Discussion

What paradox? A response to Naigles (2002)

As Naigles reports, children's early language looks item-specific in basically all of the world's languages that have been studied (see Tomasello, 2000, for a brief review). For example, children learn certain kinds of syntactic constructions with some verbs, other kinds with other verbs – sometimes with little or no overlap between them. And the learning process looks piecemeal in the sense that the constructions used with some verbs are much more complex and advanced than those used with other verbs. One methodological point that Naigles does not note is that since Tomasello (1992) was a diary study in which a relatively complete record of one child's entire language output was recorded (i.e. all new constructions were recorded), from this study we can make claims with some confidence about what the child did not do. This is important in establishing the item-specificity of early language since these data characterize the structure of each construction not only positively but negatively as well. Nevertheless, data from spontaneous speech are never definitive in identifying underlying cognitive or linguistic representations; experiments are needed.

The largest body of studies are nonce verb production studies in which children are taught a novel verb in one linguistic construction and encouraged to produce it in another. Depending on exactly how one counts, there are over a dozen such studies (with several times that many experimental conditions) done by a handful of different experimenters – although our labs have done most of those with the youngest children. The basic finding is very clear. No matter the kind of construction in which they learn the novel verb, and no matter how they are encouraged to produce utterances themselves with the novel verb, young children – below 3–4 years of age – are very conservative; they are loathe to use the novel verbs in constructions in which they have not heard them used. Even when the most generous scoring criteria are used (a child uses a novel verb creatively one time no matter how many times she uses it in the construction in which she heard it used), less than one-quarter of 2.5 year olds are productive with the novel verbs and it is not until after 3 years of age that the majority of children can produce one syntactically productive utterance with the novel verb (see Tomasello, 2000, for a review).

Naigles lodges two criticisms of these experiments. The first is that the children are not learning the meanings of the verbs the way the experimenters want them to, and since some verbs cannot be used in some constructions children are doing what they should do with these verbs given the meanings they have attached to them. But the first point is that older children do not have this problem at all; they are quite productive with the novel verbs. Why aren't they misled about the verb's meaning in the same way as the younger children? Furthermore, in several studies that Naigles does not cite (e.g. Brooks and Tomasello, 1999, Tomasello and Brooks, 1998) younger children participate in control conditions in which they hear the novel verbs used in the same target constructions that are being pulled for in the experimental conditions (typically, the English transitive construction). And the younger children have no problems in this case. Also, in some of our studies (e.g. Akhtar and Tomasello, 1997, Dodson and Tomasello, 1998, Tomasello et al., 1997 – the latter two not cited by Naigles) we tried to make sure that during training children get a transitive meaning from the syntax of the model without hearing the verb in the transitive construction itself; for example, we say to them “Look at what Ernie's doing to Bert. It's called dacking.” But this still does not help them to use the novel verb creatively in the transitive construction.

The other criticism is that because children are hearing and learning the novel verb multiple times in one construction, the verb's usage in this construction becomes entrenched (primed) and this inhibits them from using it in a novel construction. This is possible, but again it must be noted that if this effect is indeed at work, it does not hold for older children who are productive with the novel verbs. And there are no data to suggest that younger children are more subject to priming effects than older children.

The priming explanation does not work at all for the weird word order studies (Abbot-Smith et al., 2002, Akhtar, 1999). In these studies children hear novel verbs used in weird word orders (e.g. “Ernie the car dacking”) and then later are encouraged to use them with new characters. Older children tend to correct these novel word orders, saying things like “Bert's dacking the tree” whereas younger children are much more likely to continue using the novel word order. On the surface priming could be a problem here also since children hear the weird word order multiple times (though again it doesn't seem to bother the older children), however there is an important control condition that Naigles neglects to mention. Children are also given known English verbs in weird word orders exactly the same number of times (e.g. “Ernie the car pushing”), and even the youngest subjects correct these to SVO order almost all the time. This effectively rules out priming as the sole explanation of the findings in the main experimental conditions.

With the nonce verb comprehension studies, the issue is again verb meaning (and possibly priming) in the sense that in most comprehension experiments children have to first learn the verbs and then act them out in response to requests to do so. Perhaps during learning they do not learn the right meaning. But because we were worried about this ourselves, in Akhtar and Tomasello (1997) we used a new method. We did not really teach the verb at all in any construction. What we did was teach the activity that one does with an apparatus and two characters: on repeated occasions one character acted in a particular way on another on this apparatus, with the characters being manipulated on many occasions by the children themselves. The children thus knew the game. Then, on their first hearing of the novel verb, we pushed the apparatus in front of them in preparation for a new round of the game, handed them two characters and told them “Make Ernie meek Bert”. The only question here was which character was made to be the agent and which the patient; the children knew the game. But again the pattern of results was the same: the older children did this well, whereas the younger children were basically at chance. We believe this study is a particularly strong demonstration since (i) it does not involve children having to produce just-learned words or overcome other performance demands, (ii) there are no priming effects, and (iii) there is no question of problems with intended verb meaning.

Finally, we have one study that is still in press that looks at priming directly (Savage, Lieven, Theakston, & Tomasello, in press). Following Bock and others who have studied priming in adults (e.g. Bock & Griffin, 2000), we looked for evidence of structural priming in young children. The basic idea is that children look through a book containing pairs of pictures. For each pair they first hear (and in some conditions produce themselves by repeating) one described with, for example, a passive sentence (e.g. “The car got smashed by the tree”); then they are asked to describe the second picture of the pair. The second picture contains different characters and different actions, and so the question is whether children will also use a passive sentence for this picture even though there will be very low lexical overlap with what they just heard for the first member of the pair (if they are going to construct a passive sentence it will have to be something like “The balloon got popped by the pin”). The answer is, once again, older children are subject to such structural priming effects, but younger children – below 4 years of age – are not. However, and importantly, in other experimental conditions the younger children were subject to priming effects when the sentence they heard/used for the first member of the pair contained pronouns and other closed class morphemes that they could repeat in their own sentence for the second picture (e.g. in both cases “He got VERBed by it”). This is actually relatively direct evidence that children's underling representations contain specific lexical content, in this case pronouns as arguments and some closed class morphemes constitutive of particular syntactic constructions.

Although there is only this one study using this methodology with young children, we believe it is especially telling because in this study children do not have to learn anything new; they simply use English verbs they already know in constructions they already know with those verbs. And priming is of course a widely accepted method for establishing the nature of underlying cognitive representations in Cognitive Science in general.