imai & Haryu, 1998), the shape bias and the taxonomic assumption are indistinguishable. The shape bias and the taxonomic assumption yield similar consequences for words denoting basic-level objects, since at the basic level, category members are highly similar in shape (Rosch, 1978). Thus we consider the shape bias without considering the taxonomic assumption separately. The taxonomic assumption and the shape bias yield different results when category membership and shape similarity conflict. At early stages of word learning, children’s assumption about the possible extension of a word meaning is better characterized as a shape bias rather than a taxonomic assumption, but this early shape bias gradually shifts into a taxonomic assumption (Imai et al., 1994). © Japanese Psychological Association 1999. M. Imai since Japanese grammar does not make a distinction between these two ontological classes. A strong interpretation of Quine yields a prediction that it is not possible for Japanese children to learn the ontological distinction between objects and substances. In this case, Japanese children may not possess the wholeobject assumption and the shape bias at all, or, even if they do, they may not be able to exert the ontologically appropriate application of these principles. In contrast, if Soja et al.’s universal ontology view is correct, Japanese children should possess the ontological distinction between objects and substances, and be able to utilize this knowledge in inferring word meanings. We thus extended Soja et al.’s paradigm cross-linguistically, and compared monolingual Japanese-speaking and English-speaking children and adults. A cross-linguistic study of word-meaning projections for objects and substances Our subjects were monolingual Japanesespeaking and English-speaking 2-year-olds, 21⁄ 2-year-olds, 4-year-olds, and adults. The paradigm was a forced-choice triad task, where a novel label was given to the standard entity, and the subject was asked to determine which of the two alternatives could share the same label as the originally named entity. One of the alternatives, the shape alternative, had the same shape, but was made out of a different material. The other alternative, the material alternative, consisted of a portion of the standard entity. The stimulus materials included three types of entity as the referents of novel labels: complex objects, simple objects, and substances. The first type, which we characterized as complex objects, were real objects (artifacts) that had fairly complex shapes and distinct functions, although the functions were unknown to the children (e.g., a T-shaped plumbing fixture). The second type, which were called simple objects, were made of a solid substance such as clay or wax, and were formed into a very simple shape such as the shape of a kidney. These entities were somewhat between © Japanese Psychological Association 1999. the complex objects and substances, allowing a double construal. They could be construed as individuated objects, since they met the physical properties of so-called “Spelke objects.” If you picked up a part of the entity and moved it, the whole thing moved. At the same time, it could be construed as a lump of some substance as well. You could continue breaking the entity into pieces until each piece was no longer visible to the naked eye, and a micropiece of it was still the “same” substance as the original. The third type of entity, the substances, were nonsolid substances such as sand or hair gel. However, unlike typical substances in the real world, they were arranged in somewhat distinct, interesting shapes when presented (see Figure 1). For both language groups, novel labels were embedded in a syntactic frame that would not reveal the noun’s count/mass status. As mentioned earlier, in the case of Japanese, this exactly reflects the language’s structure. The label was uttered in a sentence, such as “Kore wa dax desu,” and dax could be interpreted as a name for a type of object, a name for a substance, a name for a particular individual, or even a name for a property. In English, we had to ensure that the noun’s count/mass status would not be revealed. The experimenter said, “Look at this dax! Can you find the tray that also has the dax on it?” In this task, if subjects interpreted the novel noun of ambiguous syntactic status as a name for a type of object, they would extend the label on the basis of shape identity; if they interpreted it as a substance name, they would extend it on the basis of material identity, suspending the whole-object assumption and the shape bias. To repeat, whether the noun was an object name or a substance name could not be detected from the syntax in this task. Thus we should be able to assume that the pattern of word-meaning projection for a given entity, that is, whether the child extends the noun on the basis of shape or material, reflects the child’s construal of that entity. Although the experimental results had several important aspects, in this paper I will focus on the application of the whole-object Constraint on word-learning constraints Complex objects Substances Porcelain lemon juicer Hand cream Wooden lemon juicer Pieces of porcelain Simple objects Hair gel Blobs of hand cream Kidney-shaped wax Kidney-shaped plaster Figure 1. Pieces of wax Sample materials used in the Imai and Gentner (1997) study. © Japanese Psychological Association 1999. M. Imai Complex objects 100 assumption and the shape bias (for other aspects of the study, see Imai & Gentner, 1997; Imai & Mazuka, 1997). Are the whole-object assumption and the shape bias universally used in young children’s word learning? If so, are Japanese children able to restrict the application of these word-learning principles to the appropriate kind of entity, that is, preindividuated objects? Further, is there any influence of the input language in the application of these principles, given that Japanese children could “constrain” the application of the whole-object assumption and the shape bias by ontological knowledge about objects and substances? Universal application of the whole-object constraint and the shape bias. The answers to the first two questions raised above are affirmative. Japanese children, just as Englishspeaking American children, extended a novel label on the basis of shape when the label was associated with an object of complex shape. In both language groups, there was a large difference in the proportion of shape responses between the complex-object trials and the substance trials, and this difference held across all age groups (see Figure 2). These results suggest that children as young as 2 years apply the whole-object assumption and the shape bias independent of the structure of the input language. Furthermore, they also suggest that children universally know from the very early stages of word learning that objects and substances belong to fundamentally distinct ontological classes. Application of the whole-object assumption and the shape bias is constrained by this universal ontological knowledge, as suggested by Soja et al. (1991) and Soja (1992) (see also Hall, 1996). Influence of the language in applying the principles. However, as to the third question, the answer is also affirmative but in a restricted sense. That is, we found a substantial crosslinguistic difference for simple objects and substances. When a label was associated with a simple object such as a kidney-shaped lump of wax, English-speaking children and adults unanimously interpreted the label as an object name, while Japanese-speaking children © Japanese Psychological Association 1999. 80 % shape response 40 American 20 Japanese 21/2 Age Adults Simple objects 80 % shape response 40 American 20 Japanese 2 21/2 Age 100 Substances American 80 Japanese 4 Adults % shape response 21/2 Age Adults Figure 2. The results of the Imai and Gentner study: Proportion of shape responses on (a) complex-object trials, (b) simpleobject trials and (c) substance trials. Constraint on word-learning constraints showed a random response. That is, Englishspeaking children did not hesitate to apply the whole-object assumption for this type of entity while Japanese children did not know whether it was appropriate to do so. Interestingly, when Japanese children heard a label associated with a nonsolid substance configured into a shape, they were not at all affected by the interesting shape, extending the label exclusively on the basis of material. This time, it was Englishspeaking children who had trouble determining whether the whole-object assumption should be applied. Universality, linguistic relativity, and the perceptual salience of entities What conclusions can we draw from these results? First, Quine’s view was bluntly refuted, since children as young as 2 years were able to appreciate the ontological distinction between objects and substances, and projected word meanings differently for the two kinds of entity, regardless of whether or not the grammatical categorization in their native language corresponded to the ontological class. This ontological knowledge successfully constrained the application of the whole-object assumption and the shape bias even though their native language did not provide them with syntactic cues. However, the cross-linguistic data also suggested that the linguistic structure of the speaker’s native language influenced the construal of individuation, especially when the perceptual “affordance” of the entity was weak and ambiguous. Perceptual properties are often very powerful indicators for categorization. Often, people can determine what category a given entity X belongs to just by looking at it, without knowing details of its nonapparent properties. However, the degree to which perceptual affordance on the part of entities invites humans to classify a given entity into a particular category (in this case, either an object class or a substance class) varies across different entities. With some entities, their perceptual properties are highly indicative of only one particular category, while with others, their perceptual properties do not point toward a particular categorization. In our case, the perceptual properties of complex objects (such as those used in our experiments) strongly suggest to us their preindividuated nature, while the perceptual affordance of simple objects is much lower, allowing us to construe them either as preindividuated objects or nonindividuated substances. English-speakers uniformly construed such simple-shape solid entities as individuated objects, while Japanese children originally had no conception of their status of individuation. Furthermore, Japanese speakers had no trouble construing the nonsolid substances as nonindividuals even though they were formed into complex, interesting shapes. However, this manipulation – making substances appear in complex shapes – made it difficult for English speakers to determine whether these entities should be construed as individuals or nonindividuals. As noted by the linguistic anthropologist Lucy (1992), shape is perhaps the best indicator for individualization. The distinct, complex shapes in which the nonsolid substances appeared may have led the English-speaking subjects to think that the label referred to a particular shape rather than the substance per se, even though they were fully aware that the standard entity could not be an object. These results tell us that there is an intricate interaction between internal constraints (i.e., biases we possess internally) and external constraints (i.e., constraints the world places on us). The internal constraints involve two types of bias, one of which is language-independent cognitive disposition, and the other a bias built up by language-specific grammatical categories. At the same time, our construals are also constrained by a third factor, which lies in the outside world: namely, the perceptual affordance of each entity by which humans are invited to make a particular categorization (see Malt, 1995). When a given entity’s perceptual properties strongly suggest object-ness or substance-ness, language-specific grammatical categorization plays a relatively small role in our classification behavior. It is when the entity’s perceptual affordance is weak that © Japanese Psychological Association 1999. M. Imai language-specific categorization influences our construal more directly.2 To summarize, the cross-linguistic research demonstrated that young Japanese children, just as their English-speaking counterparts (Hall, 1996; Soja et al., 1991), use wordlearning principles such as the whole-object assumption and the shape bias at the first stages of word learning and, furthermore, that they are able to control the application of these principles by using their ontological knowledge about objects and substances as well as by using the perceptual properties of the named entities. Constraints on the application of the shape bias/taxonomic assumption: Learning of proper names and nonbasic category terms We now know that Japanese children, from 2 years old onward, can appropriately restrict the application of the whole-object assumption and the shape bias even when their native language does not tell them whether a given noun should be interpreted as a count noun or mass noun. The next question is whether Japanese children successfully control the disposition to Interestingly, it is these types of entity whose construal was most greatly affected by syntax among English speakers. Mazuka and I (Imai & Mazuka, 1997; in preparation) investigated to what extent English count/ mass syntax would influence the construal of the three types of entity. In this study, the English-speaking 4-year-olds and adults were assigned either to the count-noun condition or to the mass-noun condition. In the count-noun condition, novel labels were presented as count nouns (e.g., “This is a dax.”) for all 12 trials. In the mass-noun condition, labels were presented as mass nouns (e.g., “This is some dax.”). We found that, when the label’s syntactic class severely conflicted with the target entity’s perceptual affordance (i.e., a label in the count-noun syntactic frame was associated with a nonsolid substance or noun in the mass-noun syntax was associated with a complex object), the subject’s performance became random. That is, the subjects could not replace the default construal with a different construal suggested by syntax. In contrast, for the simple objects, mass-noun syntax easily shifted the English speakers (both 4-year-olds and adults) from their default object construal to a substance construal. © Japanese Psychological Association 1999. extend labels to similar objects (i.e., the taxonomic assumption/the shape bias) in order to learn proper names and common names when a label is given to a discrete object. In learning proper names, this disposition must be suspended. In a sense, this problem seems to be more challenging for children than learning object names and substance names. The object/ substance distinction is largely detectable by the entity’s perceptual properties. Typical objects (such as the complex objects in Imai and Gentner’s study) are solid and have a distinct shape that is not easily changed just by touching. In contrast, nonsolid substances (such as sand) are not rigid. As mentioned in the earlier section, one can tell, just by its appearance, that the shape of a particular portion of sand is only transient and can be easily changed when someone touches it. In determining whether a novel noun is a proper noun or a common noun, no such perceptual information is available, since the referent of the word is identical. For Englishspeaking children, syntax can resolve this problem, and children can use this information from a very early age (Gelman & Taylor, 1984; Katz, Baker, & Macnamara, 1974; see also Bloom, 1994). Importantly, however, the use of syntactic cues interacts with children’s internal bias to interpret novel nouns as basic-level category names. On the one hand, it has been reported that the taxonomic assumption overrides syntactic information for unfamiliar objects. Hall (1991) introduced a novel word in a proper-name syntactic frame (“This is X”) for an unfamiliar, monster-like toy animal whose name had been unknown to the children. He found that the children were willing to extend the word to similar toy monsters in spite of the proper-noun syntax. On the other hand, for familiar objects, (English-speaking) children even as young as 2 years can sometimes use syntax to override the taxonomic assumption (and the mutual exclusivity assumption as well), interpreting the second label as a proper name (Gelman & Taylor, 1984). However, importantly, Gelman and Taylor also demonstrated that the proper-noun interpretation was Constraint on word-learning constraints constrained by semantic appropriateness: The children made a proper-noun interpretation when the labeled object was an animal but showed random performance when the label was given to an inanimate object such as a ball (see also Hall, 1994). From these findings, the following pictures emerge. Children are universally biased to assume that novel nouns are category names (in other words, the taxonomic assumption/ shape bias is universally used), and this bias tends to override syntactic information when the latter is in conflict with the former (Hall, 1991; Markman & Wachtel, 1988); but children can make use of cues from syntax when the newly labeled object already has a name, provided that the interpretation is semantically plausible. But how can Japanese children select the single most appropriate interpretation out of so many competing alternative interpretations? If a Japanese child hears a new noun in a typical sentence frame “Kore wa X desu,” it could be interpreted like any of the following English sentences: “This is an X” (object category name); “This is X” (property name); “This is X” (proper name); or “This is some portion of X” (substance name). However, we know that Japanese children would not make the last interpretation for complex objects (at least when an unfamiliar novel object is named). It is true that English syntax does not disambiguate all possible interpretations. For example, hearing a noun in the count-noun syntactic frame (such as “This is an X”), children may assume that the noun refers to a class of individuals, but the syntax does not tell children which level of lexical hierarchy the noun should be mapped on to (Hall & Waxman, 1993). In hearing a noun in a syntactic frame such as “This is X,” X could refer to either a particular individual or a property of the named object (Hall, 1994). Despite this, English count/mass syntax does significantly reduce the original vast search space: Englishspeaking children at least do not have to consider the possibility of noun X referring to a category in hearing “This is X,” or in the case of hearing “X” in the count-noun frame they can abandon the possibility that “X” may be a proper name or a property name. No such help is available for Japanese children from syntax. How do they get around this problem? To examine this question, Etsuko Haryu and I recently studied how Japanese 2-year-olds interpret novel labels that are associated with familiar and unfamiliar animals and inanimate objects (Imai & Haryu, 1998). Study 1: Interpretation of novel nouns for unfamiliar objects We first examined how Japanese children interpret novel nouns associated with unfamiliar objects. Two-year-old monolingual Japanesespeaking children were randomly assigned to the animal condition or to the inanimate condition. The structure of the stimuli and the procedure were identical across the two conditions. An unfamiliar object was named in a sentence frame something like “Kore wa neke desu,” where neke is the target noun. As mentioned above, the noun neke could be an object category name, proper name, property name, or substance name. It is impossible to infer whether the noun is a proper noun or a common noun from the structure of the sentence, although we know that, based on the results of Imai and Gentner (1997), the child would not think that the noun refers to a portion of the named object or to the material it is made of, since the labeled objects in this study were all “complex objects.” The named object was taken out of the child’s view after the naming session, and was then presented again with four other objects. The four objects included a subordinate-level item, a basic-level item, a superordinate-level item, and a distractor (see Figure 3).3 The subordinate item was identical to the original in shape, size, and material. When the original was a toy animal, the subordinate item was Note that, in this setup, the item allowing the interpretation of the word as a property, which is equivalent to Hall’s (1994) property-match item, was not included because selecting appropriate item(s) out of five objects was already very demanding, and this interpretation seemed unlikely given the results of Hall (1994) as well as Imai and Gentner (1997). © Japanese Psychological Association 1999. M. Imai Figure 3. (a) Sample stimulus set used in the animal condition and (b) sample stimulus set used in the inanimate condition in Imai and Haryu (1998). distinguishable from the original object by clothes and/or accessories (e.g., a hat, a ribbon, or hair band). For the inanimate object sets, the original and the subordinate item differed only in color. The basic-level item was very similar (but not identical) to the original in shape, but © Japanese Psychological Association 1999. was different from it in material, color, and/or size. The superordinate item had a very different appearance (both in shape and color) from the original but it came from the same superordinate category. A distractor item was drawn from a different ontological category (i.e., when Constraint on word-learning constraints the named object was a toy animal, then the distractor object was an inanimate object, and vice versa). The distractor was included so that we could detect whether the child was responding randomly. The five objects (the original and the four variations) were all placed in front of the child, and the location of each object relative to the child was counterbalanced. The child was asked to “neke o sagashite,” which could mean “find a neke/nekes/Neke/some neke.” The subject could select either a single object or multiple objects at one time. Since Japanese does not mark the singular/plural distinction, the instruction would not bias the child toward selecting only one or more than one object. The selected object(s) were put into a box, leaving the nonselected objects in front of the child, who was then asked whether there was any more neke there. This procedure was repeated until the child said “no” to the prompt. If children interpreted the noun as a proper name, they would be expected to select only the named object. If they interpreted the noun as a common noun, they should select multiple objects. In the latter case the selected objects should always include the target and the object(s) that were more similar to the original than the least similar object in the set. The shape bias predicts that children will extend the label up to the basic-level item (i.e., selecting the original, the subordinate item, and the basic-level item) but not to the superordinate item. There were three stimulus sets in each condition. When the child went through all three sets, the procedure was repeated. Each child thus received six trials. What pattern is predicted? One possible pattern is that Japanese children face difficulty in determining whether the noun refers to a particular individual or to a class of individuals. In this case, the difficulty may be greater in the animal condition than in the inanimate condition since proper-noun and common-noun interpretations are both possible for animals while the proper-noun interpretation is implausible for inanimate objects (Gelman & Taylor, 1984; Hall, 1994). Alternatively, Japanese children may interpret the noun as a common noun in either case, guided by the bias to assume that a label refers to a class of objects (i.e., the taxonomic assumption/shape bias). This second possibility would be consistent with Hall’s (1991) finding with English-speaking 2-year-olds. The child’s response in each trial was classified as one of five mutually exclusive response categories: proper-noun response, subordinate response, basic-level response, superordinate response, and unclassifiable response. For example, a child’s response was counted as a proper-noun response when only the original object was selected. Likewise, the response was counted as subordinate when the child selected both the original and the subordinate item, but no other objects. For each child, we counted the number of trials (a maximum of six) that were classified into each of the five response categories. The results were clear. Whether the novel label was given to an animal or an inanimate object, the 2-year-old Japanese children interpreted it as a common name. The children made common-noun interpretations (at the subordinate, basic, or superordinate level) more than 85% of the time for both conditions (animal, 87.5%; inanimate, 94.2%; see Table 1). Among the possible common-noun interpretations (i.e., subordinate, basic-level, superordinate responses), the basic-level interpretation was made most frequently (47.2% in the inanimate condition and 54.2% in the animal condition). This suggests that when children hear a novel noun associated with an unfamiliar object, either animate or inanimate, they assume by default that the noun refers to a kind of object rather than to a particular instance of the object, relying on shape similarity in determining the extension of the category. But what do children do when they hear a novel label associated with a familiar object, that is, one they already know the name of? This question was examined next. Study 2: Interpretation of novel nouns for familiar objects The structure of the stimulus sets was identical to that used in the above study except that the © Japanese Psychological Association 1999. M. Imai Table 1. Results of Studies 1 and 2 in Imai and Haryu (1998): Percentage of responses in each condition in 2-year-olds in the animal condition and inanimate condition Proper Subordinate Basic Superordinate Unclassifiable Study 1 (unfamiliar) Animal condition Inanimate condition Study 2 (familiar) Animal condition Inanimate condition 11.1 1.4 59.7 (0) 9.7 (0) 23.6 25 27.7 (0) 48.6 (2.6) 54.2 47.2 8.3 (77.8) 16.7 (83.3) 9.7 22 0 (22) 22.2 (13.9) 2.8 4.2 4.2 (0) 2.8 (0) Note. The percentages in parentheses indicate the proportion of response falling into each response type when familiar words were provided. named objects were now familiar. Novel labels were given to instances of bear, penguin and monkey in the animal condition. In the inanimate object condition, novel labels were given to instances of ball, cup, and spoon. As in Study 1, Japanese 2-year-olds were examined; half of them were assigned to the animal condition, and the other half were assigned to the inanimate object condition. To make sure that our stimuli were valid, we had another group of Japanese 2-year-olds classify the objects in each set using the familiar label – bouru (ball), koppu (cup), supuun (spoon) for the sets in the inanimate object condition; pengin (penguin), kuma (bear), saru (monkey) for the sets in the animal condition. For example, the experimenter said to the child, “kuma [bear] o [accusative participle] sagashite [find],” meaning “Find a bear/bears.” We expected the children to select the original, the subordinate item, and also the basic-level item, but not the remaining two items. The children were highly successful in this task (see the percentages given in parentheses in Table 1). What pattern is predicted in this study? Because children already know the basic-level names, the mutual exclusivity assumption may make it difficult for them to accept the novel labels. If this is the case, we might then expect the children to exhibit inconsistent behavior, often showing the “unclassifiable” response or failing to show a distinct preference in response patterns. © Japanese Psychological Association 1999. Note that the principle of contrast (Clark, 1987) does not predict difficulty in learning novel names for objects that already have a name. Rather, it predicts that children should think that the extension of the novel label does not exactly overlap with that of the extension of the basic-level category that they already know. In this case, children will face three possibilities: (a) the extension of the novel noun may be restricted to the named object; (b) the extension may be broader than the basic-level category; (c) the extension may be narrower than the basic-level category, but not restricted to the named object. For animals, all three are possible. For inanimate objects, the first possibility is implausible. The question now is whether young Japanese children have a systematic bias toward any particular solution. The results showed that even 2-year-old Japanese children did not have much difficulty in accepting novel labels given to the familiar objects: the proportion of responses that could not be categorized as any of the four consistent response types (i.e., “unclassifiable”) was very low in spite of the fact that the base probability for an “unclassifiable” response is much higher (28/32) than that for each of the four consistent response types, for which the base probability was only 1/32 each. Consistent with the principle of contrast, the children rarely interpreted a novel label as a complete synonym of a word they already knew. That is, they rarely showed a basic-level interpretation in either Constraint on word-learning constraints condition. Furthermore, the interpretation of novel nouns was greatly influenced by the named object’s animacy status. The children who heard a novel label associated with familiar animals interpreted the noun as a proper name 59.7% of the time (i.e., they chose the original object and no other), whereas those who heard a novel label for inanimate familiar objects made a proper-name response only 9.7% of the time (see the bottom half of Table 1). The children in the inanimate condition showed a subordinate interpretation (choosing the original and that most closely resembling it) most often (48.6%), and the proportions of other response types were all low. Constraints on the taxonomic assumption/ the shape bias and the mutual exclusivity assumption in learning proper names and nonbasic category terms The above research presents a fairly comprehensive picture of how Japanese children assign meanings to novel nouns. Japanese children as young as 2 years showed very reasonable, orderly behavior in word learning even though the grammar in their native language does not provide any cues for narrowing down multiple, competing possibilities for the meaning of a given noun. When a novel label is given to an object that does not yet have a label, children assume that the label is a name for an object category, whether the referent is an animal or an inanimate object. If the named object already has an established name, and if the object is an animal, children tend to interpret the label as a name for the particular individual rather than interpreting it as a name for a narrower or a broader category.4 However, when a novel label was given to an inanimate object, they did not interpret it as a name for a particular instance any more. Instead, they mapped the noun onto a narrower category, that is, a subordinate category. It is interesting that, for the novel labels for the familiar inanimate objects, the children rarely interpreted the novel noun as a superordinate category term, although the superordinate and subordinate interpretations were equally plausible. How should we account for this preference for the subordinate over the superordinate interpretation? Does this mean that the subordinate level is privileged relative to the superordinate level? I conjecture that this is not the case; rather, the answer may lie in children’s bias to extend nouns to objects of similar shape. In Imai and Haryu’s research, the subordinate item had identical shape to the original, whereas the superordinate item had a very different shape. In contrast, both the American children in Imai et al. (1994) and the Japanese children in Imai and Uchida (1995) willingly extended novel labels to objects of similar shape which were drawn from a different ontological class from the original (e.g., a banana and a feather). This suggests that the preference for the subordinate interpretation does not necessarily mean that children always assume that the second label for an inanimate object refers to the subordinate category rather than the superordinate category. Conclusions and implications: The mechanism of word learning In this paper, I have considered the question of how Japanese children deal with the hard problems of induction in word learning without much aid from syntax, aid which is available to children learning English. I have argued that, based on the empirical evidence from two series of studies, Japanese children do not go astray in the maze of induction problems in word learning, contrary to Quine’s concern. At a very early stage of word learning, Japanese children are able to get around the Quinian induction problems quite easily and learn new words very efficiently. But how? © Japanese Psychological Association 1999. However, this conclusion warrants some caution. It may not be animacy alone that induces the proper-noun interpretation. The fact that the toy animals wore clothes may have led the children to personify these animals (see Hall, 1994). If they hear a novel label for a familiar animal in the wild (e.g., neke for a real bear in a forest), it is unclear whether Japanese children would interpret the word to be the proper name of the animal or a nonbasic category name. In this respect, children are likely to utilize general knowledge of the world and pragmatic knowledge in addition to semantic knowledge in disambiguating the meaning of novel words. M. Imai I would like to propose that children are not only equipped with word-learning principles which guide them in mapping new words onto their meanings, but are also equipped with a system – we may call it a “meta-mechanism” – that enables them to control the application of the principles. I propose that it is this system as a whole, rather than each of the word-learning principles alone, that makes such efficient word learning possible in young children. In this system, multiple factors, including semantic/ ontological knowledge (e.g., Carey, 1997; Gelman & Taylor, 1984; Hall, 1994, 1996; Imai & Gentner, 1997; Jones, Smith, & Landau, 1991; Soja et al., 1991), perceptual salience on the part of labeled entities (e.g., Imai & Gentner, 1997), knowledge of how syntactic classes map onto conceptual classes (e.g., Bloom & Kelemen, 1995; Imai & Mazuka, 1997; Soja, 1992), knowledge of functions and other properties about the labeled entities (e.g., Kemler-Nelson, 1995; Kobayashi, 1997), and pragmatic knowledge such as ability to infer other people’s communicative intentions (e.g., Nelson, 1985; Tomasello, 1997), all contribute to constrain word meanings. More importantly, however, these different sources of constraint are appropriately coordinated so that children will not go astray when two or more factors are in conflict. For example, semantic knowledge of animals and inanimate objects and the syntactic cue (if available) do control the application of the taxonomic assumption so that children can learn proper names. On the other hand, results in the literature (Hall, 1991; Imai & Haryu, 1998; Markman & Wachtel, 1988) suggest that the suspension of the taxonomic assumption is difficult unless the labeled object is unfamiliar: when the syntactic information is in conflict with the taxonomic assumption for an unfamiliar object (i.e., a novel label for an unfamiliar object appears in the proper-noun syntactic frame), the taxonomic assumption seems to take precedence over the syntactic information. Thus, syntax, semantic knowledge, and the mutual exclusivity assumption control the application of the taxonomic assumption in an interactive fashion. © Japanese Psychological Association 1999. Furthermore, even when one of the important sources of constraint is unavailable in a particular language, children seem to find a way to compensate for the missing information. Although a very important source of information for constraining word meanings, that is, cues from syntax, is not available to Japanese children, they are able to learn word meanings as efficiently as children learning English and other languages with count/mass syntax by utilizing word-learning principles and other available sources of information. However, the fact that Japanese children can deal with the induction problems in word learning without the aid of syntax does not mean that syntactic information is not useful for word learning in children whose native language does provide cues from syntax. In fact, a number of studies suggest that syntax influences the construal of the named entity in English-speaking children (e.g., Bloom & Kelemen, 1995; Hall, 1994; Gelman & Taylor, 1984; Imai & Mazuka, 1997; Katz, Baker, & Macnamara, 1974; Soja, 1992; also see footnote 2). I believe that the implication from the Japanese data is that children are very flexible and are good at making use of whatever helpful information is available to them. If the syntax of their language provides useful information, they will become attentive to this information at a very early age, together with other types of linguistic and nonlinguistic constraint. If the syntax does not provide useful information, they then find a way to constrain and determine word meanings by other means, by establishing heuristics that efficiently coordinate available resources such as semantic knowledge, pragmatic/contextual knowledge, and perceptual cues from the world. Importantly, this also suggests that syntactic cues might be of a relatively weak, secondary status as a constraint, which can assist the process of constraining word meanings but which may not be indispensable (see also Haryu & Imai, 1999, for a relevant discussion). In closing, I would like to propose that future research in word learning needs to explore exactly how multiple sources of constraints, including word-learning principles and other linguistic and nonlinguistic factors, are Constraint on word-learning constraints coordinated, and how this coordinated system interacts with the linguistic structure of different languages.

Journal

Japanese Psychological Research – Wiley

Published: Mar 1, 1999

There are no references for this article.