SPECIFIC EFFECT OF MODELING ON YOUNG CHILDREN'S WORD PRODUCTIONS

：芭芭拉

SPECIFIC EFFECT OF MODELING ON YOUNG CHILDREN'S WORD PRODUCTIONS
Harrie Boelens, Berend Hofman, Taiss Tamaddoni, Katja Eenink. The Psychological Record. Gambier: Winter 2007. Vol. 57, Iss. 1; pg. 145, 22 pgs

Abstract (Summary)
In naturalistic studies of early language development, researchers often require evidence of referent specificity before including an utterance as a word in the child's productive vocabulary (e.g., Bloom, Margulis, Tinker, & Fujita, 1996; Dromi, 1987; Harris, Barrett, Jones, & Brookes, 1988). Numbers of correct responses as high as these are extremely unlikely when experimental names are produced regardless of the experimental stimulus presented.

：芭芭拉

Full Text (8693  words)

Copyright The Psychological Record Winter 2007

[Headnote]
Effects of object-name contiguity on word production were examined in 11/2- to 2-yr-old children. Two objects and two spoken names were presented in each of three experiments. Each object was the referent of 1 spoken name. An object was presented either together with its spoken name (modeling trial) or alone (test trial). Modeling trials and test trials were mixed with trials that presented familiar stimuli. Referent-specific productions of the experimental names developed on the test trials. This was found Immediately after modeling (Experiment 1), and at least 1 day later (Experiments 2 and 3). Further, it was found not only when children interacted with an experimenter (Experiments 1 and 2) but also when they saw a computerized animation (Experiment 3). These findings support theoretical accounts that emphasize a role for modeling rather than consequences of behavior in the learning of object-name productions, and they have implications for the design of word modeling studies.


Children learn to produce hundreds of words during the second year of life (for reviews, see Bloom, 2000; Dromi, 1999; Tomasello, 2003; Woodward & Markman, 1998). Observers recognize the productions of the children as words, not only because they sound like conventional words, but also because they occur on particular occasions, which resemble those on which the words are conventionally appropriate. This is especially clear in the case of object names. For example, a mother may note that her 1year-old child says car in response to cars, trains, and buggies, but not in response to animals, plants, and furniture. An utterance that is produced only on a certain range of occasions (situations, conditions, contexts, etc.) may be called referent specific or may be said to be under stimulus control. Referent specificity is an important criterion when deciding whether a child's utterance should be viewed as a word or not (Vihman & McCune, 1994). In naturalistic studies of early language development, researchers often require evidence of referent specificity before including an utterance as a word in the child's productive vocabulary (e.g., Bloom, Margulis, Tinker, & Fujita, 1996; Dromi, 1987; Harris, Barrett, Jones, & Brookes, 1988).

Accounts of word learning have to explain the roles of experience in the development of referent-specific word productions. To continue the example given above: Which experiences can make a 1-year-old child say car in response to cars and to objects that resemble cars, but not in response to other objects? In empirical research, most attention has been devoted to the possible roles of hearing a word used by others. Naturalistic studies have revealed that the contexts of children's early word uses closely resemble the contexts in which mothers use the same words most frequently (Harris et al., 1988), and that the more language children hear the larger their vocabularies, or the faster their vocabulary will grow (e.g., Hart & Risley, 1995; Huttenlocher, Haight, Bryk, Seltzer, & Lyons, 1991; Pearson, Fernandez, Lewedeg, & oiler, 1997). All of these findings suggest influences of hearing words used by caregivers on word production in 1-year-olds.

Training studies have examined the kinds of word uses, demonstrated by others, that might support word learning in young children. The demonstration of the use of a word is often termed word modeling. In word modeling, participants are exposed to a relationship between a word and a target stimulus. The relationship studied most frequently is the contiguous presentation of an object and a spoken name. This may be designated object-name contiguity. For example, a mother might say car when her child is looking at a car. It has repeatedly been found that object-name contiguity can lead to comprehension of the name (i.e., pointing to or looking at the object in response to its name) in children just over 1 year old (see Schafer& Plunkett, 1998; Woodward & Hoyne, 1999; Woodward, Markman, & Fitzsimmons, 1994). One might therefore expect a similar effect to be present in production. Indeed, much research has examined effects of object-name contiguity on word production in 1-yearolds. The findings, however, lend themselves to different interpretations.

Training the production of object names in 1-year-olds: Problems with interpreting the findings

Many training studies have examined effects of object-name contiguity on word production in 1-year-olds (Leonard, Chapman, Rowan, & Weiss, 1983; Leonard, Schwartz, Folger, Newhoff, & Wilcox, 1979; Leonard, Schwartz, Morris, & Chapman, 1981; Moore, Angelopoulos, & Bennett, 1999; Namy & Waxman, 2002; Schwartz & Leonard, 1982; Schwartz & Terrell, 1983; Tomasello & Farrar, 1986; Tomasello, Strosberg, & Akhtar, 1996). It has often been found that object-name contiguity can make 1year-olds utter words similar to those that have been modeled, but it is not entirely clear how these findings should be interpreted. One problem is that few of the studies have revealed evidence for referent specificity in 1 -year-olds' word productions. This has to do with the nature of the measures employed. Early work in this area has distinguished imitative from spontaneous word productions. Imitative word productions are immediate repetitions of words uttered by others. Spontaneous word productions are those not given as immediate repetitions. Several studies have examined relationships between imitative and spontaneous word productions, or effects of independent variables on spontaneous word productions (e.g., Leonard et al., 1983; Schwartz & Leonard, 1982; Schwartz & Terrell, 1983). One important finding is that spontaneous productions of a word can be shown without any preceding imitative productions of the same word (e.g., Leonard et al., 1983). Later work has restricted the measurement of children's spontaneous word productions to the presence of referents (e.g., Tomasello & Farrar, 1986; Tomasello et al., 1996). These productions may be called referential (Namy & Waxman, 2002). This work has established that referential word productions can be brought about as a result of both contiguous and noncontiguous arrangements of words and objects (Tomasello et al., 1996). Both spontaneous and referential word productions, however, can be echoes of the names heard earlier, given regardless of the present occasion. They need not be referent-specific.

A number of studies with 1-year-olds have collected measures that allow an assessment of referent specificity. The outcomes have been mixed. Leonard et al. (1981, 1983) presented 16 experimental names, 8 for actions and 8 for objects. Each name was presented five times during each of 10 play sessions. Thereafter, each of the actions and objects was presented once, and the experimenter asked for its name (posttest). Leonard et al. determined the number of names used correctly on the posttest. This number could vary between O and 16. The means across children were 4.6 (Leonard et al., 1981), 5.9 (Leonard et al., 1983, Experiment 1), and 3.2 (Leonard et al., 1983, Experiment 2). Numbers of correct responses as high as these are extremely unlikely when experimental names are produced regardless of the experimental stimulus presented. Thus, at least some of the names must have been produced in a referent-specific manner. Moore et al. (1999) credited a child with the exclusive use of a word when the child produced it at least once in the presence of its referent, and never on other occasions. Moore et al. found such uses after contiguous object-name modeling in 2-year-old, but not in 1-year-old, children. The 1-year-olds rarely produced the experimental object name. Namy and Waxman (2002) studied the production of vocal versus gestural object names in 1- and 2-year-old children. The response measure was the proportion of object name productions given in the presence of the referent of the name. This proportion can be viewed as a graded measure of exclusivity. In 2-yearolds, the proportions were higher for vocal than for gestural object names.

In 1-year-olds, the proportions could not be determined reliably, because too few children produced the object names. The negative findings obtained with 1-year-olds (Moore et al., 1999; Namy & Waxman, 2002) suggest that it will be worthwhile to develop more sensitive techniques for revealing referent specificity in word productions. A promising alternative for the assessment of referent specificity is suggested by signal detection theory (Green & Swets, 1966; Thomas, Campos, Shucard, Ramsay, & Shucard, 1981). Signal detection theory compares the rate or probability of responding to a target stimulus (hit) with the rate or probability of responding to other, nontarget stimuli (false alarm). Thus, signal-detection theory takes the amount of exposure to target and nontarget stimuli into account, and it employs graded measures of both target and nontarget responding. Such measures were also collected in the experiments reported below.

Another problem with the interpretation of word productions in 1-yearolds concerns the sufficiency of word modeling. Leonard et al. (1981,1983) found that a modeling-based training led to referent-specific production of names for objects and actions in 1-year-olds. Their findings do not, however, allow the conclusion that modeling sufficed for this outcome. The training lasted many sessions, and word modeling was embedded in play that involved the experimental stimuli. In these circumstances, a role for the consequences of producing a word seems especially plausible. As an everyday example, consider again the child's utterance of car in the presence of a car. There are many natural reactions to this. A caregiver or experimenter may respond verbally, with a word such as yes, or car, or with a longer phrase. She may also respond nonverbally by nodding, smiling, pointing to another car, handing over a car, joining in the child's play, and so forth. Leonard et al. (1981, 1983) instructed their experimenters to avoid using experimental names in response to the behavior the children showed during play (see also Leonard et al., 1979). This minimizes a role for some consequences, but it leaves room for others, especially nonverbal ones. Natural, nonverbal consequences, such as handing over a toy, have been used successfully in word training studies with older and developmentally delayed children (e.g., Hart & Risley, 1968; Koegel, O'Dell, & Koegel, 1987; McGee, Krantz, & McClannahan, 1985; Neef, Walters, & Egel, 1984; Whitehurst & ValdezMenchaca, 1988). These findings, like those of word modeling, suggest experiences that could play a role in early normal development.

Interplay of word modeling and consequences of uttering words

Little is known about the interplay of word modeling and consequences of uttering words in the development of word productions. There are many possibilities for this interplay. Two exhaustive and mutually exclusive categories of possibilities will be distinguished here. In the first category, word modeling makes the child capable of producing the word with at least some referent specificity. For example, a child could hear somebody else say car in the presence of a car, and this, by itself, could increase the chances of the child's saying car in response to cars and to objects that look like cars, but not to other objects. Thereafter, experiencing consequences could make the child's production of the word more or less likely, and it could increase or decrease referent specificity. Many modern theoretical treatments of early word learning (e.g., Bloom, 2000; Meltzoff & Gopnik, 1-9-8-9; Smith, 2000; Tomasello, 2003) suggest that this effect of word modeling is present in the word learning of 1-year-olds.

In the second category of possibilities, word modeling does not make the child capable of producing the word with referent specificity. Other experiences would be necessary for that. These could be experiences of consequences of utterances. In this effect, presentation of the word car, in the presence of a car, could make the child more inclined to say car, but this would occur without any referent specificity. The circumstances (the presence or absence of something that looks like a car) would not yet make a difference in the child's utterances of car. One special role for consequences involves immediate imitations. Suppose a mother says car when her child is looking at a toy car. The child then says car (immediate imitation), and the mother allows her child to play with the car (consequence). As a result, there could be an increase of the child's saying car in the presence of the toy car. This is the main productive word learning process of early reinforcement formulations (Miller & Dollard, 1941; Skinner, 1957). The process remains present in modern reinforcement formulations, but these allow other possibilities too (e.g., Home & Lowe, 1996; Moerk, 1990; Whitehurst & DeBaryshe, 1-9-8-9).

The present experiments

The purpose of the present experiments was to examine whether object-name contiguity, by itself, can make 11/2- to 2-year-old children capable of producing an object name with at least some referent specificity. Two objects and two spoken names were presented in each of three experiments. Each object was the referent of one spoken name. An object was presented either together with its spoken name (modeling trial), or alone (test trial). The two kinds of trials were presented repeatedly, mixed with trials that presented familiar stimuli. The test trials made it possible to compare the probability of a hit with the probability of a false alarm in the child's production of each experimental name. In Experiment 1, two test trials followed immediately after two modeling trials. This experiment was done in a tabletop setting, with an experimenter facing a child. The experimenter was instructed to avoid procedures other than presenting the names in the presence of their referents. Under these circumstances, if the children would learn to say each name in the presence of its target stimulus, and not in the presence of the other stimulus, then that would suggest that object-name contiguity, by itself, can bring about referent specificity.

There could still, however, be important limitations on these findings, and two other experiments were conducted to explore them. One possibility is that the effects of object-name contiguity on referent specificity do not last long. Experiment 2 therefore arranged delays between modeling and testing that lasted at least 1 day. Another possibility is that consequences were provided unintentionally. In Experiment 3, therefore, the main procedures were automated. The development of referent specificity after long delays, in automated procedures, would suggest that modeling is a powerful mechanism in 1-year-olds' learning of productive object names.

：芭芭拉

☆Experiment 1

Method

Children and Settings

Day-care centers in the vicinity of Leiden, Netherlands, were asked to provide children. Children could participate if they were (a) between 18 months and 24 months old and (b) capable of naming at least eight depicted objects. The second requirement was made so that the experimental (modeling and testing) procedures could be embedded in a naming task the children were capable of. This was done to maintain an interest in the experimental procedures. In addition, permission by the parents was required.

Three day-care centers together provided 9 children in the appropriate age range (M = 21 months 16 days; range 19 months 20 days to 23 months 15 days; 4 girls, 5 boys). We asked the parents of these children for the words that they had ever heard their children say in response to depicted objects. All children had at least eight of these words.

A female experimenter visited the day-care centers. She played for some time with a child in the room that the child shared with peers and caregivers. She then accompanied the child to a room available especially for the experiment. Here, the child was seated at a table, facing the experimenter.

Materials

Fifty-four picture cards of the Ravensburger Dick Bruna Lottino game, and two similar cards with pictures of strange objects (S^sub 1^ or S^sub 2^) were used. Figure 1 shows S^sub 1^ and S^sub 2^; they differed not only in shape but also in coloring. All cards were 54 × 54 mm. Each card was presented in a lightblue cardboard frame (148 × 105 mm). The frame was presented with its long side horizontally. In the center of each frame was a recess of 58 × 58 mm, where the cards could be temporarily fixed with a spray. Each recess had a cover that could be folded away.

Procedure

The experiment was divided into two phases: (1) Vocabulary Assessment and (2) Modeling and Testing.

Vocabulary assessment. This phase was arranged to find pictures that the children had names for. Sixteen different Lottino pictures were used. The pictures were presented on trials. At the beginning of each trial, the experimenter presented a frame and folded the cover away so that a picture became visible. She then said Wat is dat? "What is that?" Now there were three possibilities; consider the picture of a car as an example. (1) The child said auto "car" within 2 s. The experimenter then said Goed zo, 't is een auto! "Good, it's a car!" (2) The child said something other than auto "car" within 2 s. The experimenter then said 't Is een auto. "It's a car." (3) The child said nothing within 2 s. The experimenter then said 't Is een auto. "It's a car". After any of these three possibilities the experimenter removed the frame with the card. In cases (2) and (3), if the child imitated the experimenter, the experimenter said Goed zo, 't is een auto! "Good, it's a car!" Similar procedures were followed with other pictures. Two sessions were arranged. Each session had 16 trials (16 different pictures) in a random order. Random orders differed between sessions.

Modeling and testing. This phase was arranged to assess the effects of word modeling. Stimuli S^sub 1^ and S^sub 2^, and all stimuli that had been named appropriately in both of the vocabulary assessment sessions, were used. Three kinds of trial were presented: modeling trials, test trials, and familiar-picture trials. Modeling trials went as follows. The experimenter showed a frame and folded the cover away so that a stimulus (either S^sub 1^ or S^sub 2^) became visible. As soon as the child looked at the stimulus, the experimenter said een /en/ "a" followed by either /tat/, designated W^sub 1^, or /mum/, designated W^sub 2^ (not Dutch words). For 5 children, W^sub 1^ was presented in the presence of S^sub 1^, and W^sub 2^ in the presence of S^sub 2^. For the other 4 children, W^sub 1^ was presented in the presence of S^sub 2^, and W^sub 2^ in the presence of S^sub 1^. The experimenter took away the frame with the card 2 s after she had presented W^sub 1^ or W^sub 2^. Test trials also presented either S^sub 1^ or S^sub 2^, but now the experimenter said Wat is dat? "What is that?" as soon as the child looked at the stimulus. She took away the frame with the card 2 s after this question. Familiar-picture trials presented stimuli the children had names for. These trials went as in vocabulary assessment.

Ten sessions were arranged. Table 1 (middle column) illustrates their structure. Each session consisted of two blocks of eight trials. Trials 1, 2, 7, and 8 of each block were familiar-picture trials. Trials 3 and 4 were modeling trials (S^sub 1^ or S^sub 2^, each once, random order), and Trials 5 and 6 test trials (S^sub 1^ or S^sub 2^, each once, random order). Random orders varied from block to block. The experiment ended after 10 sessions of modeling and testing. Thus, 40 modeling trials and 40 test trials were presented. Throughout the experiment, the interval between taking away a card and presenting the next card lasted at least 3 s. The experimenter presented the experimental words only on modeling trials, as described above, and avoided responding to the children's utterances. Sessions were held 1 to 4 days a week, at most once a day, and at approximately the same time of the day.

Data Collection, Dependent Variables, and Reliability

All sessions were recorded on audiotape. Two observers listened to the audiotapes. One observer listened to all recordings; the other to the recordings of 5 randomly selected children. The observers did not know which visual stimuli were presented. W^sub 1^ was said to have occurred on a test trial when /ta/ or /at/ was heard among the speech sounds uttered by the child; W^sub 2^ when /mu/ or /um/ was heard. All speech sounds uttered within 5 s from the question Wat is dat? "What is that?" were taken into consideration. Note that both words could occur on the same trial. Multiple utterances of the same word could also occur, but these were not counted. The basic dependent variables of the experiment were four proportions: the proportion of test trials with S^sub 1^ that evoked W^sub 1^, the proportion of test trials with S^sub 1^ that evoked W^sub 2^, the proportion of test trials with S^sub 2^ that evoked W^sub 1^, and the proportion of test trials with S^sub 2^ that evoked W^sub 2^. These may be abbreviated P(W^sub 1^|S^sub 1^), P(W^sub 2^|S^sub 1^), P(W^sub 1^|S^sub 2^) and P(W^sub 2^|S^sub 2^). Each proportion could be either 0, 0.5, or 1.0 for each session. Two of the four proportions are measures of the tendency to utter a word in the presence of its target stimulus (i.e., the stimulus that had been paired with the word). The other two proportions are measures of the tendency to utter a word in the presence of the other, nontarget stimulus.

Figure 2 illustrates the calculation of dependent variables for a child hearing W^sub 1^ in the presence of S^sub 1^ and W^sub 2^ in the presence of S^sub 2^. For this child, the target proportions are P(W^sub 1^|S^sub 1^) and P(W^sub 2^|S^sub 2^); the other, nontarget proportions are P(W^sub 1^|S^sub 2^) and P(W^sub 2^|S^sub 1^). Hypothetical data are shown for the test trials of two sessions. On Session 1, the first test trial presents S^sub 1^, and it evokes W^sub 1^ and W^sub 2^ successively, both within 5 s from the question Wat is dat? The other test trials present S^sub 2^, S^sub 1^, and S^sub 2^, and they evoke no utterances. As a result, P(W^sub 1^|S^sub 1^) and P(W^sub 2^|S^sub 1^) are 0.5; P(W^sub 2^|S^sub 2^) and P(W^sub 1^|S^sub 2^) are zero. The mean across the target proportions, abbreviated P(W|target), and the mean across the other, nontarget proportions, P(W|Other), are both 0.25. The difference between these two means (i.e., target minus other) is a measure of referent specificity (see Thomas et al., 1981, for a similar analysis, but for word comprehension). Here, the difference is zero, and that expresses that there is no evidence for referent specificity in these data. On Session 2, the first test trial (with S^sub 2^) evokes utterances of W^sub 1^, W^sub 2^, and W^sub 1^, successively, the second test trial (S^sub 1^) one utterance of W^sub 1^, the third test trial (S^sub 1^) no utterances, and the fourth test trial (S^sub 2^) two utterances of W^sub 2^. Multiple utterances of the same word are not counted. As a result, P(W^sub 1^|S^sub 1^) and P(W^sub 1^|S^sub 2^) are 0.5, P(W^sub 2^|S^sub 2^) is 1.0, and P(W^sub 2^|S^sub 1^) is zero. The mean across the target proportions is 0.75 and the mean across the other, nontarget proportions is 0.25. Here, the difference is 0.5, and that does indeed suggest referent specificity.

Reliability measures were obtained by dividing the number of agreements between the two observers by the number of agreements plus the number of disagreements. Reliability was 92% for W^sub 1^ and 96% for W^sub 2^ (both percentages based on 200 test trials). The data obtained by the first observer are reported below.

Results and Discussion

Figure 3 shows P(W|Target) and P(W|Other). Data points represent means across 9 children, two words, and two consecutive sessions. At Time 1 (first two sessions), the proportions were .32 (target) and .24 (other). Target responding increased, and other responding decreased, after that time. An analysis of variance revealed a significant effect of Target (i.e., target vs. other), F(1, 8) = 57.26, p < .001, and a significant Target × Time interaction, F(4, 32) = 7.16, p < .001, but no effect of Time, p > .05. The differences between target and other responding were significant at Times 2 through 5, t(8) ≥ 5.34, ps < .001.

Nonparametric analyses had similar outcomes. Wilcoxon matchedpairs signed-ranks tests revealed significant differences between target and other responding at Times 2 through 5, Wilcoxon T = 0, ps < .05. Further, the differences were seen in almost all children. The final pairs of proportions (target, other) for individual children at Time 5 were (.88, 0), (1, .5), (.5, 0), (.75, .13), (.5, 0), (.88, .13), (.25, .25), (1, 0), and (.75, .13). Thus, 8 of 9 children showed more target than other responding at Time 5. One child (.25, .25) showed no difference at that time.

Similar outcomes were also obtained with a more conservative coding scheme, in which only the complete words /tat/ and /mum/ were scored. This led to lower levels or responding, with mean levels of .49 (target) and .08 (other) at Time 5. The pattern of the findings remained the same. An analysis of variance revealed a significant effect of Target, F(1, 8) = 20.33, p < .01, and a significant Target x Time interaction, F(4, 32) = 3.83, p < .05, but no effect of Time, p > .05. The differences between target and other responding were significant at Times 2 through 5, t(8) ≥ 4.12, ps < .01. Some sounds excluded by the more conservative coding scheme were /ta/, /dat/, /pum/, and /mu/. The sounds /ta/ and /dat/ were uttered primarily when /tat/ was correct; /pum/ and /mu/ when /mum/ was correct. Other sounds were excluded by both schemes. Examples were /mam/ and /pu/. One child produced /mam/ on 17 test trials with the target of /mum/ and never on other test trials. Another child produced /pu/ on seven test trials with the target of /mum/ and on three test trials with the target of /tat/.

The findings suggest that object-name contiguity, by itself, can bring about referent specificity in 1-year-olds' word productions. It is possible, however, that the modeling procedures were only short-acting. Two test trials followed two modeling trials, in each of two blocks of trials. In each block of trials therefore, the delay between modeling and testing was only in the order of seconds. It might be the case that longer delays-more representative of word learning in everyday life-do not lead to referentspecific productions. Experiment 2 was conducted to investigate this. Test trials were now presented before modeling trials. The delay between one session and the next was at least 1 day. Therefore, at least 1 day elapsed between modeling and testing.

：芭芭拉

☆Experiment 2

Method

Children and Settings

Children had to meet the same criteria as those of Experiment 1. Eight children were found in two day-care centers (M age = 20 months 27 days; range 19 months 10 days to 22 months 0 days; 6 girls, 2 boys). We asked the parents of these children for the words that they had ever heard their children say in response to depicted objects. Each child had at least eight of these words.

Materials

Materials were the same as in Experiment 1.

Procedure

The procedure was the same as in Experiment 1, with the following exceptions. Session 1 of modeling and testing arranged only one block of eight trials. Trials 3-6 of this block were all modeling trials. Sessions 2-9 of modeling and testing all arranged two blocks of eight trials. Table 1 (last column) illustrates the structure of these sessions. Trials 3-6 of the first block were test trials; Trials 3-6 of the second block were modeling trials. Session 10 again arranged only one block of eight trials. Trials 3-6 of this block were all test trials. In each block of trials arranged during modeling and testing, S^sub 1^ and S^sub 2^ each occurred twice, in a random order. Thus, 36 modeling trials and 36 test trials were presented. For 4 children, W^sub 1^, was presented in the presence of S^sub 1^, and W^sub 2^ in the presence of S^sub 2^. For the other 4 children, W1 was presented in the presence of S^sub 2^, and W^sub 2^ in the presence of S^sub 1^.

Data Collection, Dependent Variables, and Reliability

Data collection was the same as in Experiment 1. Proportions P(W^sub 1^|S^sub 1^), P(W^sub 2^|S^sub 1^), P(W^sub 1^|S^sub 2^), and P(W^sub 2^|S^sub 2^) were determined for blocks of two or three consecutive sessions with test trials. The blocks were Sessions 2 and 3, Sessions 4 and 5, Sessions 6 and 7, and Sessions 8 through 10 (note that Session 1 did not have test trials). A second observer listened to the audiorecordings of 3 children. Reliability was 94% for W^sub 1^ and 93% for W^sub 2^ (both percentages based on 108 trials).

Results and Discussion

Figure 4 shows P(W|Target) and P(W|Other). Data points represent means across 8 children, two words, and two or three consecutive sessions (see Method). There were almost no word productions at Time 1 (Sessions 2 and 3). This may be compared with the mean levels of .32 (target) and .24 (other) at Time 1 in Experiment 1 (see Figure 3). The findings obtained at Time 1 in the two experiments together suggest that nonspecific word production is especially likely immediately after early modeling. Target responding increased, and other responding remained low, after Time 1. An analysis of variance revealed significant effects of Target, F(1, 7) = 33.33, p < .001, Time, F(3, 21) = 51.07, p < .001, and Target x Time, F(3, 21) = 19.10, p < .001. The differences between target and other responding were significant at Times 2 through 4, t(7) ≥ 4.63, ps < .01.

Nonparametric analyses had similar outcomes. Wilcoxon matchedpairs signed-ranks tests revealed significant differences between target and other responding at Times 2 through 4, Wilcoxon T ≤ 1, ps < .05. Further, the differences were seen in almost all children. The final pairs of proportions (target, other) for individual children at Time 4 were (1, 0), (1, 0), (.92, .08), (.92, 0), (.67, .17), (1, 0), (.17, .25), and (.75, 0). Thus, 7 of 8 children showed more target responding than other responding at Time 4. The remaining child (.17, .25) showed more other responding at that time, but the absolute difference between the two proportions was the smallest for this child.

Similar outcomes were obtained when only the complete words /tat/ and /mum/ were scored. This led to lower levels or responding, with mean levels of .57 (target) and .04 (other) at Time 4. The pattern of the findings remained the same. An analysis of variance revealed significant effects of Target, F(1, 7) = 23.82, p < .01, Time, F(3, 21) = 24.70, p < .001, and Target × Time, F(3, 21 ) = 13.71, p < .001. The differences between target and other responding were significant at Times 2 through 4, t(7) ≥ 4.02, ps < .01.

The findings of Experiments 1 and 2 show that modeling during interaction with an experimenter can have immediate as well as delayed effects on word production in 1&frac12;- to 2-year-old children. Further, the findings suggest that modeling by itself can bring about referent specificity in the word productions of these children. It could still be argued, however, that consequences may have been provided unintentionally. Therefore, the involvement of an experimenter was strongly reduced in Experiment 3.

☆Experiment 3

Method

Children and Settings

Children had to meet the same criteria as those of the previous experiments. Eight children were found in two day-care centers (M age = 22 months 5 days; range 20 months 3 days to 22 months 29 days; 5 girls, 3 boys). We asked the parents of these children for the words that they had ever heard their children say in response to depicted objects. All children had at least eight of these words.

Apparatus and Materials

As in Experiments 1 and 2, Lottino cards were used to find pictures the children had names for. Further, the pictures used in the previous experiments were scanned and shown on a Windows computer.

Procedure

As in Experiments 1 and 2, there were two phases: (1) Vocabulary Assessment and (2) Modeling and Testing. Vocabulary assessment went as before. Modeling and testing, however, was done with computerized animations. The animations were programmed in Macromedia Flash and presented via the computer's Media Player.

The child was seated between the computer screen and the experimenter (child and experimenter both faced the screen). An animation began with the presentation of a puppet, who introduced himself as Charlie. This was followed by a block of eight trials. As in Experiments 1 and 2, there were three kinds of trials: modeling trials, test trials, and familiar-picture trials. Modeling trials began with Charlie shown slightly to the left of the middle of the screen and facing the child (1 s). Charlie then looked and pointed to the right. At the same time, he presented a label in an exclamation. This was either Kijk, een /tat/! "Look, a /tat/!" (W^sub 1^), or Kijk, een /mum/! "Look, a /mum/!" (W^sub 2^). Immediately thereafter, S^sub 1^ or S^sub 2^ appeared in the top-right corner of the screen. For 4 children, W^sub 1^ was presented in the presence of S^sub 1^, and W^sub 2^ in the presence of S^sub 2^. For the other 4 children, W^sub 1^ was presented in the presence of S^sub 2^, and W^sub 2^ in the presence of S^sub 1^. The stimulus moved down, circling around its center, in 2 s. It remained for 3 s in the bottom-right position and then disappeared. Charlie then withdrew his hand and looked towards the child again (intertrial interval, 1 s). Test trials were similar, but now the presentation of a label was replaced by the question Wat is dat? "What is that?" Familiar-picture trials went like test trials, but now the experimenter responded to the child, as in Experiments 1 and 2. On all kinds of trials, the experimenter pointed to the screen, and said Kijk "Look", when the child stood up from the chair, or turned around.

The present procedures did not ensure that a child saw an experimental stimulus when a label was presented. Sometimes a child looked elsewhere, and then (within 5 s) oriented to the screen. This occurred on 11% of all modeling trials (mean taken across children; range 3 - 20%). Sometimes the target stimulus was not seen at all. This occurred on 4% of all modeling trials (range 0 -18%).

Each session presented one or two blocks of eight trials. The structures of the experimental sessions (arrangements of trials in blocks) were as in Experiment 2, but now there were 10 sessions with test trials (11 experimental sessions in total). Thus, 40 test trials were arranged, in 10 sets of four, and each set of four test trials followed at least 1 day after four modeling trials.

Data Collection, Dependent Variables, and Reliability

All sessions were recorded on videotape, with the camera placed so that the screen was not visible. Two observers saw all recordings. Dependent variables were defined as in Experiment 1. Reliability was 98% for W^sub 1^ and 99% for W^sub 2^ (both percentages based on 320 trials in total). The children talked less than those studied in the previous experiments. We therefore calculated reliabilities again using only those trials on which at least one observer had heard an experimental word. Reliability was 88% for W^sub 1^, when only those trials were included on which at least one observer had heard W^sub 1^ (59 trials). Reliability was 96% for W^sub 2^, when only those trials were included on which at least one observer had heard W^sub 2^ (68 trials).

Results and Discussion

Figure 5 shows P(W|Target) and P(W|Other). Data points represent means across 8 children, two words, and two consecutive sessions (see Method). Levels of responding were low at Time 1 (target: .17; other: .06). Thereafter, target responding showed a slow increase, while other responding remained low. An analysis of variance revealed significant effects of Target, F(1, 7) = 8.92, p < .05, Time, F(4, 28) = 7.45, p < .001, and Target × Time, F(4, 28) = 5.35, p < .01. The differences between target and other responding were significant at Times 4 and 5, f(7) ≥ 2.62, ps < .05.

Nonparametric analyses had similar outcomes. Wilcoxon matchedpairs signed-ranks tests revealed significant differences between target and other responding at Times 4 and 5, Wilcoxon T = 0, ps < .05. Further, the differences were seen in almost all children. The final pairs of proportions (target, other) for individual children at Time 5 were (.75, 0), (1, 0), (.5, 0), (.25, .13), (.13, .13), (1, 0), (.38, .25), and (.5, .25). Thus, 7 of 8 children showed more target than other responding at Time 5. The remaining child (.13, .13) showed no difference at that time.

Similar outcomes were obtained when only the complete words /tat/ and /mum/ were scored. This led to slightly lower levels or responding, with mean levels of .44 (target) and .08 (other) at Time 5. The pattern of the findings remained the same. An analysis of variance revealed significant effects of Target, F(1, 7) = 11.17, p < .05; Time, F(4, 28) = 5.12, p < .01 ; and Target × Time, F(4, 28) = 6.09, p < .01. The differences between target and other responding were significant at Times 3 through 5, t(7) ≥ 2.39, ps < .05.

The involvement of an experimenter was strongly reduced in the present experiment, and it is not plausible that consequences of utterances had a role in the development of referent specificity. Therefore, the present findings suggest more strongly than those of the previous two experiments that modeling by itself can bring about referent specificity in the word productions of 1-year-olds. Computerized animations seem interesting not only because they allow one to reduce the involvement of an experimenter, but also because children encounter similar situations in their everyday lives (when watching television, or playing with computers or electronic toys). The present findings suggest that referent specificity of children's word productions can develop in these situations (see Meltzoff, 1988a, for effects of televised stimulus presentations on referent specificity in nonverbal behavior). Further, the findings extend the generality of referent specificity produced by word modeling to wordreferent relations in which a word is presented before its referent (for similar procedures, embedded in play, see: Akhtar, Jipson, & Callanan, 2001 ; Akhtar & Tomasello, 1996; Tomasello & Barton, 1994; Tomasello & Kruger, 1992; Tomasello et al., 1996).

：芭芭拉

General Discussion

Effects of object-name contiguity on word production were studied in 1&frac12;- to 2-year-old children. Two objects and two spoken names were presented in each of three experiments. An object was presented either together with its spoken name (modeling trial), or alone (test trial). The children interacted with an experimenter in Experiments 1 and 2, but the experimenter was instructed to avoid responding to the child's utterances. The involvement of an experimenter was minimized in Experiment 3, which exposed the children to a computerized animation. Efforts were thus made to exclude a role for consequences in the development of children's productions of experimental names on the test trials. Nevertheless, referent-specific word productions developed in all experiments. These findings suggest that modeling on its own can bring about referentspecific speech in 1&frac12;- to 2-year-old children.

The present findings support accounts that emphasize a role for word modeling in the learning of 1-year-olds' word productions. There are many of these accounts. Their terms and viewpoints vary widely, but they share the assumption that learning takes place as a result of exposure to a relation between a word, spoken by a caregiver or experimenter, and a referent (e.g., Bloom, 2000; Meltzoff & Gopnik, 1-9-8-9; Smith, 2000; Tomasello, 2003). These accounts do not deny roles for the consequences of utterances, but they would not view consequences as critical in the early stages of learning to produce a new object name.

This lack of a critical role for consequences helps to make sense of children's language learning when nobody is deliberately trying to teach it. Children often seem to learn language in unstructured situations that do not involve a caregiver who provides both antecedent stimuli and consequences (for reviews, see Akhtar et al., 2001; Lieven, 1994). For example, children can learn correct uses of personal pronouns by overhearing father and mother (Oshima-Takane, 1988). In these unstructured situations it will be useful to be able to learn to produce a word more or less correctly on the basis of modeling only. Caregivers may experience the more-or-less correct uses on later occasions and then provide consequences that support the utterances. Thus, the referent specificity brought about by word modeling can bridge a gap between an earlier experience, when the child hears a word for the first time, and a later experience, when a caregiver is capable of responding appropriately.

Consequences of behavior might be important not only in the maintenance of word productions, but also in the development of prerequisites for word learning. Consider the prerequisite of being able to learn by observation. There is much evidence for this ability in children younger than 1&frac12; years (e.g., Meltzoff, 1985, 1988a, 1988b, 1988c). The ability to learn by observation might itself be learned, and the consequences of behavior might be important in this. It has often been found that children's imitations can be strengthened as a result of their consequences, and that this effect spreads to other, untrained imitations. The spread to untrained imitations has been termed generalized imitation (for reviews, see Baer & Deguchi, 1985; Poulson, Kyparissos, Andreatos, Kymissis, & Parnes, 2002). Again, the evidence for generalized imitation includes children younger than 1&frac12; years (Poulson, Kymissis, Reeve, Andreatos, & Reeve, 1991; Poulson et al., 2002). Thus, it could be the case that consequences of behavior, even though not critical during the early stages of learning new object names, are important before that time.

The present experiments successfully employed methods suggested by signal-detection theory (Green & Swets, 1966; Thomas et al., 1981). The production of two words was registered in the presence of each of two stimuli. This made it possible to compare, for each of two words, the probability of production of a word in the presence of a target stimulus (hit) with the probability of production of a word in the presence of a nontarget stimulus (false alarm). The comparisons led to the finding of referent-specific word productions in each of three experiments, each with a small number of participants. The technique holds promise for extension in at least two directions. First, the technique could be applied to kinds of word modeling other than object-name contiguity. More complex, noncontiguous arrangements could also be studied, and these might involve referents other than objects (e.g., Akhtar, Carpenter, & Tomasello, 1996; Akhtar & Tomasello, 1996; Baldwin, 1991; Tomasello & Barton, 1994). Second, the technique could be applied to observations of spontaneous speech. For example, consider the procedures of Akhtar et al. (1996), who presented four experimental toys. One of these was the referent of an experimental word during a modeling period. This was followed by free access to the toys during a play period. In this case, a suitable registration procedure could be to divide the free play period into small intervals of equal duration, and to assign these intervals to the experimental toys on the basis of the child's contact with the toys (e.g., eye contact or manipulation). Referent specificity can then be assessed by comparing the proportion of intervals with target utterances with the proportion of intervals with nontarget utterances.

The relationship between the sounds presented by the experimenter (or apparatus) and the sounds produced by the children deserves further study. In our original coding scheme, we allowed incomplete approximations of the experimental words (/ta/ or /at/ instead of /tat/; /mu/ or /um/ instead of /mum/). We obtained similar patterns of effects with a more conservative scheme, but there are reasons for preferring our original scheme, or one that accepts even more (such as /pu/ instead of /mum/; see Experiment 1). Perhaps, the main issue should be whether or not the child's speech sounds are the result of word modeling. The nature of the coding scheme employed then becomes unimportant (see Rescorla & Holland, 1976, for a similar argument in connection with classical conditioning).

In conclusion, the present findings suggest that modeling on its own can bring about referent-specific word productions in 1&frac12;- to 2-year-old children. The arrangement of two word-object pairings, with measurement of both words in the presence of each stimulus, seems to be a promising method for studying early word production.

[Reference]
References
AKHTAR, N., CARPENTER, M., & TOMASELLO, M. (1996). The role of discourse novelty in early word learning. Child Development, 67, 635-645.
AKHTAR, N., JIPSON, J., & CALLANAN, M. A. (2001). Learning words through overhearing. Child Development, 72, 416-430.
AKHTAR, N., & TOMASELLO, M. (1996). Two-year-olds learn words for absent objects and actions. British Journal of Developmental Psychology, 14, 79-93.
BAER, D. M., & DEGUCHI, H. (1985). Generalized imitation from a radical-behavioral viewpoint. In S. Reiss & R. Bootzin (Eds.), Theoretical issues in behavior therapy (pp. 179-217). New York: Academic Press.
BALDWIN, D. A. (1991). Infants' contribution to the achievement of joint reference. Child Development, 62, 875-890.
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, P. (2000). How children learn the meanings of words. Cambridge, MA: MIT Press.
DROMI, E. (1987). Early lexical development. Cambridge: Cambridge University Press.
DROMI, E. (1999). Early lexical development. In M. Barrett (Ed.), The development of language: Studies in developmental psychology (pp. 99-131). Philadelphia: Psychology Press.
GREEN, D. M., & SWETS, J. A. (1966). Signal detection theory and psychophysics. New York: Wiley.
HARRIS, M., BARRETT, M., JONES, D., & BROOKES, S. (1988). Linguistic input and early word meaning. Journal of Child Language, 15, 77-94.
HART, B., & RISLEY, T. R. (1995). Meaningful differences in the everyday experience of young American children. Baltimore: Paul Brookes.
HART, B. M., & RISLEY, T. R. (1968). Establishing use of descriptive adjectives in the spontaneous speech of disadvantaged preschool children. Journal of Applied Behavior Analysis, 1, 109-120.
HORNE, P. J., & LOWE, C. F. (1996). On the origins of naming and other symbolic behavior. Journal of the Experimental Analysis of Behavior, 65, 185-241.
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.
KOEGEL, R. L, O'DELL, M. C., & KOEGEL, L K. (1987). A natural language teaching paradigm for nonverbal autistic children. Journal of Autism and Developmental Disorders, 17, 187-200.
LEONARD, L B., CHAPMAN, K., ROWAN, L E., & WEISS, A. L (1983). Three hypotheses concerning young children's imitations of lexical items. Developmental Psychology, 19, 591-601.
LEONARD, L B., SCHWARTZ, R. G., FOLGER, M. K., NEWHOFF, M., & WILCOX, M. J. (1979). Children's imitations of lexical items. Child Development, 50, 19-27.
LEONARD, L B., SCHWARTZ, R. G., MORRIS, B., & CHAPMAN, K. (1981). Factors influencing early lexical acquisition: Lexical orientation and phonological composition. Child Development, 52, 882-887.
LIEVEN, E. V. M. (1994). Crosslinguistic and crosscultural aspects of language addressed to children. In C. Gallaway & B. J. Richards (Eds.), Input and interaction in language acquisition (pp. 56-73). Cambridge: Cambridge University Press.
MCGEE, G. G., KRANTZ, P. J., & MCCLANNAHAN, L E. (1985). The facilitative effects of incidental teaching on preposition use by autistic children. Journal of Applied Behavior Analysis, 18, 17-31.
MELTZOFF, A. N. (1985). Immediate and deferred imitation in 14-month-old and 24-month-old infants. Child Development, 56, 62-72.
MELTZOFF, A. N. (1988a). Imitation of televised models by infants. Child Development, 59, 1221-1229.
MELTZOFF, A. N. (1988b). Infant imitation after a 1-week delay: Long-term memory for novel acts and multiple stimuli. Developmental Psychology, 24, 470-476.
MELTZOFF, A. N. (1988c). Infant imitation and memory: Nine-month-olds in immediate and deferred tests. Child Development, 59, 217-225.
MELTZOFF, A. N., & GOPNIK, A. (1-9-8-9). On linking nonverbal imitation, representation, and language learning in the first two years of life. In G. E. Speidel & K. E. Nelson (Eds.), The many faces of imitation in language learning (pp. 23-51). New York: Springer-Verlag.
MILLER, N. E., & DOLLARD, J. (1941). Social learning and imitation. New Haven: Yale University Press.
MOERK, E. L. (1990). Three-term contingency patterns in mother-child verbal interactions during first-language acquisition. Journal of the Experimental Analysis of Behavior, 54, 293-305.
MOORE, C., ANGELOPOULOS, M., & BENNETT, P. (1999). Word learning in the context of referential and salience cues. Developmental Psychology, 35, 60-68.
NAMY, L. L., & WAXMAN, S. R. (2002). Patterns of spontaneous production of novel words and gestures within an experimental setting in children ages 1;6 and 2;2. Journal of Child Language, 29, 911 -921.
NEEF, N. A., WALTERS, J., & EGEL, A. L. (1984). Establishing generative yes/no responses in developmentaly disabled children. Journal of Applied Behavior Analysis, 17, 453-460.
OSHIMA-TAKANE, Y. (1988). Children learn from speech not addressed to them: The case of personal pronouns. Journal of Child Language, 15, 95-108.
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.
POULSON, C. L., KYMISSIS, E., REEVE, K. F., ANDREATOS, M., & REEVE, L. (1991). Generalized vocal imitation in infants. Journal of Experimental Child Psychology, 51, 267-279.
POULSON, C. L., KYPARISSOS, N., ANDREATOS, M., KYMISSIS, E., & PARNES, M. (2002). Generalized imitation within three response classes in typically developing infants. Journal of Experimental Child Psychology, 81, 341-357.
RESCORLA, R. A., & HOLLAND, P. C. (1976). Some behavioral approaches to the study of learning. In M. R. Rosenzweig & E. L. Bennett (Eds.), Neural mechanisms of learning and memory (pp. 165-192). Cambridge, MA: MIT Press.
SCHAFER, G., & PLUNKETT, K. (1998). Rapid word learning by fifteen-montholds under tightly controlled conditions. Child Development, 69, 309-320.
SCHWARTZ, R. G., & LEONARD, L. B. (1982). Do children pick and choose? An examination of phonological selection and avoidance in early lexical acquisition. Journal of Child Language, 9, 319-336.
SCHWARTZ, R. G., & TERRELL, B. Y. (1983). The role of input frequency in lexical acquisition. Journal of Child Language, 10, 57-64.
SKINNER, B. F. (1957). Verbal behavior. Englewood Cliffs, NJ: Prentice-Hall.
SMITH, L. B. (2000). Learning how to use words: An associative crane. In R. Michnik-Golinkoff, K. Hirsh-Pasek, L. Bloom, L. B. Smith, A. L. Woodward, M. Tomasello, & G. Hollich (Eds.), Becoming a word learner: A debate on lexical acquisition (pp. 51-80). Oxford: Oxford University Press.
THOMAS, D. G., CAMPOS, J. J., SHUCARD, D. W., RAMSAY, D. S., & SHUCARD, J. (1981). Semantic comprehension in infancy: A signal detection analysis. Child Development, 52, 798-803.
TOMASELLO, M. (2003). Constructing a language: A usage-based theory of language acquisition. Cambridge, MA: Harvard University Press.
TOMASELLO, M., & BARTON, M. (1994). Learning words in nonostensive contexts. Developmental Psychology, 30, 639-650.
TOMASELLO, M., & FARRAR, M. J. (1986). Joint attention and early language. Child Development, 57, 1454-1463.
TOMASELLO, M., & KRUGER, A. C. (1992). Joint attention on actions: Acquiring verbs in ostensive and non-ostensive contexts. Journal of Child Language, 19, 311-333
TOMASELLO, M., STROSBERG, R., & AKHTAR, N. (1996). Eighteen-month-old children learn words in non-ostensive contexts. Journal of Child Language, 23, 157-176.
VIHMAN, M. M., & MCCUNE, L. (1994). When is a word a word? Journal of Child Language, 21, 517-542.
WHITEHURST, G. J., & DEBARYSHE, B. D. (1-9-8-9). Observational learning and language acquisition: Principles of learning, systems and tasks. In G. E. Speidel & K. E. Nelson (Eds.), The many faces of imitation in language learning (pp. 251-276). New York: Springer Verlag.
WHITEHURST, G. J., & VALDEZ-MENCHACA, M. C. (1988). What is the role of reinforcement in early language acquisition? Child Development, 59, 430-440.
WOODWARD, A. L., & HOYNE, K. L. (1999). Infants' learning about words and sounds in relation to objects. Child Development, 70, 65-77.
WOODWARD, A. L., & MARKMAN, E. M. (1998). Early word learning. In D. Kuhn & R. S. Siegler (Eds.), Handbook of child psychology, 5th éd., Vol. 2: Cognition, perception, and language (pp. 371-420). New York: John Wiley & Sons.
WOODWARD, A. L., MARKMAN, E. M., & FITZSIMMONS, C. M. (1994). Rapid word learning in 13- and 18-month-olds. Developmental Psychology, 30, 553-566.

[Author Affiliation]
HARRIE BOELENS, BEREND HOFMAN, TAISS TAMADDONI, and KATJA EENINK
Leiden University

[Author Affiliation]
Correspondence may be sent to Harrie Boelens, Department of Psychology, Leiden University, Wassenaarseweg 52, 2333 AK Leiden, Netherlands. (E-mail to boelens@fsw. leidenuniv.nl).