Review of How Children Learn the Meanings of Words
Date: Fri, 31 May 2002 09:51:24 -0400
From: Katharine Beals
Subject: Bloom (2002) How Children Learn the Meanings of Words
Bloom, Paul (2002) How Children Learn the Meanings of Words. MIT Press, paperback ISBN 0-262-52329-9, vii+300pp, $19.95, A Bradford book.
[The hardback edition ISBN 0-262-02469-1 (2000) was reviewed at
Katharine Beals, full time stay-at-home mom
In How Children Learn the Meanings of Words, Paul Bloom argues that word learning occurs not through mental faculties dedicated to this task, but through various abilities that exist for other, not specifically linguistic, purposes.
He begins by reviewing the mysteries of word learning, citing Quine's "Gavagai!" riddle, in which a linguist immersing himself in an unfamiliar language overhears this outburst after a rabbit scurries by. Quine points out that there is an infinity of possible meanings for "gavagai". It might refer only to a part of the rabbit, or to a specific type of rabbit, or to a more general category of animal, or to some property the rabbit has, or to a time slice of the rabbit that exists only while "gavagai" is being uttered. It might not be a single word, or a word at all. And even if "gavagai" does specifically refer to the whole rabbit as such, how do we know that its actual meaning doesn't encompass one of an infinite number of bizarre but logically possible generalizations, for instance "rabbits-but only on Tuesdays; otherwise carrots."
Augmenting the mystery is empirical evidence about how, and how many, words are learned. Estimates show that American children, by the time they graduate from high school, have acquired at least 60,000 "minimal free forms," or words and idioms whose meanings cannot be predicted by their morphological components. But most of the word meanings children learn are not explicitly demonstrated to them; many words, like verbs and abstract nouns, are typically not, or simply cannot be, spoken at the same time that their referents are observable; and even in the case of words denoting concrete objects, between 30% and 50% of the time the child is not attending to the referent of the word as it is spoken-when asked "Do you want a cookie?" she may be looking at her parent's face. Nor do children in many cultures receive any explicit feedback about their own uses of words.
The first word learning tool Bloom discusses is fast mapping. This involves the ability to grasp, fairly quickly, "aspects of the meaning of a new word on the basis of a few incidental exposures, without any explicit training." (p. 26). Bloom notes that mapping speech to current circumstances is not limited to word learning, or to children, but also applies when we acquire factual information. He also debunks, through an examination of longitudinal data, the notion that a critical period for word-learning occurs during childhood, wherein a child's first few words are followed by a vocabulary spurt. What differences there are between how children and adults learn words stem, rather, from a gradual maturation of general learning mechanisms-memory, attention, conceptual awareness, ability to recognize a speaker's referential intentions-as well as our increasing familiarity with the language, which means that there are, over time, more context clues out there, but fewer new words to learn.
Bloom next addresses how children map words to their intended referents. Emphasizing again how rarely the utterance of a word co-occurs with its intended referent, and how poorly other animals that are capable of associative learning acquire words, he debunks the popular notion that word learning emerges from statistical correlations between words and their contexts of use. Does this show that humans have a unique word-learning mechanism? No, for there is another ability, unique to humans but not specific to word learning, that we use in mapping words to their intended referents: our Theory of Mind, or ability to infer the intentions of others. When a child hears an adult use a new word, studies show, he checks to see what the adult is looking at rather than assuming that the adult is referring to whatever he himself has been attending to. Autistic children, whose core deficit is an absent or defective Theory of Mind, make the wrong assumption; their vocabularies are often much smaller than normal, and riddled with precisely this kind of associative error.
Two other mapping abilities that have been treated as specific to word learning can be derived from Theory of Mind: the principle of lexical contrast, which tells a child that a new word probably does not have the same meaning as a word he already knows, and the principle of the linguistic sign, which tells him that the words he hears others use are bidirectional symbols that he himself can use in the same way. Bloom views lexical contrast as a principle about speakers rather than about words, namely that speakers tend not to use terms interchangeably, and as a consequence of our naïve theories of human psychology. He points out that this assumption applies in any interpersonal communication system, not just one involving single words, and doesn't just stem from a general preference for one-to-one mappings, for children will allow that two different speakers might describe a given object in different ways.
The child's recognition that linguistic signs are bidirectional, a notion that other intelligent animals never master, stems from her more general ability to infer a person's goal from his behavior and to reason that she herself can achieve the same goal if she successfully imitates this behavior. As Bloom puts it (p. 76), "Once a child believes that the adult's use of the word DOG was used with the intent to refer to a dog, she can infer that if she herself has the same intent (to refer to a dog), then she could use the same means (saying "dog") to satisfy this goal."
Having addressed this half of how children map words to their intended referents, Bloom turns to how they learn what a word's possible referents are, given what the speaker is looking at. In other words, how can they reasonably suppose that "Gavagai!" in Quine's example might refer to the whole rabbit? As many studies show, babies, like adults, see the world as composed of whole objects, i.e. cohesive stretches of matter, and deem all other properties as secondary. This object bias is apparent before children learn their first words, and is not specific to word learning, for it also helps them track and count objects. Children, thus, are predisposed to view a new word as referring to whatever whole object the speaker is looking at, rather than, say, its color or materials.
This creates another mystery: how do children learn words for these other properties? As Bloom points out, a number of circumstances can override the object bias. Sometimes there aren't any object-like candidates around-for example when the speaker refers to a non-solid substance like water. In other cases the principle of lexical contrast applies: if a new word is used in reference to an object that the child already knows a word for, he may conclude that it refers to one of the object's properties. As the child becomes more familiar with syntax, he may also rule out object reference on the basis of a word's part of speech.
What about entities that aren't objects? Some of these, while lacking the core property of bounded cohesion, still have some object-like properties. Object parts, like eyes and legs, are internally connected, and some can move independently. Collections, like flocks or families, are spatially bounded, and can appear cohesive if, like schools of fish, they move in concert. Also bounded and connected are negative spaces like holes; if we shift from space to time, so are sounds and actions. But Bloom argues that not everything can be viewed as a metaphorical extension of concrete objects. Sometimes what parses motion and matter and their metaphorical extensions into individuals is, rather, our Theory of Mind, or our "understanding of goal, function, and intent." (p. 114). This accounts for our individuation of entities like conferences, fights, and parties, and more abstractly, chapters, stories and jokes. It explains why we view a bikini as a single entity, since "it is created and used for a single purpose." (p. 112).
Having sketched out the second half of the mapping of words to intended referents, Bloom turns to how children deduce from this mapping the actual meanings of words, or range of potential reference. He begins with two word categories at either extreme. Personal pronouns, he suggests, have a referential range so vast that, as studies indicate, children can learn them only by attending to the conversations of others. This is particularly true of the second person pronoun: "Only when a child hears other people called YOU can she... reasonably infer that it is not her name." (p. 125). Once again, thus, Theory of Mind is key, and, accordingly, autistic children find pronouns especially difficult. Proper names, with their much narrower ranges of reference, also involve Theory of Mind. Young children know that things conceived of as having mental states, or as being personal pets, can have proper names even if they are inanimate. Also key is lexical contrast: if the child already knows the common noun, a new word that refers to the same thing may be its proper name.
Since different objects generally get different proper names, learning these also means appreciating object identity. Underlying this is a sensitivity to spatio-temporal continuity. This sensitivity, in turn, is not specific to word learning, but, as with our parsing the world into objects, also applies when we track things, count them, and record their individual histories--in the case of people, who owes you what, and who has demonstrated what kind of personality.
Bloom turns next to the concepts and categories that underlie the meanings of other words, particularly nouns, verbs and adjectives. Here he argues against the popular notion that concepts and categories emerge from a "perceptual similarity space," modified through experience, wherein clusters of objects (those that share many observable properties) are assigned the same category. This account meshes nicely with the tendency of children to classify solid objects by shape. But it fails to explain how abstract concepts, with no observable properties, arise, or how people, children and adults alike, classify things not just by their superficial properties, but, as studies show, also according to presumed essential properties that may not be observable. The ability to classify objects by criteria other than superficial appearance, Bloom points out, is highly adaptive, and not just for language learning, because it permits classifications by chemical or biological properties that, in turn, allow predictions about how particular members of a class behave. This raises the possibility that such essentialist thinking is innate.
Unlike many philosophers who assume that only natural kinds have essential properties, Bloom argues that we also view artifacts this way. Their essences aren't chemical or biological, but social and psychological. A chair is not defined solely by its appearance (what about a beanbag chair or a doll's chair?), or by its function (one can sit on the floor) or even by its intended function (what about a chair that was never intended to be sat in?). Rather, whether adults view something as a chair depends on the "creator's intent and how it relates to the design of the object," where the intent is "to create something that falls into that kind." (p. 161), and some studies suggest that even young children use this criterion. Representations of objects, e.g. drawings, are also categorized by children and adults alike according to their creators' intentions, as Bloom shows in a following chapter specifically devoted to how we name representations.
Next, Bloom discusses the role of linguistic context in word learning, focusing on syntactic clues. Part of speech can filter possible word meanings; so does the distinction between count and mass nouns, and the correspondence between the syntactic structure of predicates and their conceptual structure-- i.e., how many entities participate in the action a verb denotes. But for specific meanings that are hard to otherwise deduce (for example that of "nightmare"), more general linguistic context clues, not specifically syntactic, are key.
Bloom turns next to how we learn number words, addressing why so much time elapses between when children understand both the concepts of small numbers such as twoness, threeness, and the principle of counting, and when they grasp the precise meanings of larger numbers. He proposes two discrete mechanisms: one an innate, non-recursive "accumulator" that underlies the awareness that babies, and other animals, have of small numbers (and is also involved in measuring duration), and the other a generative, recursive system of number that children grasp only after sufficient exposure to the similarly recursive counting system of their language. This is the first example Bloom entertains of language influencing conceptual development. Young children vary cross-linguistically in how well they understand number, in a way meshes with differences among counting systems. But this sort of influence is not specific to language: with reference to notation systems for music and chess, Bloom observes that "an inspection of the properties of a symbolic system that refers to the domain can lead to insights about the domain itself." (p. 238).
Aside from this, however, the main way in which language affects concept formation, Bloom argues, is as a vehicle for conveying ideas. While vocabulary and linguistic structure can call attention to different properties, ways of construing something, or ways of carving up space and time that are already available to us, there is little evidence, he says, that language per se actually structures the way we view reality. One exception may be within the domain of subordinate kinds: learning the names of types of wine might help you "organize the 'flux of impressions' that you experience into discrete categories and to appreciate the ways in which wines differ." (p. 253). But in general, Bloom concludes, the abilities that many scholars have claimed are products of language learning-the object bias, essentialist thinking, aspects of the Theory of Mind-are, rather, its prerequisites.
How Children Learn the Meanings of Words is a well-written, even-handed account, full of thought-provoking insights. It is generally well-organized, though in such a way as to recapitulate, somewhat confusingly, certain issues: e.g., when Bloom criticizes a competing theory after reviewing all the evidence for his own, or when he treats objects and individuals, which occasion many of the same issues, in separate chapters. His careful review of experimental methodology and results, of competing claims, and of possible objections to his own accounts, however, make Bloom's conclusions highly convincing. Where there are gaps in data, ambiguous results, or open questions, we can generally trust him to say so. For all the praise that scholars heap on falsifiability in the abstract, it is rare to see someone so clearly demonstrate, as Bloom does in his concluding chapter, how his particular theory may be proved wrong.
I was, however, struck with some unaddressed questions. One of the first facts that Bloom notes about language learning is that it involves internalizing a series of arbitrary facts that is longer than any other list we memorize (at least 60,000 word-meaning correspondences for the average high school graduate). It's hard to believe that we are even capable of retaining an equivalent number of other arbitrary facts-consider, for example, the extreme difficulty that most of us have with all the names and dates in history or paleontology. Furthermore, as Bloom notes, "while the recovery of most arbitrary facts is slow and hard, access to words and their meanings is fast and effortless." (p. 6). But then where does the almost universal ability to memorize and rapidly recall so many facts about word meaning come from, especially if, as Bloom's account implies, it is not specific to language learning?
Bloom's account of essentialist thinking in categorization also raises questions. Given that essences are often hidden, how are we able to categorize according to them and thus learn words that denote these categories? One possibility is that children are verbally informed of an individual's essential properties, as they are in the study of dinosaur categorization Bloom discusses (where the property in question is "cold blood"). But is this typically how children learn about category membership? That is, in order to understand what is and isn't a dinosaur, must the child be told, for each type of dinosaur, that it has cold blood? The dinosaur study, however, demonstrates an alternative clue: here children learn that the triceratops has cold blood by hearing it described as a dinosaur. But this result conflicts with Bloom's later claim that words don't affect category creation-unless he intends a distinction between category creation and category membership, and would allow that words can affect our learning of the latter. In any case, there appears to be a complex, poorly understood interaction between words, essences and categories that underlies the learning of each.
As the mother of an autistic child with impaired language, I was particularly struck by Bloom's insights about Theory of Mind and word learning. My son does not notice what others are looking at, does not attend to their conversations, and doesn't infer referential intent unless it is extremely obvious. He has, accordingly, required an explicit demonstration of the meaning of each of his many words. Pronouns remain elusive. As for proper names, I have long noticed that he overextends these according to personal appearance, as if he sees people as belonging to different sub-categories of person. My assumption had been that, because he thinks of people as mere objects rather than as sentient beings, and because individual identity is generally more important with the latter than the former, he simply doesn't notice people as individuals. Bloom's analysis of how normal children learn proper names raises another possibility: perhaps my son individuates people more than we think he does, but doesn't understand the proper use of proper names. If proper names are specifically reserved for sentient entities, then those who don't recognize anything as sentient, if they use proper names, use them not as such but as common nouns.
Further study of differences among autistic children would help illuminate what Bloom takes to be one of the biggest factors, and mysteries, in word learning: Theory of Mind. Do autistic children with different degrees of Theory of Mind impairment have different patterns of word learning? How aware are they of the principle of lexical contrast and of the bidirectionality of the linguistic sign? To what degree do they classify artifacts by the creator's intent? And what do they take to be the function of proper names?
| ABOUT THE REVIEWER:
ABOUT THE REVIEWER Katharine Beals received her Ph.D. in linguistics from the University of Chicago in 1995. From 1995 to 2000 she worked as a Senior Software Engineer with the Natural Language Group at Unisys. She is currently at home with her baby daughter and at work on a book about her deaf, autistic son, which explores such issues as language modality, cochlear implants, and language and consciousness in autistic people.