Method

Testing procedure

Each child was individually tested in a separate room in her kindergarten. Sessions took about 15–20 min and were videotaped. Each session began with a warm-up period lasting a few minutes. During this period children were asked questions about objects generally similar to the ones they would get during testing, with no feedback from experimenters.

The experiment was framed as a series of three games in which a toy clown (manipulated by E1) showed the child activities that he performed in each of the three different environments (i.e., his circus, the playground, his living room). Within a given environment, each activity was performed with one particular target object that is naturally associated with the activity, for example, for cleaning the floor of the living room the clown used a broom, for putting his dog to sleep he had a pillow. The natural relation between activity and target object was designed to help children remember the objects and their names during elicitation of reference in the test phase. After demonstrating an activity (e.g., sweeping the floor), E1 remarked that the target object (the broom) can be put away now, because the clown wants to show the child another activity. Having said this, E1 put the target object up on the shelf, making sure that the child saw where she put it, and then went on to demonstrate the next activity with the next target object. Each environment had associated with it three activities and corresponding target objects. For the test, after E1 and the child had finished with the three activities/targets in a given environment, E2 came over and asked a single question about each of the activities in that environment in the same order that the activities had been engaged in (each representing one of the three experimental conditions; see below). After the first game and its testing phase were completed, play then moved on to the next two environments in turn.

In the test phase, it was most natural for E2 to be ignorant about either the existence or the location of the target objects. Therefore, during the play phase, she sat in a far corner of the room pretending that she did not notice anything of the games being played by E1 and the child (she was busily engaged in writing things down on the protocol sheet). After each environment, E2 entered the scene and started the test phase, for which she was the main interactant with the child. She immediately explained that she had not been able to watch (because she had to work, take notes, etc.), but that this all looked like fun. She asked the child whether she would like to play the game once more with her. Experimenter 2 then sat down with the child, mentioned the kind of activity necessary for eliciting reference to a particular object (e.g., “It's really dirty here, I bet the clown wants to clean the floor”), and then immediately asked one of test questions, depending on experimental condition (see below). If children named a target object before the actual test question had been asked (which happened infrequently), E2 would go on and ask the question, pretending she had not heard what the child just said (e.g., “Hmm? I did not get what you said, so, what happened to the car?”). Conversely, if a child did not answer a question immediately, E2 repeated it up to two times.

There were three types of test questions, representing the three experimental conditions. In the specific question condition, E2 indicated with her question knowledge about which object was needed, but ignorance about what happened to it (e.g., Was ist mit dem Besen? “What happened to the broom?”). A natural answer to this question involves use of a pronoun (Der/Er ist im Regal. “It is on the shelf,”) or null reference (Da oben, “Up there,”). In the general question condition, E2's question indicated that she knew that something was needed, but she did not know what (Was müssen wir holen? “What do we need to get?”). A likely response to this question involves use of a noun (Den Besen, “the broom”), although a pronoun (das, “this”) together with a pointing gesture is also possible. In contrast, null reference specifying only a location (da oben, “up there”) would mostly not be sufficiently informative here. In the contrast question condition, E2 also knew that something was needed, but she asked for the wrong thing: an object that one might typically expect in the given situation, that is, that is in the same semantic field as the target object (e.g., Hatte der Clown einen Staubsauger? “Did the clown have a vacuum cleaner?”). A natural answer here is to negate the question and mention the correct (target) object with a noun (Nein, einen/den Besen,“No, a/the broom”). Use of a pronoun/demonstrative is not so informative here, but a null reference would be much worse. The conditions and expected responses were thus as follows:

Results

In the main analysis we looked at the nature of the nominal forms children used in the different conditions. Because children sometimes refused to answer or provided an irrelevant response, children had different numbers of scoreable responses; thus, proportions were used; that is, the value for a given child for a given response type (e.g., nouns in the contrast condition) was the number of responses of that type (e.g., nouns) over the total number of responses for that child in that condition (e.g., contrast condition). Children provided scoreable responses in about 80% of all trials (with fewest in the contrast condition, because they sometimes simply answered, incorrectly, “Yes”); if a child gave no scoreable responses in a condition, s/he is not included in the relevant analyses. Table 1 provides the basic results. Nouns with and without determiners were combined into one category, as were pronouns and demonstratives (numbers in the uncombined categories were sometimes very low). “Nonverbal” refers mostly to pointing gestures given with no verbal accompaniment.

For the main statistical analysis we performed a 2 (Age) × 3 (Condition) analysis of variance on the mean proportion of nouns used. Two-year-olds used a higher proportion of nouns overall than did 3-year-olds, F(1, 37) = 5.52, p < .05. There was also a significant effect for condition, F(2, 74) = 77.54, p < .001. Of most interest was the difference among the three conditions. Using Fisher least significant difference (LSD) tests, we found that children used proportionally fewer nouns in the specific condition (.26) than in either the general (.87) or contrast (.94) conditions (p < .001, in both cases). They used marginally fewer nouns in the general condition than in the contrast condition (p < .07). There was no interaction between age and condition.

Inspection of the values for the other reference types is consistent with this analysis (statistical analysis was not useful because of low frequencies). For both age groups children used more pronouns and null references, by several orders of magnitude, in the specific compared with the general and contrast question conditions. Indeed, there is not a single instance at either age of a child using a null reference in the contrast question condition, whereas this is children's most frequent response in the specific question condition, and they only use it infrequently in the general question condition. Using only nonverbal means of reference occurs almost exclusively in the specific question condition.

The specific question condition was thus very different from the other two conditions (less than one-third the number of nouns and three to five times the number of pronouns and null references), whereas the other two differed from one another only somewhat. One explanation for this pattern of results is that certain kinds of questions elicit from children certain kinds of grammatical constructions. For example, when children are asked a specific question such as “What happened to the broom?,” they are drawn to make some reference to the broom as the subject of their response. Because subjects are most often pronouns in the constructions children habitually use, they might simply choose the utterance “It's over there,” “It's on the X,” or something similar. On the other hand, when they are asked a more general question such as “What do we need?,” a natural response, because the target object is not specifically mentioned in the question, is to say something with reference to the target in object position such as “We need a broom” (or even just “A broom.”). This construction also has a pronoun subject, but it repeats the reference to “we” from the adult question, with the needed object specifically named postverbally. Contrast questions such as “Did the clown have a vacuum cleaner?” also elicit responses with a pronoun subject for clown and a noun for the target object postverbally, for instance, “No, he had a broom.” (or just “No, a broom.”).

Figure 1 depicts how the target referent was indicated in the different conditions as a function of the nominal type (noun, pronoun, null) and its syntactic role in the child's utterance (subject, object). The overall picture is very clear. In accordance with the above analysis, in the general and contrast question conditions children used a noun by itself or a postverbal noun almost three-quarters of the time, whereas in the specific question condition they used a noun by itself or a postverbal noun only about 10% of the time. In contrast, in the specific question condition children used a null reference or pronoun in subject position about 65% of the time (noun as subject an additional 13% of the time), whereas in the general and contrast question conditions they used a null reference or pronoun in subject position less than 10% of the time. Thus, it is clear that the specific question pulls for target as subject (about three-quarters of the time), and children are obeying some kind of “light subject constraint” (Chafe, 1994) or “preferred argument structure” in which the argument is expressed with something other than a lexical noun (DuBois, 1987). In the other two conditions, the questions pull for nominal reference, often using lexical nouns, in something other than subject position (about three-quarters of the time).

Syntactic position (S, subject; O, direct object) and referential form (N, noun; Pro, pronoun; Null, null) of children's reference to target objects as a function of experimental condition in Study 1

Table 2 shows the proportions of responses in which children used a nonverbal gesture (with or without language) in each of the experimental conditions: more than three-quarters of their responses overall. Children used gestures most often in the specific question condition, F(2, 74) = 3.45, p < .05, with this condition being higher than each of the other two (p < .05, Fisher LSD tests), which did not differ from one another. The effect was similar for children of both ages (no main effect of age or interaction of age and condition). This finding further supports the idea that the specific question condition is different from the other two.

Finally, a related subject of interest is the question of whether children point equally when they are using different nominal types (see Table 3). The answer is that they do not. As can be seen in Table 3 (all ages and experimental conditions combined, proportion of total responses indicated), when they use a noun they point about three times more often than not. However, when they use a pronoun or null reference they point about six times more often. This result is statistically significant according to a binary logistic regression using repeated measures, e^β = 2.3, p < .01 (Mehta & Patel, 2002). Thus, although they pointed quite a bit overall, children pointed most often when they were not using a noun, perhaps suggesting that they understand that pronouns and null references need pointing more urgently than nouns.

Discussion

The most general finding of the current study is that specific questions (e.g., “What happened to the broom?”) elicit very different referential choices from young children than do more general questions (e.g., “What do we need?”) or contrast questions (e.g., “Do we need a mop?”). Prototypically, the child answers specific questions by placing the entity being asked about as the subject of the sentence in her answer (about three-quarters of the time), and this entity is indicated very infrequently with a noun but much more often with either a null reference or a pronoun (almost three-quarters of the time). In contrast, children tend to answer (again about three-quarters of the time) general questions either with (a) a lexical noun as a single word utterance (Q: “What do we need?” A: “A broom.”), or (b) a sentence with one of the interlocutors (either “we” or “the clown”) as subject and the target lexical noun in a postverbal position (Q: “What do we need?” A: “We need a broom.”). Contrast questions elicited somewhat similar referential choices, although perhaps for different reasons. Thus, when children were asked “Do we need a mop?,” they most often answered (again about three-quarters of the time) with something like “No, a broom,” or “No, we need a broom.”

This pattern of responding makes it quite clear that children at 2.5 and 3.5 years of age are in some ways sensitive to the knowledge states of their interlocutor. In the context of the game being played in the current study, if the adult asked a specific question it meant that she knew the object needed but simply did not know where it was. In her answer, the child thus made only fleeting reference to the object of joint discourse attention (null reference or pronoun, as topic) and then said where it was. On the other hand, in this same context if the adult asked a general question it meant that she did not know what object was needed, and so the children needed to inform her specifically with a lexical noun, which they mostly did. Finally, if the adult asked a contrast question, it meant that she had wrong information about what object was needed and so the child had to correct her by saying “No,” and then informing her of the correct object with a lexical noun. The specific question also elicited more pointing, but this is very likely due to the fact that children used more pronouns and null references in response to the specific question and they seem to know that pronouns and null references “need” indicative gestures more than nouns. That is, our analysis showed that across conditions children used more pointing in combination with pronouns and null references than with lexical nouns, a replication, at a younger age, of the findings of Tomasello, Anselmi, and Farrar (1985).

It is possible that these findings could be interpreted without crediting children with such deep knowledge of other people's knowledge, but it is not clear exactly how that would work. Presumably, the only possibility would be that children have learned some “mindless” discourse rules of the type “when someone asks about an object, begin your sentence with reference to that object” combined with some mindless grammatical rule of the type “begin sentences with pronouns or null references.” However, children make appropriate referential choices in a wide variety of discourse circumstances, for example, continuing a conversation on topic (when the adult is only making statements and not asking questions), using pronouns for topic and lexical nouns for nontopics (preferred argument structure; Clancey, 2002). Thus, a much more plausible view is that young children, at least from 2.5 years of age, know how to make referential choices appropriate for the knowledge states of the listener in many, although certainly not all, communicative contexts. This age range fits perfectly with O'Neill's (1996) findings that children use the pointing gesture differentially depending on the adult's knowledge states, a bit at 2.0 and much more consistently at 2.5. The fact that children cannot talk about knowledge states coherently in false belief and other theory of mind tasks only means that expressing knowledge explicitly in language is something different from expressing it implicitly in one's social and communicative interactions.

Method

Materials, design, and procedure

The experimental setup and procedures were like those of Study 1, with the following main changes. Most importantly, pilot subjects seemed to be distracted by the comings and goings of E2, so in this study the only human interactant was E1, who manipulated the clown (with the clown asking the questions). In addition, to lighten memory demands, each target object had associated with it a specific reference object (e.g., the target object pen had the reference object paper). This meant that E1 and the clown interacted with the child with, for example, the pen and paper and then placed the pen somewhere nearby, as in Study 1. To enact the experimental conditions, E1 and the clown then later found the paper and asked for the pen so they could draw. In addition, to make things easier the target objects were placed in more salient positions (e.g., on top of boxes) rather than always on a shelf. Finally, the test questions were simplified, to fit with 24-month-olds' language skills, as follows:

The wording of the test question was not changed in the specific question condition.

Scoring and reliability were done as in the first study, and Cohen's kappa was determined to be .91.

Results

The 24-month-old children provided scoreable responses on only 45% of the trials (compared with almost 80% for the older children in Study 1). They clearly did not have the same discourse skills as the older children.

Nevertheless, in the main analysis we again compared the proportion of lexical noun use across the three experimental conditions. Results are shown in Table 1. The younger children of this study responded to the altered contrast question as had the older children in Study 1, that is, almost exclusively with lexical nouns. However, in response to the other two kinds of questions, these younger children behaved very differently. Whereas the older children from Study 1 made a sharp distinction between the specific and general questions (providing nouns more than three times as often in response to general questions), the younger children made no such distinction. They responded with lexical nouns approximately half the time in both of these conditions. The statistical finding was thus a main effect for condition, F(2, 25) = 6.90, p < .01, with post hoc analyses (Fisher LSDs) showing that the contrast condition was different from the specific and general conditions (p < .05 in both cases), which did not differ from one another. However, it can also be seen in Table 1 that the children made many null references in the specific question condition, more than twice as many as in the general question condition. This would suggest (statistical analysis was not possible because of low frequencies) that they do differentiate the two kinds of questions, at least to some degree.

With regard to pointing, the younger children of this study again behaved differently from the older children of Study 1 (see Table 2). There was a statistical main effect for experimental condition, F(2, 25) = 6.01, p < .01, with the contrast condition differing from the other two, which did not differ from one another. These younger children only pointed about 14% of the time in the contrast question condition. However, this is very likely due to the nature of the question as modified for this study (viz., “Is this an X?”), which simply pulls for a yes or no answer (if the answer was no, children often pointed to the X on the shelf). In the general and specific question conditions, these younger children pointed 79 and 66% of the time, respectively, values that do not differ from one another reliably.

Discussion

The findings of this study clearly indicate that 2.0-year-old children do not take the knowledge states of their listener into account in the same way as do 2.5- and 3.5-year-old children. Although the contrast question again elicited almost all lexical nouns, it was a different question from the one used with the older children in Study 1, because pilot subjects could not deal with the form “Did the clown have a comb?” (they mostly just said “Yes,” which was factually incorrect). They answered the other two types of questions (specific and general) in very similar ways in terms of the proportion of nouns that were used. This was not true for the older children in Study 1, who differentiated these two sharply by answering general questions mostly with nouns but specific questions mostly with pronouns and null references. Although these younger children did seem to use more null references to specific questions than to general questions, children of this age quite often omit subjects from their utterances (Bloom, 1990). This might thus indicate that they knew that in their answer the mentioned noun should be the subject of the sentence in their response (although they did not make an active choice as to its form). The 24-month-old children used pronouns hardly at all in any of the three conditions. Perhaps this is because the pronoun needed would be the inanimate “it,” and young children only use that as subject in certain constructions, for example, “It's an X” (see Budwig, Stein, & O'Brien, 2001). The children also did not differentiate strongly in their pointing behavior between the general and specific question conditions as did the older children in the first study.

By comparison with the older children of Study 1, therefore, the 24-month-old children in this study seemed much less sensitive to the knowledge states of their interlocutor, at least those relevant for their choice of a referring expression. This conclusion must be taken with caution, however, because these barely 2-year-olds provided scoreable responses to less than half of the questions asked them. In contrast, this result itself could be taken as further evidence of the weaker discourse skills of children at their second birthday.

Method

Results

The results of this study, also analyzed in terms of mean proportions, are very straightforward (see Table 4). Children used lexical nouns at high rates in all conditions, especially the older children. Statistically there was no effect of experimental condition, but there was an effect of age, F(1, 46) = 12.10, p = .001, such that older children produced more lexical nouns than younger children (about 96–76% overall). Children did use their pointing gesture differentially in the three conditions, however, F(2, 92) = 13.19, p < .001. Children of both ages pointed less when the object was not visible than when it was visible, either with or without alternatives present (p < .05 in both cases, Fisher LSD tests). Children pointed equally often in the visible and alternatives conditions.

Discussion

The basic finding of this study was that children did not use different kinds of nominals depending on the perceptual availability of the referent object. There are two important qualifications, however. The first is that the discourse situation in all experimental conditions was one that encouraged the use of lexical nouns. The child was instructed to go ask an adult for an object. Although we thought that children might relatively often request the object by simply pointing to it or saying “That!,” instead what they most often did, the older children on almost every occasion, was simply go over and say something like “We need the car” or “Could I have the car?” or even just “The car.” Presumably, what was going on here was that the child was supposed to ask someone with whom she had established no previous discourse context to retrieve a particular object. This is apparently a very strong discourse context in determining how children choose referring expressions; that is to say, it calls for nouns, regardless of the physical arrangement of the referent object. Children did point more nonverbally to the visible objects, both close to and not close to alternatives, but there are some fairly straightforward explanations for this in terms of perceptual salience; children are simply more inclined to point directly to the object they need than to the location in which it is hidden. The fact that there was no distinction between the visible and alternatives conditions argues for this interpretation not involving an assessment of the listener's knowledge states.

The second point is that we only manipulated perceptual availability in a physical sense: the object was in the open on a shelf, with alternatives on a shelf, or hidden from view. However, it is possible that we may have observed something different if we had manipulated more social aspects of the perceptual context, such as joint attention. Allen and Schroder (2003) have examined naturalistic uses of language by young Inuktitut children, and have found that they use something less specific than lexical nouns (often null references) when they and their interlocutor are both simultaneously looking at the target object. There was basically no joint attention in the current study, at least initially, because the objects were always on the shelf behind the adult when the child asked for them. Thus, it is possible that perceptual context plays an important role in the child's choice of referring expressions, but not perceptual context in a physical sense but rather in a social sense of things that she and her interlocutor are or are not perceiving together.