The acquisition of word meaning involves, at minimum, the process of mapping concepts onto sounds or signs. These mappings differ across languages — siy is the phonological string associated with the meaning “if” in French, “yes” in Spanish, and “apprehend visually” in English — and hence cannot be innate. Since antiquity, scholars interested in word learning have assumed that the source of information for acquiring the meaning of new words is the child’s observation of the linguistic and extralinguistic contexts of their use. Recent theorizing, as we shall describe, contributes new insights and findings that argue that the process is dramatically more complex than thought in the past.
Children understand a few words starting at about 9 months, with first spoken words typically appearing at about 12-14 months. These milestones are identical for spoken and signed languages (Petitto 1992; see SIGN LANGUAGES and SIGN LANGUAGE AND THE BRAIN). The acquisition of new words is at first quite slow, averaging about 0.3 words per day, but this rate gradually increases, eventually to about 15 new words per day in the school years (Anglin 1993). Some children show a sudden spurt in vocabulary growth between about 20 to 30 months, and there is in general a high correlation between syntactic and lexical development (Fenson et al. 1994). The onset of SYNTAX also coincides with a change in the character of the vocabulary: Earliest words, even in very different linguistic communities, are mainly animal sounds, “routines” like bye-bye, and common nouns, with a relatively high proportion of object names. Verbs, adjectives, and function words are rare (compared to their proportions in maternal usage; Fenson et al. 1994; Gentner 1982).
The vocabulary of a monolingual high school graduate is in the neighborhood of 80,000 words (Miller 1996). This number is impressive — 80,000 arbitrary sound/meaning pairings is a lot to learn and store in memory — but word learning is impressive for other reasons as well. Quine (1960) gives the example of hearing a word, “gavagai,” in an unknown language under the most transparent of circumstances; let us say, only and always while viewing a rabbit. A first thought is that the listener would be warranted in supposing that this word means “rabbit.” But Quine points out that gavagai has a logical infinity of possible meanings, including rabbit tails or legs, rabbit shapes, colors, or motions, temporal rabbit-slices, and even undetached rabbit parts. Because all of these conjectures for the meaning are consistent with the listener’s experience, how can he or she zoom in on the single interpretation “rabbit”? (See RADICAL INTERPRETATION.) In actual fact, real children are apparently faced with just this problem, yet they seem to converge just about unerringly on the adult interpretation — that the word refers to the whole animal. But so saying leaves the problem raised by Quine unanswered.
Matters get worse. Contrary to what is often assumed, words are typically not presented in such “transparent circumstances.” Adult speech to young children even in middle-class American homes is frequently about the past or the future, when the word’s referents often are not in view (Beckwith, Tinker, and Bloom 1989). In some cultures, there is no explicit labeling of objects, and yet children have no problem learning object names (Schieffelin 1985). Blind children, whose observational opportunities are limited compared to the normal case, acquire word meanings — even the meanings of color words and verbs of perception — at about the same rate as sighted children (Landau and Gleitman 1985). And philosophers such as Plato have noted the puzzles that arise for the learning of abstract terms, such as those for numbers or ideal geometrical forms. (Consider also terms that describe mental states, such as “idea” and “know.”) In sum, the sorts of words that children must learn and the conditions under which they must learn them suggest that word learning cannot succeed solely by raw observation of the co-occurring world.
The solution to these puzzles must involve attributing certain powers to very young children. Some of these are conceptual. Surely children enter the language-learning situation equipped with natural ways of interpreting many things, properties, and events in the world around them. These “natural ways” include perceiving objects and even rabbits, but it is likely that they do not include perceiving undetached rabbit parts. In consequence, the task of early word learning ordinarily comes down to connecting these preexisting conceptions with phonetic sequences. In support of this, consider the speed and effectiveness of early word learning. Children grasp aspects of the meanings of new words with very few exposures, without training or feedback and without ostensive naming. This process has been dubbed “fast mapping” (Carey 1978). For instance, if a 3-year-old hears one object out of ten being referred to casually as “a koba,” over a month later she will know that this object is a koba (Markson and Bloom 1997). Children under the age of two can fast map new nouns (Waxman and Markow 1995), and the meanings of these early acquired words seem to be the same as they are for adults (Huttenlocher and Smiley 1987). This suggests that word learning is indeed supported by a preexisting conceptual repertoire. This position is bolstered by research with prelinguistic infants that shows that they possess a rich understanding of objects, actions, and other ontological kinds (e.g., Spelke 1994).
But even if children have the required conceptual structure for word-meaning acquisition, making the mappings to the right phonetic sequences remains to be explained. For example, very often in the child’s immediate environment “open” is uttered when nothing is opening, and often there is no utterance of “open” when something is saliently opening. How then do children make the connection between “open” and opening? Quite surprisingly, part of the solution is that even infants under two years of age will not passively associate all and only new sounds with all and only newly observed objects or actions. Rather, they regularly inspect the situation co-occurring with a new word to find out what the speaker intended to refer to when he or she used that word. For example, if 18-month-olds hear a novel word while they are playing with a novel toy, they will assume that the word names that toy only if the speaker is also attending to it: if the speaker is looking at a different object, children will spontaneously follow her line-of-regard and assume that the object she is looking at is the referent of the new word (Baldwin 1991). This pragmatic understanding might also underlie certain expectations about how the LEXICON works, such as the assumption that different words should not have the same meaning (Clark 1985; though see Woodward and Markman 1997 for a different perspective). And children’s understanding of the actual meanings of many words — particularly for artifacts, such as “toy” or “clock,” but also for certain collections such as “family” and “army” — might require an appreciation of the goals and motivations of others (Bloom 1996).
Finally, much of word learning results from the child’s emerging appreciation of properties of language itself. As noted above, there is a strong relationship between the onset of syntax and the nature and development of word learning. In many regards (see SYNTAX, ACQUISITION OF), young children understand certain properties of language-specific grammar and balk when these are violated by adult speakers (Shipley, Smith, and Gleitman 1969). The ability to take into account more than one word in the adult sentence offers significant new clues to the word-learning child. One important cue comes from identification of a word’s grammatical category, a process called “syntactic bootstrapping.” For example, children who hear “This is ZAV” expect the word to refer to a specific individual, as with a proper name like “Fred” (Katz, Baker, and Macnamara 1974). Children who hear “a ZAV” expect it to refer to a kind of individual, such as “dog”; those who hear “some ZAV” expect it to refer to a substance name such as “water” (Brown 1957; Bloom and Kelemen 1995). And children hearing “John ZAVS Bill” expect the word to have a meaning similar to that of “hit,” whereas those who hear “John and Bill ZAV” expect it to have a meaning similar to that of “stand” (Gleitman 1990; Naigles 1990) — that is, they can make inferences about the number and type of arguments that a new predicate encodes for. Ultimately, the view that structure can be informative for word learning is one that commits theorists to the view that there are links between syntax and semantics to which young learners are privy very early in their language development (e.g., Gleitman 1990; Pinker 1989).
This brings us back to a finding mentioned earlier — the large proportion of object names in children’s very early vocabularies. One explanation that has been offered for this property of early lexicons is that object categories are conceptually simpler than categories such as parts and actions (Gentner 1982). Another proposed solution is that there are special default biases in word learning that guide children towards an object-kind interpretation of new words (Woodward and Markman 1997). A focus on syntactic development suggests a third possibility: no matter how conceptually sophisticated they are, learners who have no access to syntactic cues may be able to acquire only words whose referents are relatively “concrete,” those whose referents can be directly observed in the extralinguistic environment. In support of this, Gillette et al. (1997) found that adults shown videotaped mother-child interactions with the audio turned off and with beeps used in place of words (that is, adults who were deprived of syntactic support) acquired only “concrete” terms, typically object names — just like presyntactic infants.
In sum, there are three recent discoveries that have increased our understanding of word learning. First, a range of ontological categories, including objects, are available to very young children. Second, in addition to being conceptual creatures like us, children are intentional creatures like us: far from being at the mercy of passive associations when words occur in extralinguistic contexts, they are sensitive to subtle cues to a speaker’s referential intentions. Third, two findings about the form-meaning relations that children are acquiring may help explain how they perform their pyrotechnic learning feats — the increasing linguistic understanding of the SYNTAX-SEMANTICS INTERFACE; and the growing evidence that young children use cues at this level to constrain the interpretation of novel words.