How Children Use Input to Acquire a Lexicon

Child Development, March/April 2002, Volume 73, Number 2, Pages 418–433 How Children Use Input to Acquire a Lexicon Erika Hoff and Letitia Naigles The contributions of social processes and computational processes to early lexical development were evaluated. A re-analysis and review of previous research cast doubt on the sufficiency of social approaches to word learning. An empirical investigation of the relation of social–pragmatic and data-providing features of input to the productive vocabulary of sixty-three 2-year-old children revealed benefits of data provided in mother– child conversation, but no effects of social aspects of those conversations. The findings further revealed that the properties of data that benefit lexical development in 2-year-olds are quantity, lexical richness, and syntactic complexity. The nature of the computational mechanisms implied by these findings is discussed. An integrated account of the roles of social and computational processes to lexical development is proposed. INTRODUCTION The process of acquiring a lexicon is clearly a process of learning from experience, and the relevant experience must be conversational interaction, because that is the context in which exposure to language occurs. It is not yet clear, however, just how word-learning benefits from participating in or overhearing conversations. What do children find in conversation that is useful to word learning, and what is the nature of the word-learning mechanism that makes those particular things useful? Two types of answers have been proposed. The first focuses on the social–pragmatic aspects of conversation and the social–cognitive abilities of children. According to this view children figure out the meaning of the words they hear to a substantial degree by inferring the speaker’s attentional focus and communicative intent (Akhtar & Tomasello, 2000; Baldwin, 2000). The routinized and jointly engaged nature of the conversations that children experience support this process by making the speaker’s communicative intentions transparent (Bruner, 1974/1975; Pinker, 1984; Tomasello, 1990, 2001). The second answer focuses on the data-providing aspects of conversation, arguing that the lexical content and syntactic structure of the utterances themselves, along with the accompanying nonlinguistic context, provide considerable information that children use in figuring out word meaning (Carey, 1978; Gillette, Gleitman, Gleitman, & Lederer, 1999; Gleitman, 1990; Siskind, 1996). These two proposals are not mutually exclusive. Recent work acknowledges that multiple sources of information must contribute to word learning and seeks to build an integrative account of how word learning occurs (e.g., Akhtar & Tomasello, 2000; L. Bloom, 1993, 2000; P. Bloom, 2000; Hollich, HirshPasek, & Golinkoff, 2000). The purpose of this study was to add to this effort by evaluating the roles of social–pragmatic and data-providing functions of conversation on both the theoretical and empirical levels. The Social–Pragmatic View of the Role of Input Early formulations of the social–pragmatic proposal argued that the recurrent social interactions between mother and child establish each participant’s intentions throughout a given routinized activity, allowing the child to predict “where the adult’s attention is currently focused and where it is likely to be focused next. Therefore, any language the adult may use in such a context is likely to be immediately meaningful to the child” (Tomasello & Todd, 1983, p. 199). Additionally, if, in nonroutinized interaction, mothers talk about the aspect(s) of the activity that the child is focused on, then the meaning(s) the child is harboring should be consistently expressed, with the result that “mothers who follow their children’s leads in determining the topics of conversation may help their children’s language learning by increasing the likelihood that their children will be able to construct semantic representations of the sentences they hear” (Hoff-Ginsberg, 1987, p. 147). Thus, the social– pragmatic argument is that by virtue of either routinization or maternal attentiveness, children often know what their mothers are saying without understanding the language, and they can use that nonlinguistically acquired knowledge to figure out the meaning of the language they hear. More recently, the social–pragmatic proposal has focused on the social–cognitive abilities and inclinations of children (Akhtar & Tomasello, 2000; Baldwin, 2000; P. Bloom, 2000). There is evidence that children are not at the mercy of adults’ following their attentional focus for word meaning to be made transparent © 2002 by the Society for Research in Child Development, Inc. All rights reserved. 0009-3920/2002/7302-0006 Hoff and Naigles because children have the ability to discern their mothers’ communicative intentions. According to this view, word learning begins once children understand others as intentional agents, assume that there is some communicative intention behind the vocalizations others make, and successfully figure out what those communicative intentions are. In both social–pragmatic accounts, mutual engagement or joint attention provides support for word learning, but in the early formulation the social–pragmatic skill resides in the minds of the adults who structure the learning environment, whereas in the more recent formulation the social–pragmatic skill that contributes to word learning is located in the minds of the children who do the learning. Empirical support for the notion that word learning is aided when the learner and speaker are mutually engaged is plentiful. Thirteen-month-old infants whose mothers are more verbally responsive during toy play demonstrate earlier onset of their first word and of their first 50 words in both comprehension and production, and this benefit of verbal responsiveness is specific to language development (TamisLeMonda, Bornstein, Baumwell, & Damast, 1996; Tamis-LeMonda, Bornstein, Kahana-Kalman, Baumwell, & Cyphers, 1998). That is, maternal responsiveness to child vocalizations predicts child language outcomes, and responsiveness to play predicts measures of children’s play (Tamis-LeMonda et al., 1996). There is evidence of even more specific links between mother–child interaction and subsequent word learning. For example, children who are part of mother– child dyads that engage in greater amounts of joint interaction have larger production vocabularies between 12 and 18 months of age than children who engage in less joint interaction with their mothers (Tomasello & Todd, 1983). Mothers of normally developing children more frequently make reference to objects currently in their children’s focus of attention than do mothers of children who are developing language slowly (Harris, Jones, Brookes, & Grant, 1986). Children whose mothers engage in more directives that specifically follow their children’s focus of attention (as opposed to lead it, or just describe it) when the children are 13 months old have production vocabularies that include more nouns and more words at 22 months (Akhtar, Dunham, & Dunham, 1991). Twelve-month-olds whose interactions with their mothers include more joint engagement and maternal follow-in subsequently manifest larger comprehension vocabularies at 15 months of age (Carpenter, Nagell, & Tomasello, 1998). Fifteen- to 21-month-olds whose joint attention episodes with their mothers include considerable maternal verbal follow-ins also 419 are the ones who have the larger production vocabularies, assessed via a checklist (Tomasello & Farrar, 1986). In an experimental setting in which the match between input and children’s attention was manipulated, 17-month-olds were more likely to learn words that label their current focus of attention than words that label something they are not attending to (Tomasello & Farrar, 1986). In sum, it is clear from a substantial body of naturalistic evidence and one experimental demonstration that when mothers structure children’s experiences so that input is responsive to the children’s verbalizations and matches children’s attentional focus, vocabulary development benefits. There is also evidence that children contribute to establishing the joint engagement that benefits word learning. There are direct relations between 14-monthold children’s joint attentional behaviors and their mothers’ sensitivity (Laakso, Poikkeus, Katajamaki, & Lyytinen, 1999); and individual differences in 14month-olds’ abilities to follow another into a joint attentional state predict later language development (Laakso et al., 1999; Mundy & Gomes, 1998). These relations between very young children’s joint attentional skills and subsequent language development are most likely mediated by the effects of children’s skill on the mutual engagement they achieve. Other experimental evidence suggests that after 18 months, children have social–cognitive abilities that make them less dependent on such mutual engagement. At 18 months, children seem to know that if the speaker and listener are not attending to the same thing, the speaker is more likely to be talking about what the speaker is attending to than about what the listener is attending to, and children use that information to guide word learning (Baldwin, 1993; Hollich et al., 2000). Children at 24 months of age have demonstrated even more powerful abilities to use clues to speakers’ communicative intentions as sources of information about word meaning. For example, Tomasello and Barton (1994) found that telling 2-year-old children “I’m going to hoist Big Bird,” and then doing something but saying “Woops,” resulted in the children not taking “hoist” as a label for the action demonstrated. Presented with the same sequence of events, but with “There” replacing “Woops,” children took “hoist” as the label for the action. There have been similar effects of speakers’ indications of their intentions on object label learning (see also Golinkoff, Hirsh-Pasek, & Hollich, 1999; Tomasello, 2001; Woodward & Markman, 1999). This evidence leaves no doubt that children as young as 12 months have social skills that enable them to participate in mutually engaged social interactions (Akhtar et al., 1991; Carpenter et al., 1998; 420 Child Development Tomasello & Todd, 1983), that by 18 months children can discern speakers’ intentions even in the absence of mutual engagement (e.g., Baldwin, 1993; Tomasello & Barton, 1994), and that these abilities support word learning. Even the strongest social–pragmatic view allows that social information alone is insufficient to account for word learning (e.g, Akhtar & Tomasello, 2000), however, thus raising the question of what other sources of information children use. A complete account of lexical development requires specifying just what contribution social–pragmatic information makes to word learning, when in the course of development these contributions are made, and where else in their experience children find information about the meaning of the words they hear. In the next sections we begin to explore the parameters of the role of social-pragmatic information in word learning in three ways: (1) we reconsider the implications of previous findings of correlations between indices of mutual engagement and children’s vocabulary development, (2) we analyze the task of word learning for the potential contribution of social-pragmatic information, and (3) we consider evidence regarding the degree to which children’s conversations are actually characterized by mutual understanding. The Correlational Findings, Reconsidered Not all positive relations between social–pragmatic support and language development are necessarily evidence that children find information about word meaning by being provided with or by figuring out the communicative intentions of their conversational partner. The most general positive correlations may merely reflect that social–pragmatic support is an index of supportive parenting and vocabulary growth is an index of healthy development. After all, Monnot (1999) found that the degree to which mothers used characteristics of infant-directed speech in talk to their 3- to 4-month-old infants was positively correlated with infant weight gain, yet no one would argue that infants find calories in the fundamental frequency of their mothers’ speech. Other correlations, such as those observed by Tamis-LeMonda et al. (1996) and Bornstein, Haynes, and Painter (1998) may reflect the fact that mothers who are responsive to their children’s verbalizations benefit language development in general by encouraging further verbal interaction, and the greater information provided by this greater amount of interaction is what benefits lexical development. The findings of benefits of maternal follow-in that are specific to the learning of particular words could simply be benefits of the temporal contiguity of words and children’s attention, rather than a reflec- tion of any socially based process in the mind of the child. Finally, the correlations between maternal social– pragmatic support and lexical development rely frequently on mother-generated checklists or diaries as the measure of infant vocabulary size or growth. It is likely that these are correlated for reasons other than language development: those mothers who are good at establishing and following their children’s joint attention may be better at assessing their children’s vocabulary, whereas mothers who are less responsive overall may also be less good at such an assessment. In sum, responsive mothering might be statistically associated with indicators of more rapid language development for reasons that have nothing to do with the role of social–pragmatic understandings in word learning. The Word-Learning Task, Divided into Three Parts A consideration of what the task of word learning consists of suggests that social–pragmatic information must be supplemented by other types and sources of information. Current research and theory suggest that the process of word learning consists of at least the following three components: word segmentation; an initial fast mapping of the new word onto a referent; and a longer, extended process of completing the lexical entry. There is no evidence, nor indeed any proposals, that social understandings contribute to word segmentation. In contrast, there is evidence that several physical properties of input do contribute, including stress patterns, prosody, and repetition of words in combination with a variety of different words (see references in Aslin, Saffran, & Newport, 1999; Morgan & Demuth, 1996). Not only are social understandings and communicative intentions irrelevant to this process, anything to do with utterance or word meaning also appears to be irrelevant. Although a theory is not expected to account for everything, it is worth noting that despite strong claims that language acquisition is a social process (e.g., Carpenter et al., 1998), there are components of language acquisition that social-process accounts have not addressed. In addition to isolating the sound sequences that constitute words, children must map those sound sequences onto meanings. The social–pragmatic approach argues that children’s socially based understandings of speakers’ intentions aid in this latter task: Children know what their interlocutors are referring to because utterances follow children’s attentional focus or because the children can infer their interlocutors’ communicative intentions. Thus, children are guided to the referents of new words by their in- Hoff and Naigles terlocutors’ skill in following children’s attentional focus and/or by their own understandings of their interlocutors’ intentions. Indeed, the evidence suggests that social–pragmatic information is useful in the identification of referents, at least in experimental settings. Word learning is not complete once the initial mapping of word onto referent has been accomplished, however. Although children’s ability to make a mapping on the basis of as little as one presentation of a word is an impressive feat and important to word learning, it is not the case that children learn the entirety of a given word’s meanings on the basis of a single or very few hearings. The initial fast mapping of a word results in only a partial lexical entry and is followed by a process in which the lexical entry is “completed slowly as the child encounters the word again and contrasts it with other words” (Carey, 1978, p. 292). Indeed, many accounts describe the protracted nature of word learning (e.g., L. Bloom, 1993; P. Bloom, 2000; Gropen, Pinker, Hollander, & Goldberg, 1991; Naigles, 2000a; Rice, 1990). Thus, even if social understandings are responsible for fast mappings, they do not fully explain lexical development. There must be other sources of information as well. The Degree of Mutual Understanding in Adult–Child Conversations The third reason to look for additional sources of information that children might use in word learning comes from a consideration of how frequently social interaction in joint attention actually occurs for most children and how able children actually are to divine their mothers’ meanings during mother–child interactions. Some studies provided data that allowed for the calculation of the percent of time that mothers and children spent in joint attention, which yielded estimates of 11.6% (Carpenter et al., 1998) and 20% (Tomasello & Todd, 1983) for children under 18 months, and 29% for children aged 2 years (Hoff-Ginsberg, 1998). It appears that episodes of joint attention did not comprise much of the time these dyads spent interacting. Tomasello and Farrar (1986) provided no raw data with which to make such calculations, but they reported that their dyads, taped for 15 min during toy play at home, spent two thirds of their time in joint attention episodes. Potential sources of the variability in the proportion of time spent in joint attention are both the setting and the age of the child. Higher estimates come from studies of older children, although studies of younger children have produced the findings that variability in time spent in joint attention predicts vocabulary development (Laakso et al., 1999; Mundy & Gomes, 1998). 421 Another indicator of mutual engagement is the degree to which topics are continued across conversational turns. The social–pragmatic account might be construed to predict that such continuations are common, both on the parents’ side (i.e., parents frequently understanding their children and continuing their topics) and on the children’s side. The data suggest otherwise, however. The percent of maternal utterances that follow children’s topics has been reported as 18.7% (Hoff-Ginsberg, 1987), 21.6% (Akhtar et al., 1991), 32.2% (Hoff-Ginsberg, 1987), and 36.3% (Carpenter et al., 1998). Again, these different estimates come from different settings and children of different ages—in general, there is more topic following with older children. Additionally, there are findings that 45% of parental responses are nonsequiturs (Brown & Hanlon, 1970) and that 10% to 20% of maternal turns and 20% to 30% of paternal turns reflect communicative breakdowns (Tomasello, Conti-Ramsden, & Ewert, 1990). Moreover, only 31% of the signals that preverbal infants produce lead to immediately successful comprehension on their mothers’ parts (Golinkoff, 1986). When topic continuations are calculated from the children’s point of view, it appears that only half of children’s speech receives a topic-related response, and these responses most often simply acknowledge, repeat, or clarify, rather than build on what the children said (Bloom, Margulis, Tinker, & Fujita, 1996). Furthermore, children appear to be no more engaged with mothers than mothers are with children. The percent of all child utterances that continue the topic of a prior maternal prior utterance has been estimated as 21% for children at 20 months of age (Bloom, Rocissano, & Hood, 1976), 33% for children with a mean age of 24 months (Hoff-Ginsberg, 1998), and 47% for children at 36 months (Bloom et al., 1976). It is possible that despite being a small percentage of children’s time in conversation, the time that is spent in mutual engagement is when word learning occurs. This argument, however, requires children to ignore more than half of their linguistic input (because some input is being provided when mutual engagement is not occurring). Moreover, there is evidence that children do learn aspects of their language outside of episodes of mutual engagement; in particular, from overhearing speech among others. In both experimental and naturalistic research (Oshima-Takane, 1988; Oshima-Takane, Goodz, & Derevensky, 1996) found that children acquire aspects of the personal pronoun system from overheard speech. Abundant anecdotal evidence that children go through a stage during which they call their parents by their given names, against the family convention of variants of “Mom” and “Dad,” also implies that they learn from 422 Child Development speech that is not addressed to them (Naigles, 2000b). In sum, evidence that mutual engagement is necessary for word learning is weak and, in many cases, observed benefits of mutual engagement to word learning may be attributed to general effects of positive interaction on child development. Moreover, the social–pragmatic account does not address the segmentation or the lexical entry completion aspects of word learning. The Data-Providing View of Input An alternative to the view that conversation is a social experience and language development is a social process is the view that language acquisition is a datacrunching process and conversation is a delivery mechanism whose value lies, to a substantial degree, in the nature of the data that it delivers. This view of the contribution of conversation to language development has previously been proposed with respect to syntactic development (Hoff-Ginsberg, 1986, 1990, 1999). We propose it in this article with respect to lexical development. Although the data-providing account of the role of conversational experience in lexical development has not, to our knowledge, been previously labeled as such, there are several lines of research that support the notion. The relevant and potentially helpful properties of the speech signal for the word segmentation component of word learning were mentioned earlier and have been explored in detail in work in Morgan and Demuth (1996; see also Aslin et al., 1999). With respect to the initial referent mapping and subsequent lexical entry completion components, the conversation-as-data argument is that children can use their (sometimes partial) understanding of the other words and the structure of the utterance in which an unknown word is placed to make conjectures about the referent of that novel word. Evidence for the potential usefulness and actual use of the rest of the utterance for learning a new word comes from computational and human simulations of word learning and from experimental and naturalistic studies of young children’s word learning. Computer simulations have demonstrated that the use of partial linguistic knowledge to constrain hypotheses, combined with the ability to extract commonalties across different situations of use, can result in lexical acquisition by a system that has no access to speaker intentions at all (Siskind, 1996). To illustrate, knowledge of what the word ball means, combined with knowledge about what kinds of entities do what kinds of things, indicates to the learning device that if the word ball is in the utterance, then the unknown word in that utterance is more likely to mean roll than eat. Evidence that humans can similarly make inferences about word meaning from information in the utterance containing a novel word comes from Gillette et al.’s (1999) simulation of word learning with human (adult) participants. The learners in this case were shown a series of silent videoclips of real mother– child interactions during which a specific verb had been spoken by the mother. The participants were provided with various “clues” to the identity of the verb, including (1) just the videoclips; (2) the videoclips plus the nouns in the mother’s utterance; (3) just the nouns in the utterance; (4) just the sentence frames in which the verb was placed; (5) the sentence frames plus the nouns; or (6) the videoclips, the sentence frames, and the nouns. With only the videoclip information, the participants made correct identifications of the target verbs only 7.7% of the time. Each additional bit of information raised this level of accuracy significantly, until those with complete information, that is, condition 6, made correct identifications 90.4% of the time. The relevance of these simulations to understanding how children actually acquire a lexicon is supported by experimental studies that have demonstrated that children can also use the rest of the sentence as a source of information about word reference and meaning. Two- and three-year-old children are better able to identify the referent of “Susie” if they are told “Susie is painting Jill” than if they are told “Susie is in this picture,” and their choice of a referent for “Susie” is different depending on whether they hear “Susie is painting Jill” or “Susie and Jill are painting,” thus demonstrating that they use information in the structure of the sentence to figure out who, in a picture that they are presented, is “Susie” (Prasada & Choy, 1998; see also Goodman, McDonough, & Brown, 1998). Children’s ability to use information in the structure of sentences as clues to novel word meaning has been amply demonstrated with respect to nouns, verbs, and adjectives (e.g., Naigles, 1990; and see P. Bloom, 1996; Woodward & Markman, 1998 for summaries). Furthermore, Naigles and Hoff-Ginsberg (1998) found that the use of syntax was part of children’s conventional word-learning process, in that the diversity of syntactic frames in which 25 verbs appeared in maternal input predicted the subsequent order of acquisition of those verbs. A more general suggestion that the syntax of input is a source of information that contributes to lexical development comes from findings that maternal mean length of utterance (MLU) is positively related to the size of 1.5-year-old children’s comprehension and production vocabularies as assessed with a checklist (Bornstein et al., 1998). Another relevant feature of input, in addition to Hoff and Naigles lexical and structural clues to meaning, is its sheer amount. One naturalistic source of evidence that amount of input matters is the finding that the relative sizes of the Spanish and English vocabularies of bilingually developing 1- to 2-year-olds in South Florida are related to the relative amount of input they receive in each language (Pearson, Fernandez, Lewedeg, & Oller, 1997). In addition, children’s overall rates of vocabulary growth are related to the amount of speech they hear (Hart & Risley, 1995; Huttenlocher, Haight, Bryk, Seltzer, & Lyons, 1991). Amount of input is, of course, related to the frequency with which each word is presented and there is substantial evidence that frequency affects word learning. Children’s first uses of a word are likely to match the most frequently occurring use of that word by the children’s mothers (Harris, Barrett, Jones, & Brookes, 1988), and the order in which words appear in children’s vocabularies is predicted by their frequency in input (Huttenlocher et al., 1991; Naigles & Hoff-Ginsberg, 1998). There are complementary experimental demonstrations that frequency in input is positively related to word learning (Schwartz & Terrell, 1983; Smith, 1999). One reason that frequent presentations may benefit word learning is that multiple presentations are likely to vary in the accompanying nonlinguistic and linguistic contexts; thus, each presentation provides somewhat new information about word meaning. With respect to what types of variety matter, several researchers have argued that children require crosssituational information (Fisher, Hall, Rakowitz, & Gleitman, 1994; Pinker, 1989) because this reveals more aspects of a given word’s meaning. In a direct assessment of the value of cross-situational information, Akhtar and Montague (1999) gave 2-, 3-, and 4-year-old children the sentence, “This is the modi one,” paired with three different objects selected from a visible array of nine objects. Across all three age groups, the children were able to identify the common attribute of those three objects and use it as a basis for extending the new term, modi, to novel objects. For similar reasons, hearing words (particularly verbs) in a variety of syntactic environments should also be useful because each syntactic frame in which a verb appears supports additional conjectures about the semantics of that verb (Gleitman, 1990; Naigles, 1996; Naigles & Hoff-Ginsberg, 1998; see also Waxman & Markow, 1998, for similar findings with adjectives). The finding that the syntactic diversity of the context in which verbs appeared in mothers’ speech uniquely accounted for variance in both the frequency and the syntactic diversity of the children’s use of those verbs is consistent with that argument (Naigles & HoffGinsberg, 1998). 423 Other data suggest that children who hear a greater number of different words produce a greater number of different words. The number of word types in their mothers’ speech has been found to be positively related to the size of children’s comprehension and production vocabularies (Bornstein et al., 1998). Crosslinguistic comparison shows that when mothers produce more verb types (i.e., Chinese mothers who speak Mandarin), children’s spontaneous speech contains more verb types and tokens, and when mothers’ speech contains more or equal numbers of noun types (i.e., mothers who speak English and mothers who speak Italian), children’s speech includes more noun types and tokens (Tardif, Shatz, & Naigles, 1997; see also Choi & Gopnik, 1995). In these data, the difference in the relative usage of verbs and nouns was only in types, not in tokens; therefore the data only provide evidence that the number of types has a unique effect on children’s vocabularies. That is an important finding, however, because within a language, the number of types and number of tokens tend to be highly correlated (e.g., Hart & Risley, 1992, 1995; Weizman & Snow, 2001). Thus, any relations between the amount of input provided to children and children’s vocabulary size that is found in naturalistic data could be either effects of word frequency, effects of the number of different words modeled, or some combination of the two. In sum, the results of both experimental and naturalistic studies suggest that not only can children use the data-providing aspects of their conversational input, but also that the availability of data in input affects children’s lexical development. Taken together, these findings suggest that three different data-providing properties of input are related to children’s vocabulary development: sheer frequency of presentation, number of different words, and richness and variety of linguistic environments in which the words are placed. Summary and Prospectus Research findings suggest that the degree to which mother and child are mutually engaged in their conversations as well as the data provided in the utterances that mothers produce in conversation are related to children’s lexical development. Integrating the accounts of how these two sources of information are used into a single theory of the process by which children use input to acquire a lexicon has been hampered by the fact that little research has investigated the influence of both sources of support within a single study. The present study was designed to do just that by looking for correlations between properties of 424 Child Development the input that children hear and the size of their subsequent vocabularies, focusing on both properties of input that plausibly index the extent to which social engagement makes meaning transparent and properties of input that index the amount of nonsocial information in the utterances that children hear. The goal was to investigate the unique and combined contributions of these properties of input to children’s vocabulary growth. erage duration of the Time 1 tapings was 42.4 minutes (SD 5 8.4). The interactions were transcribed by trained research assistants into the format required by the Systematic Analysis of Language Transcripts (SALT) software (Miller & Chapman, 1985). The measures of input were based on the transcripts of the first visit (Time 1). The measures of child language were assessed from transcripts of both visits (Time 1 and Time 2). METHOD Measures Participants Input measures. The measures of the data-providing properties of maternal speech included the total number of utterances produced, the number of word tokens (i.e., the total number of words) in the input samples, the number of word types (i.e, the number of different words), and the MLU. This latter measure indicates the degree of syntactic complexity in the utterances, and as such may be considered an index of the richness of the linguistic environment. In counting word types, different forms of the same root were treated as the same word. Thus, for example, walk, walked, and walking were counted as one word type, as were table and tables. For the measures of the number of utterances, word types, and tokens, no correction was made for the individual differences in the duration of interaction, because these reflect real differences in the children’s conversational experiences (see Hoff-Ginsberg, 1992, for a more complete argument to this effect). The number of word tokens, the number of word types, and the MLU in morphemes were calculated using SALT. Two measures were selected to index the degree to which maternal speech was likely to be referentially transparent by virtue of the social engagement of mother and child. The first was the number of maternal utterances produced during episodes of joint attention. Joint attention was coded only for the toy play interaction, because judging attentional focus can require eye gaze information, and only the toy play interaction was recorded in a manner that ensured that both the mother’s and child’s faces were visible. Joint attention was defined following Tomasello and Farrar (1986) and Tomasello and Todd (1983) as periods lasting at least 3 s during which the mother and child were both focused on the same object or activity. All coding of joint attention was done by the same research assistant. Reliability was assessed by comparing the research assistant’s ratings on this measure with those produced by a second research assistant who independently coded 6 of the 63 play sessions. The two coders were always within 8 percentage points of each other in their estimates of Participants were 63 children who resided in the midwestern United States and were between the ages of 18 and 29 months at the start of the study (age: M 5 21.3 months, SD 5 3.05). Thirty-three of the children came from high-SES families in which both parents were college educated, and 30 of the children came from mid-SES families in which both parents were high school educated. Within the high-SES sample there were 9 firstborn boys, 7 firstborn girls, 7 laterborn boys, and 10 laterborn girls; within the mid-SES sample there were 8 firstborn boys, 8 firstborn girls, 8 laterborn boys, and 6 laterborn girls. All the families were White. All the mothers were native speakers of English, were the primary caretakers for their children, and were not employed outside of the home for more than 15 hours per week. The children were selected to be comparable in terms of their level of productive language use, and were all at the point at which they were just beginning to combine words. Each child was heard to produce at least three different two-word combinations during a preliminary screening visit, but no more than 50% of any child’s utterances were multiword constructions, as assessed on the basis of the first speech sample. The average MLU for the children was 1.27 (SD 5 .12). The age of the children was not a selection criterion. The sample is more fully described in Hoff-Ginsberg (1991). Procedure Children were videotaped in dyadic interaction with their mother at two time points, 10 weeks apart. The recordings were made in the participants’ homes during a mealtime, as the mothers were getting the children dressed for the day, and during toy play with experimenter-provided toys. The durations of the mealtime and dressing interactions were allowed to vary naturally and were taped in their entirety. The toy play was taped for no more than 25 min. The av- 425 Hoff and Naigles the total percent of utterances in joint attention; the correlation between the two coders’ estimates was r(4) 5 .98. The second measure of social engagement was the number of maternal utterances that were topiccontinuing replies to child speech, as an index of the contingency of maternal speech on child speech. To be coded as a topic continuing reply, a maternal utterance had to follow a child’s utterance immediately and continue the topic of the child’s utterance. Utterances were coded as topic continuing if they met one of the following conditions: the utterance (1) referred to any entity or event mentioned in the child’s utterance, (2) was an answer to a question, (3) continued some patterned speech such as reciting the alphabet or a nursery rhyme, (4) commented on objects or events referred to in the prior utterance, or (5) was a paraphrase of the prior utterance. The category of topic-continuing replies was one of several in the scheme for coding the topic relations in mother–child conversation described in Hoff-Ginsberg (1987). In the present study, interrater agreement for this code was 87%, with a Cohen’s k of .80, calculated on 220 utterances from excerpted portions of the mealtime and toy play interactions in two different transcripts. Lastly, two other measures of the social–pragmatic aspects of interaction were calculated: (1) the number of maternal utterances judged to be behavior directives, and (2) the number of maternal utterances judged to be intended as conversation-eliciting questions. Behavior directives directed either the child’s attention or behavior. Conversation-eliciting questions were utterances intended to elicit verbal replies, including several categories of questions and prompts to answer questions. It has been argued that these measures are also related to the hypothesis that mutual social engagement is the basis for language learning because the relative frequency of directives and conversation-eliciting questions that mothers use indicates the degree to which their purpose in interacting with their child is to control behavior or engage in conversation (Hoff-Ginsberg, 1991; McDonald & Pien, 1982). Furthermore, Hoff-Ginsberg (1990), in another sample of mothers and children, found that children respond to conversation-eliciting questions more than they respond to other types of maternal utterances. Thus, a higher frequency of conversationeliciting questions would appear to be a good index of interaction that involves more mutual engagement. Interrater reliability for this code calculated between the coder and the first author yielded 82% agreement and a Cohen’s k of .80, based on 201 coded utterances in excerpted portions of the mealtime and dressing segments of two transcripts. This assessment of reli- ability was done at a different time and on different transcripts than the assessment of reliability for topic continuity. Coders were blind to participants’ SES. Child language measure. To assess differences among the children with regard to the size of the vocabularies they used in spontaneous speech, all of the children’s transcripts were truncated to the size of the shortest transcript so that the estimates would not be contaminated by differences among the children in the amount of their verbal output (see Hoff-Ginsberg, 1992; Richards, 1987). This procedure resulted in transcripts for each child that were 90 utterances in length and were selected in approximately equivalent proportions from the mealtime, dressing, and toy play instructions. On the basis of these speech samples, the total number of word types produced by each child was calculated using SALT. This count of word types in a speech sample does not provide an estimate of the size of the children’s total vocabularies, but it does provide estimates of the variety of vocabulary that children use. Therefore, we investigated how social– pragmatic properties of mother–child conversation and the data-providing properties of the maternal utterances produced in these conversations compare in terms of predicting children’s vocabulary diversity within a given fixed sample, not their entire vocabulary size. These same 90-utterance samples were also used to calculate the MLU in morphemes, again using SALT. RESULTS Descriptive statistics for the child vocabulary data are presented in Table 1; Table 2 presents descriptive statistics for the input measures hypothesized to explain the child vocabulary data. Correlations were calculated between the input measures based on the Time 1 language samples and the child vocabulary measures based on the Time 2 language samples, removing the variance attributable to variance in the children at Time 1. Such partialing was necessary to avoid obtaining correlations between input at Time 1 and child language measures at Time 2 that are only reflections of the effect of the Table 1 Means and Standard Deviations for Child Vocabulary Measure Time 1 Total word types Time 2 M SD M SD 36.06 8.07 48.40 12.98 Note: Word counts are based on 90-utterance speech samples. 426 Child Development Table 2 Means and Standard Deviations for Measured Properties of Input (Time 1) Input Property Data-providing properties Number of utterances Number of word tokens Number of word types Mean length of utterance Social–pragmatic properties Number of utterances in joint attentiona Number of topic-continuing replies Number of behavior directives Number of conversation-eliciting questions a Coded M SD 614 1,882 298 3.56 232 763 82 .43 101 130 120 193 59 58 60 90 for toy play session only. children’s language levels at Time 1 on both their input and their future language levels (Newport, Gleitman, & Gleitman, 1977). This procedure, however, is a very conservative approach that will “underestimate parent effects on the child in cases where these effects have already manifested themselves by the time of the first observation” (Huttenlocher et al., 1991, p. 240). Three of the four measures of the data-providing properties of input were found to predict children’s vocabulary: the number of word tokens, the number of word types, and the MLU were all positively related to subsequent lexical growth. Of the dataproviding measures, only the total frequency of utterances bore no relation to vocabulary development. No measure of the social–pragmatic features of interaction predicted vocabulary. The values of the partial rs obtained are presented in Table 3. The next step in data analysis was to pursue the Table 3 Partial Correlations between Input Properties at Time 1 and Child Vocabulary at Time 2 Input Measure Data-providing features Number of utterances Number of word tokens Number of word types Mean length of utterance Social–pragmatic features Number of utterances in joint attention Number of topic-continuing replies Number of behavior directives Number of conversation-eliciting questions Number of Word Types in Child Speech .05 .21* .22* .55*** .02 .18 2.05 2.03 Note: Variance attributable to the number of word types in child speech at Time 1 was removed. * p , .05 (one-tailed); *** p , .001. (one-tailed). significant findings to ask how the predictive dataproviding properties of input operated together to account for variance in children’s vocabularies. Analysis of the intercorrelations among these predictive properties of input revealed that the number of word tokens and the number of word types in input were so highly correlated as to make their effects statistically unseparable, r(61) 5 .89, p , .001. For further use in the tests of predictive relations, one measure— the number of different word types—was selected. The MLU of input was also significantly related to the number of word types in input, but that correlation was more moderate, r(61) 5 .48, p , .001, and both of those variables were retained as separate predictors. The reduced set of predictors—input MLU and number of word types in input—were entered into a hierarchical regression analysis with the child vocabulary measure at Time 2 as the outcome. Child vocabulary at Time 1 was entered first, analogous to the previous partialing procedure; number of words types in input was entered second; and input MLU was entered last. Input MLU was entered last because the positive correlation between the MLU and child vocabulary was not specifically predicted by any hypothesized mechanism of word learning, although it had been observed once before (Bornstein et al., 1998). We wanted to remove from the MLU any predictive power carried by properties of input that were predicted by hypothesized mechanisms of learning. The results, presented in Table 4, confirmed that the number of word types in input was a significant predictor of child vocabulary when analyzed alone, but the MLU accounted for more variance. Furthermore, in the final model with both word types and MLU as predictors, only the MLU was a significant predictor. One last analysis was conducted to ensure that the observed relation between the MLU of input and child vocabulary was not spurious. It is possible that children who hear longer utterances also produce longer utterances, and thus use a greater number of different words in a 90-utterance speech sample not because their vocabularies are bigger, but because the number of word tokens in the speech sample is Table 4 Hierarchical Regression Predicting Time-2 Word Types in Child Speech Predictors R2 Child word types (Time 1) Input word types Input mean length of utterance .20 .24 .44 Adjusted R2 DR2 .18 .21 .42 * p , .05; ** p , .01; *** p , .001 (one-tailed). .20*** .04* .20*** Final b .31** 2.02 .53*** Hoff and Naigles Table 5 Hierarchical Regression Predicting Time-2 Word Types in Child Speech, Removing Variance Attributable to Child Mean Length of Utterance (MLU) Predictors R2 Adjusted R2 DR2 Child word types (Time 1) Child MLU (Time 2) Input word types Input MLU .20 .61 .62 .65 .19 .59 .60 .63 .20*** .41*** .01 .03* Final b .31** .53*** .03 .24** * p , .05; ** p , .01; *** p , .001 (one-tailed). greater. To test this possibility, the regression analysis in Table 4 was rerun with variance in children’s MLU at Time 2 removed before the input predictors were entered. As Table 5 shows, the outcome of this analysis was that input MLU remained a significant predictor of child vocabulary. DISCUSSION The foregoing analyses of the relation between properties of input and children’s subsequent vocabulary use provide a snapshot of children around the age of 24 months as they use language experience to build their productive vocabularies. The observed patterns of correlation suggest that at this point in development, variation in the extent to which mothers and children are mutually engaged in their conversations has little influence on the richness of the vocabulary that children will come to use. In contrast, variation in the lexical richness and syntactic complexity of the utterances that mothers produce in those conversations does account for variation among children with regard to their subsequent production vocabularies. These findings are consistent with the theoretical arguments of the present study that the process of lexical development can be understood only by considering the data-providing function of children’s conversational experience in addition to the social support for language development that conversation may provide. The Social–Pragmatic Basis of Lexical Development Previous findings that have been interpreted as evidence for the social–pragmatic basis of lexical development have, in large part, come from either correlational studies that found that differences in the degree to which children experience joint attention predicted subsequent vocabulary, or experimental demonstrations that showed that children have the ability to use social–pragmatic cues to a speaker’s communicative intentions. The correlational studies primarily used 427 maternal checklists or diaries as the basis for their estimates of child vocabulary size, whereas the present study used a fixed-size sample of spontaneous speech. Maternal checklists have not been subjected to the usual criteria of inter- and intrarater reliability (L. Bloom, 1993). One possibility, as discussed earlier, is that the observed child differences in vocabulary size assessed by checklist can be attributed to maternal differences in sensitivity to child language, because, after all, mothers fill out checklists based on the interactions that they have had with their child. To be sure, these interactions are more extensive than those recorded by experimenters; however, they are also more likely to be selectively filtered. For example, Carpenter et al. (1998) found correlations of joint engagement and maternal following with children’s subsequent word comprehension (assessed via the MacArthur Communicative Developer Inventory [CDI] checklist), but not production (also assessed via the CDI). At such a young age (12–15 months), mothers’ estimates of their children’s levels of word comprehension are extremely likely to be differentially influenced by maternal sensitivity. The foregoing arguments suggest, then, that some of the correlations between mutual engagement and child vocabulary previously observed in naturalistic data are spurious. We do not make that claim with respect to all the previously observed correlations, however. Some correlations, not involving maternal report measures of child language, are quite probably real, but it is worth noting that (1) they involved children who were younger than the participants in the present study, and (2) they involved maternally guided rather than child-guided mutual engagement. Thus, previous findings that maternal responsiveness, contingency, and follow-in behavior benefit vocabulary development may reflect the fact that these pragmatic features of maternal language are particularly important when children are too young to have the skills to discern maternal intentions and are not sufficiently linguistically advanced to use lexical and syntactic cues to meaning. At this early stage— between 9 and 18 months—the relevant property of interaction may be how well the mother structures the interaction so that children hear the labels of the things that they are already attending to. The only thing social about this support for word learning is the social skill of the mother in following her child’s attentional focus and timing her words to match. From the child’s point of view, the benefit is that of temporal contiguity between attention and input. Similarly, the finding that individual differences among children under 18 months with regard to their joint attention skills predict subsequent language de- 428 Child Development velopment (Laakso et al., 1999; Mundy & Gomes, 1998) may merely reflect the fact that being in joint attention makes temporal contiguity of the child’s attention and the word presented more likely. That is, the benefit is real and depends on a social skill—this time the child’s—but is not the social skill of reading others’ intentions. By the age of 24 months, developmental differences in these skills among children are likely to have evened out, and all children may have a sufficient level of skill such that any remaining differences have little consequence for language development. The last type of finding that is used to argue for the social–pragmatic view of lexical development is the experimental finding that children can read speaker intentions and use them as clues to word meaning (Baldwin, 1993; Golinkoff et al., 1999; Tomasello & Barton, 1994). This ability, however, has been demonstrated only in children over 18 months of age, and thus it could not be part of the early social basis of word learning. In principle, children over 18 months could use this ability to learn words, but in the present study no evidence of this ability was found in naturalistic data. The present study investigated the role of the child’s reading of speaker communicative intention only indirectly. We reasoned that if children found significant information in reading the communicative intentions of their conversational partners, then children whose mothers more often expressed readable communicative intentions—indexed by utterances spoken in joint attention and by the relatedness of the topics of maternal speech to prior child speech—should be advantaged in vocabulary development. No such advantage was found, however. It is always difficult to interpret a null result. It is possible that more direct measures of intention reading and/or other measures of vocabulary (e.g., comprehension) would have produced positive results. It is also possible that the pattern of positive findings for data-providing properties and null findings for social–pragmatic properties of conversation reflects that the data-providing properties of input can be more reliably assessed from a limited sample of interaction than can the social–pragmatic properties. This concern may particularly pertain to the measure of language use in joint attention because it was assessed only for the toy play interaction. We cannot, and do not, conclude from the current findings that there is no effect of mutual engagement on language development at this stage. Some degree of mutual engagement is necessary, if only to sustain the interactions that provide the child with input. On the other hand, the present database for assessing social–pragmatic features was as large or larger than that used in most previous studies that have found effects at younger ages, and toy play is the most widely used interactive setting in both naturalistic and experimental studies. Thus, the present findings deserve interpretation. The findings of the present study suggest a developmental progression in which social–pragmatic properties of conversation—which result primarily from maternal effort—produce information that is used by children at the very earliest stages of word learning but apparently is not used at later stages (around 24 months of age). This account has points that both agree and disagree with other recently proposed accounts. For example, Hollich et al. (2000) asked children aged 12, 20, and 24 months to learn novel words for objects, and compared the usefulness of perceptual and pragmatic information for making the correct object–word mapping at each age. Hollich et al. found that the youngest children did not use experimenter eye gaze (i.e., child-guided joint attention) as a source of information about the referent of the novel word; however, the older children (20 months and older) did. Their findings with the youngest group led them to conjecture that very young children’s initial learning of words is not based on their reading of the communicative intents of others. Our understanding of the data from naturalistic studies is in line with this view: the usefulness of social– pragmatic bases of information for child word learners around 10 to 15 months of age is due to maternal sensitivity, not children’s social–cognitive abilities. The findings in this study, however, do not converge with those of Hollich et al. (see also Baldwin, 1993) with respect to word learning in children over 18 months. That is, their experiments found that children 18 months and older can read adults’ communicative intents en route to learning novel words. In contrast, the present study’s data indicate that children are apparently not using this ability as they go about the process of actually learning words of English. That is, even though children over 18 months of age can use speaker intent, it appears not to be a particularly important source of information for actual word learning. Thus, the question of what children can use in word learning, which experiments address, and what children actually do use, which naturalistic data address, are two different questions. Moreover, the current findings suggest that these two questions have different answers for 24-month-old word learners. A similar suggestion has been made by Carpenter et al. (1998), who allowed that the relations between joint attentional engagement and children’s language development that they observed may be true during only the earliest stages of language development. Later, children can and do use more complex sources Hoff and Naigles of information. One final possibility that only can be considered at this point is that the toddlers’ newfound ability to read intentions beginning at 18 months may mean that the usual measures of social–pragmatic engagement (joint attention, maternal topic following) are no longer the right ones; what becomes needed— and what no one has yet analyzed in this vein — is a direct assessment of children’s discernment of their mothers’ communicative intents in naturalistic interaction. In sum, the previous findings, in conjunction with the findings of the present study, suggest that the social–pragmatic features of mother–child conversation matter most to vocabulary development early in the second year of life. The particular social–pragmatic features that appear to matter—maternal responsivity, contingency, and following of the child’s attentional focus—are all things that mothers do to maximize the match between the child’s attentional focus and the speech the child hears and to reinforce the child for participating in communicative interaction. This may indeed be a crucial social–pragmatic foundation for language development. The child contributes to these interactions the capacity to be engaged, but the social–pragmatic work that benefits vocabulary development in particular is done by the mother. The child’s social abilities and inclinations are a precondition for learning, but they are not part of the explanation, at that early stage, of how children figure out the meaning of the new words they hear. After 18 months, children do have the ability to read others’ intentions, but this has been demonstrated only in experimental settings. Thus far, there has been no empirical demonstration that the exercise of children’s experimentally demonstrated ability to read intentions and use them as clues to word meaning plays a significant role in accounting for conventional vocabulary development. This may be because, as was argued in the Introduction, naturalistic interaction simply does not provide the carefully constructed cues that experiments have provided, or because other sources of information prove more important. The Data-Analytic Basis of Lexical Development The analyses of correlations between properties of input and children’s subsequent vocabularies identified three data-providing features of input as positive predictors: the number of words produced, the lexical richness of the vocabulary used (i.e., the number of different words), and utterance length. The number of utterances in input was unrelated to vocabulary growth. This latter finding is consistent with the null findings with respect to the social–pragmatic mea- 429 sures of engagement. Whether or not the mother talks to the child is essentially a measure of engagement. The present findings argue that it is not the fact that mothers talk to their children, but rather the lexical and syntactic properties of what they say, that is relevant for vocabulary development. These findings are consistent with Huttenlocher et al.’s (1991) finding that the amount of input predicts vocabulary growth, because in that study amount was measured in terms of number of words, not number of utterances. In the present data, number of words (i.e., tokens) was similarly a significant predictor of child vocabulary, but it was also highly correlated with the number of different words (i.e., types). Of these two highly correlated measures, only the number of types was retained in further analyses. The observed benefit of lexical richness, however, must be interpreted as a benefit of not only lexically varied input but also of a large amount of such input. Although it is an interesting intellectual exercise to consider how the amount and lexical variety of input might independently benefit lexical development, the present data show that in naturally occurring maternal speech, the two properties are too highly correlated to be able to separate their effects. This finding is consistent with those of Hart and Risley (1992, 1995) and Weizman and Snow (2001) who found that mothers who produced more words when talking to their children also produced a greater number of different words. Relatedly, Huttenlocher et al. (1991) found that more talkative mothers did not use more word types in equivalent-sized samples of speech. Nonetheless, the by-product of talking more appears to be using a bigger vocabulary. Multiple regression analyses further revealed that this benefit of lexically rich input was not independent of the benefit of syntactic complexity. It is a property of speech to children, perhaps of speech in general, that use of a more varied vocabulary is associated with production of longer utterances. The positive correlation between the lexical richness of input and child vocabulary is carried by the effect of the MLU. Maternal MLU uniquely accounted for a significant portion of the variance in children’s vocabularies. These findings are consistent with other findings in the literature that the amount of input, the lexical richness of input, and maternal MLU are positively associated with children’s vocabulary sizes (Bornstein et al., 1998; Hart & Risley, 1995; Huttenlocher et al., 1991). The present findings suggest, for the first time, that among these correlated predictors, the MLU accounts for the greatest portion of the variance in child vocabulary sizes. What does the observed benefit of large amounts of lexically rich and, in particular, syntactically com- 430 Child Development plex input suggest about how children use input to build a lexicon? Large amounts of input provide repeated exposure to the same word, which is likely to be useful for several reasons. First, as Huttenlocher et al. (1991) argued, repeated exposures function as multiple learning trials. Second, words presented frequently are likely to be presented in a variety of situations and syntactic frames, allowing for both crosssituational learning and syntactic bootstrapping. Frequency may be particularly relevant when the outcome measure is production based, because frequent presentations of a word in input may not only help build a lexical entry by providing more opportunities to obtain information about word meaning, but also may also have a separate effect on the ease with which children can retrieve a word in the course of talking. One benefit of lexical richness seems obvious: the more words a child hears, the more words the child can potentially learn. Additionally, to the extent that children know—even partially—the meaning of the other words presented with a new word, that information constrains the possible interpretation of the new word. This finding is wholly consistent with the results of simulations that demonstrate a benefit of complex input to a variety of language-learning tasks (Gillette et al., 1999; Plaut & Kello, 1999; Siskind, 1996). What is interesting is that this contradicts the notion that simple input is better with regard to children’s language acquisition. The benefit of longer MLU also contradicts the simpler-is-better notion, but it is consistent with evidence about potential sources of clues to word meaning. Other known words in the utterance, as well as the structure of the utterance in which a new word appears, provide clues to the meaning of that new word. Longer utterances may provide more of these sources of information than do shorter utterances— at least for children old enough to take advantage of the information and within the range of utterance length to be found in maternal child-directed speech. Longer utterances may also be long because they contain explicit information about new word meaning (Clark, 1999; Koenig & Naigles, 1996). For example, hearing a new word, bat, in a sentence such as “That’s a bat,” is less useful for building a new lexical entry than hearing bat in the sentence, “Bats have big wings and they live in caves.” Such additional information may not only contribute to completing a lexical entry, but also may result in deeper processing of that new word, with all the benefits of depth-of-processing to memory.1 This explanation is consistent with Della Corte, Benedict, and Klein’s (1983) finding that chil1 The authors are grateful to Tom Sawallis for this suggestion. dren’s noun vocabularies are related to the amount of description their mothers provide, and not to the amount of task-oriented, child-oriented, or context-oriented talk. It suggests that building a vocabulary depends not only on hearing words, but also on finding information about word meaning in the speech that one hears. Summary and Conclusions Acquiring a vocabulary consists of learning mappings between sound sequences and meanings. An obvious prerequisite to this process is development of the conceptual understandings that sound sequences express (L. Bloom, 1993). Also necessary is a process for figuring out which meanings are being expressed by which sound sequences—the part of the process of lexical development addressed here. The combined results of this and previous studies suggest that the process of word learning makes use of both the human child’s social interest in and ability to interact with others and the child’s computational ability to extract information from the speech presented in those interactions. Evidence that babies as young as 7 months can learn distributional regularities in input (Saffran, Aslin, & Newport, 1996) and have actually learned some of the distributional regularities that constitute the phonotactics of their language (Jusczyk, 1997) suggests that the computational mechanism starts working early on. The fact that some severely asocial individuals, such as some individuals with autism, can acquire aspects of language suggests that the computational mechanisms alone may be sufficient for the acquisition of linguistic competence, if not communicative competence. On the other hand, the rich body of evidence on children’s capacities for joint attention and the evidence that time spent in joint attention predicts language development suggest that in the normal course of events, language acquisition is very much a social process. A critical look at the data, however, suggests that the nature and extent of the social contribution may be less than sometimes has been claimed. The social nature of humans and human interaction appear to do two things for language acquisition. One, social interaction provides the motivation and occasion for language use and thus brings the child into the context in which language-advancing data are provided. Second, children’s social cognitive capacities to infer the intentions of others may actually provide data about word reference that is used by the computational mechanism. Just how significant a source this information is in word learning remains to be demonstrated. It depends on how often the mutual understanding it requires actually occurs and on the rela- Hoff and Naigles tive value of that source of information over other information also available in the environment. Although both social and data-analytic processes contribute to lexical development, their relative importance changes over the course of development. When joint attention skills begin to develop, there are developmental differences among children that will result in individual differences in the rate of language development. Similarly, differences among mothers in the extent to which they respond to these emerging abilities will affect language development. Thus, at this point—between 9 and 18 months—variability in the experience of joint attention may be the greatest source of variability affecting lexical development. As children mature, early developmental differences in the capacity for engagement recede as even the slower developing children acquire this capacity. As all children become competent at staying engaged and following their mother’s focus, individual differences in maternal responsivity and contingency lessen in importance. Instead, what comes to matter is how much data is available and how informative are those data. The particular qualities that appear to make data useful suggest that the child’s armamentarium of computational devices includes mechanisms for extracting information from the nonlinguistic context and from the content and structure of utterances, a mechanism for accruing that extracted information across multiple presentations, and a mechanism for inducing what is common among those accrued instances. The richer the information, the faster the process occurs. The present argument does not deny the necessity of the social nature of humans nor the significance of the experience of social interaction to language development. It merely asserts that the explanation of how language develops as a result of experiencing social interaction cannot be written solely in the vocabulary of social processes. In those social interactions, children obtain data regarding the form of language and there are computational mechanisms that children bring to bear on those data to produce linguistic knowledge. Thus, a full account of language development must be written in both the vocabulary of social processes and the vocabulary of computational mechanisms. ACKNOWLEDGMENTS Data collection was supported by the National Institute of Child Health and Human Development, grant #HD20936, and by a Spencer Foundation grant to the first author. Portions of this research were presented at the 23rd Boston University Conference on Language Development, Boston, 1998; and the VIIIth In- 431 ternational Congress for the Study of Child Language, San Sebastian–Donostia, Spain, 1999. The authors thank Carol Fowler, Janellen Huttenlocher, Alan Kersten, Brett Laursen, and the anonymous reviewers for conversations on this topic and for comments on earlier versions of this article. ADDRESSES AND AFFILIATIONS Corresponding author: Erika Hoff, Department of Psychology, Florida Atlantic University, 2912 College Avenue, Ft. Lauderdale, FL 33314; e-mail: ehoff@ fau.edu. Letitia Naigles is at the University of Connecticut in Storrs. REFERENCES Akhtar, N., Dunham, F., & Dunham, P. J. (1991). Directive interactions and early vocabulary development: The role of joint attentional focus. Journal of Child Language, 18, 41–50. Akhtar, N., & Montague, L. (1999). Early lexical acquisition: The role of cross-situational learning. First Language, 19, 347–358. Akhtar, N., & Tomasello, M. (2000). The social nature of words and word learning. In R. Golinkoff & K. HirshPasek (Eds.), Becoming a word learner: A debate on lexical acquisition. Oxford, U.K.: Oxford University Press. Aslin, R., Saffran, J., & Newport, E. (1999). Statistical learning in linguistic and nonlinguistic domains. In B. MacWhinney (Ed.), The emergence of language (pp. 359–380). Hillsdale, NJ: Erlbaum. Baldwin, D. A. (1993). Infant’s ability to consult the speaker for clues to word reference. Journal of Child Language, 20, 395–418. Baldwin, D. A. (2000). Interpersonal understanding fuels knowledge acquisition. Current Directions in Psychological Science, 9, 40–45. Bloom, L. (1993). The transition from infancy to language: Acquiring the power of expression. New York: Cambridge University Press. Bloom, L. (2000). Pushing the limits on theories of word learning [Commentary]. In G. J. Hollich, K. Hirsh-Pasek, & R. M. Golinkoff, (2000). Breaking the language barrier: An emergentist coalition model for the origins of word learning. Monographs of the Society for Research in Child Development, 65(3, Serial No. 262). Bloom, L., Margulis, C., Tinker, E., & Fujita, N. (1996). Early conversations and word learning: Contributions from child and adult. Child Development, 67, 3154–3175. Bloom, L., Rocissano, L., & Hood, L. (1976). Adult-child discourse: Developmental interaction between information processing and linguistic knowledge. Cognitive Psychology, 8, 521–552. Bloom, P. (1996). Word learning and the part of speech. In R. Gelman & T. Au (Eds.), Handbook of perceptual and cognitive development (pp. 151–184). New York: Academic Press. 432 Child Development Bloom, P. (2000). How children learn the meanings of words. Cambridge, MA: MIT Press. Bornstein, M. H., Haynes, M. O., & Painter, K. M. (1998). Sources of child vocabulary competence: A multivariate model. Journal of Child Language, 25, 367–393. Brown, R., & Hanlon, C. (1970). Derivational complexity and order of acquisition in child speech. In B. Hayes (Ed.), Cognition and the development of language (pp. 11– 54). New York: Wiley. Bruner, J. (1974/1975). From communication to language: A psychological perspective. Cognition, 3, 255–288. Carey, S. (1978). The child as word learner. In M. Halle, J. Bresnan, & G. A. Miller (Eds.), Linguistic theory and psychological reality (pp. 264–293). Cambridge, MA: MIT Press. Carpenter, M., Nagell, K., & Tomasello, M. (1998). Social cognition, joint attention, and communicative competence from 9 to 15 months of age. Monographs of the Society for Research in Child Development, 63(4, Serial No. 255). Choi, S., & Gopnick, A. (1995). Early acquisition of verbs in Korean: A cross-linguistic study. Journal of Child Language, 22, 497–529. Clark, E. (1999). Metalanguage directions and their uptake in language acquisition. San Sebastian/Donostia, Spain: International Association for the Study of Child Language. Della Corte, M., Benedict, H., & Klein, D. (1983). The relationship of pragmatic dimensions of mothers’ speech to the referential-expressive distinction. Journal of Child Language, 10, 35–44. Fisher, C., Hall, G., Rakowitz, S., & Gleitman, L. (1994). When it is better to receive than to give: Syntactic and conceptual constraints on vocabulary growth. Lingua, 92, 333–375. Gillette, J., Gleitman, H., Gleitman, L., & Lederer, A. (1999). Human simulations of vocabulary learning. Cognition, 73, 135–176. Gleitman, L. (1990). Structural sources of verb meanings. Language Acquisition, 1, 3–55. Golinkoff, R. (1986). “I beg your pardon?” The preverbal negotiation of failed messages. Journal of Child Language, 13, 455–476. Golinkoff, R., Hirsh-Pasek, K., & Hollich, G. (1999). Emergent cues for early word learning. In B. MacWhinney (Ed.), The emergence of language (pp. 305–330). Hillsdale, NJ: Erlbaum. Goodman, J., McDonough, L., & Brown, N. (1998). The role of semantic context and memory in the acquisition of novel nouns. Child Development, 69, 1330–1344. Gropen, J., Pinker, S., Hollander, M., & Goldberg, R. (1991). Syntax and semantics in the acquisition of locative verbs. Journal of Child Language, 18, 115–152. Harris, M., Barrett, M., Jones, D., & Brookes, S. (1988). Linguistic input and early word meaning. Journal of Child Language, 15, 77–94. Harris, M., Jones, D., Brookes, S., & Grant, J. (1986). Relations between the non-verbal context of maternal speech and rate of language development. British Journal of Developmental Psychology, 4, 261–268. Hart, B., & Risley, T. (1992). American parenting of language-learning children: Persisting differences in family-child interactions observed in natural home environments. Developmental Psychology, 28, 1096–1105. Hart, B., & Risley, T. (1995). Meaningful differences in the everyday experience of young American children. Baltimore, MD: Brookes. Hoff-Ginsberg, E. (1986). Function and structure in maternal speech: Their relation to the child’s development of syntax. Developmental Psychology, 22, 155–163. Hoff-Ginsberg, E. (1987). Topic relations in mother-child conversation. First Language, 7, 145–156. Hoff-Ginsberg, E. (1990). Maternal speech and the child’s development of syntax: A further look. Journal of Child Language, 17, 85–99. Hoff-Ginsberg, E. (1991). Mother-child conversation in different social classes and communicative settings. Child Development, 62, 782–796. Hoff-Ginsberg, E. (1992). How should frequency in input be measured? First Language, 12, 233–244. Hoff-Ginsberg, E. (1998). The relation of birth order and socioeconomic order and socioeconomic status to children’s language experience and language development. Applied Psycholinguistics, 19, 603–630. Hoff-Ginsberg, E. (1999). Formalism or functionalism? Evidence from the study of language development. In M. Darnell, E. Moravscik, M. Noonan, F. Newmeyer, & K. Wheatley (Eds.), Functionalism and formalism in linguistics (pp. 317–340). Amsterdam, The Netherlands: John Benjamins. Hollich, G. J., Hirsh-Pasek, K., & Golinkoff, R. M. (2000). Breaking the language barrier: An emergentist coalition model for the origins of word learning. Monographs of the Society for Research in Child Development, 65(3, Serial No. 262). Huttenlocher, J., Haight, W., Bryk, A., Seltzer, M., & Lyons, T. (1991). Early vocabulary growth: Relation to language input and gender. Developmental Psychology, 27, 236–248. Jusczyk, P. (1997). The discovery of spoken language. Cambridge, MA: MIT Press. Koenig, P., & Naigles, L. (1996, April). One-word speakers interpret a novel word as an action. Poster presented at the International Conference on Infancy Studies, Providence, RI. Laakso, M.-L., Poikkeus, A.-M., Katajamaki, J., & Lyytinen, P. (1999). Early intentional communication as a predictor of language development in young toddlers. First Language, 19, 207–231. McDonald, L., & Pien, D. (1982). Mother conversational behavior as a function of interactional intent. Journal of Child Language, 9, 337–358. Miller, J., & Chapman, R. (1985). SALT: Systematic Analysis of Language Transcripts. Madison: University of Wisconsin, Waisman Center, Language Analysis Laboratory. Monnot, M. (1999). Function of infant-directed speech. Human Nature, 10, 415–443. Morgan, J., & Demuth, K. (1996). Signal to syntax: Bootstrapping from speech to grammar in early acquisition. Hillsdale, NJ: Erlbaum. Hoff and Naigles Mundy, P., & Gomes, A. (1998). Individual differences in joint attention skill development in the second year. Infant Behavior and Development, 21, 469–482. Naigles, L. (1990). Children use syntax to learn verb meanings. Journal of Child Language, 17, 357–374. Naigles, L. (1996). The use of multiple frames in verb learning via syntactic bootstrapping. Cognition, 58, 221–251. Naigles, L. (2000a). Manipulating the input: Studies of mental verb acquisition. In B. Landau, J. Jonides, E. Newport, & J. Sabini (Eds.), Cognition and language: Essays in honor of Henry and Lila Gleitman (pp. 245–274). Cambridge, MA: MIT Press. Naigles, L. (2000b). Call for information about children’s names for mom and dad, and summary thereof (7 June 2000 [call] and 8 July 2000 [summary]). Retrieved December 27, 2001, http://listserv.linguistlist.org/archives/ info-childes.html. Naigles, L. R., & Hoff-Ginsberg, E. (1998). Why are some verbs learned before other verbs? Effects of input frequency and structure on children’s early verb use. Journal of Child Language, 25, 95–120. Newport, E. L., Gleitman, H., & Gleitman, L. R. (1977). Mother, I’d rather do it myself: Some effects and noneffects of maternal speech style. In C. E. Snow & C. A. Ferguson (Eds.), Talking to children: Language input and acquisition (pp. 109–150). Cambridge, U.K.: Cambridge University Press. Oshima-Takane, Y. (1988). Children learn from speech not addressed to them: The case of personal pronouns. Journal of Child Language 15, 95–108. Oshima-Takane, Y., Goodz, E., & Derevensky, J. (1996). Birth order effects in early language development: Do secondborn children learn from overheard speech? Child Development 67, 621–634. Pearson, B. Z., Fernandez, S. C., Lewedeg, V., & Oller, D. K. (1997). The relation of input factors to lexical learning by bilingual infants. Applied Psycholinguistics, 18, 41–58. Pinker, S. (1984). Language learnability and language development. Cambridge, MA: Harvard University Press. Pinker, S. (1989). Learnability and cognition: The acquisition of argument structure. Cambridge, MA: MIT Press. Plaut, D., & Kello, C. (1999). The emergence of phonology from the interplay of speech comprehension and production: A distributed connectionist approach. In B. MacWhinney (Ed.), The emergence of language (pp. 381– 416). Hillsdale, NJ: Erlbaum. Prasada, S., & Choy, J. (1998, November). The role of syntactic structure in the interpretation of proper nouns. Paper presented at the 23rd Annual Boston University Conference on Language Development. Boston. Rice, M. (1990). Preschoolers’ QUIL: Quick incidental learning of words. In G. Conti-Ramsden, & C. Snow (Eds.), Children’s language (Vol. 7, pp. 171–195). Hillsdale, NJ: Erlbaum. Richards, B. (1987). Type/token ratios: What do they really tell us? Journal of Child Language, 14, 201–209. 433 Saffran, J., Aslin, R., & Newport, E. (1996). Statistical learning by 8-month-olds. Science, 274, 1926–1928. Schwartz, R. G., & Terrell, B. Y. (1983). The role of input frequency in lexical acquisition. Journal of Child Language, 10, 57–64. Siskind, J. M. (1996). A computational study of crosssituational techniques for learning word-to-meaning mappings. In M. R. Brent (Ed.), Computational approaches to language acquisition (pp. 39–91). Cambridge, MA: MIT Press. Smith, L. (1999). Children’s noun learning: How general learning processes make specialized learning mechanisms. In B. MacWhinney (Ed.), The emergence of language (pp. 277–304). Hillsdale, NJ: Erlbaum. Tamis-LeMonda, C. S., Bornstein, M. H., Baumwell, L., & Damast, A. M. (1996). Responsive parenting in the second year: Specific influences on children’s language and play. Early Development and Parenting, 5, 173–183. Tamis-LeMonda, C. S., Bornstein, M. H., Kahana-Kalman, R., Baumwell, L., & Cyphers, L. (1998). Predicting variation in the timing of linguistic milestones in the second year: An events-history approach. Journal of Child Language, 25, 675–700. Tardif, T., Shatz, M., & Naigles, L. (1997). Caregiver speech and children’s use of nouns versus verbs: A comparison of English, Italian, and Mandarin. Journal of Child Language, 24, 535–565. Tomasello, M. (1990). The social bases of language development. Social Development, 1, 67–87. Tomasello, M. (2001). Perceiving intentions and learning words in the second year of life. In M. Bowerman & S. Levinson (Eds.), Language acquisition and conceptual development (pp. 132–158). Cambridge, U.K.: Cambridge University Press. Tomasello, M., & Barton, M. (1994). Learning words in nonostensive contexts. Developmental Psychology, 30, 639–650. Tomasello, M., Conti-Ramsden, G., & Ewert, B. (1990). Young children’s conversations with their mothers and fathers: Differences in breakdown and repair. Journal of Child Language, 17, 115–130. Tomasello, M., & Farrar, M. J. (1986). Joint attention and early language. Child Development, 57, 1454–1463. Tomasello, M., & Todd, J. (1983). Joint attention and lexical acquisition style. First Language, 4, 197–212. Waxman, S. R., & Markow, D. (1998). Object properties and object kind: Twenty-one-month-old infants’ extension of novel adjectives. Child Development 69, 1313–1329. Weizman, Z. O., & Snow, C. E. (2001). Lexical input as related to children’s vocabulary acquisition: Effects of sophisticated exposure and support for meaning. Developmental Psychology, 37, 265–279. Woodward, A. & Markman, E. (1999). Early word learning. In D. Kuhn & R. S. Siegler (Vol. Eds.), & W. Damon (Series, Ed.), Handbook of child psychology: Vol. 2. Cognition, perception, and language (5th ed., pp. 371–420). New York: Wiley. This document is a scanned copy of a printed document. No warranty is given about the accuracy of the copy. Users should refer to the original published version of the material.