Clahsen and Felser's article (CF) is an important contribution to the field of psycholinguistics in several respects. First, it draws attention to the importance of a better understanding of the processing mechanisms utilized by child and adult language learners. Differences in these mechanisms may be responsible for the final outcome of the acquisition process. Second, the article provides an excellent summary of current first language (L1) and second language (L2) research on processing. A variety of studies, ranging from morphological off-line investigations to on-line research on syntactic development, are reviewed in a concise and accurate manner; the cross-linguistic dimension of the article makes both the review and the argument even more comprehensive and convincing. Third, based on their own experimental results, and the results from other studies, the authors propose a novel account of L2 processing, the so-called shallow syntax hypothesis (SSH).
In this short commentary, I would like to focus on a particular line of research that is relevant specifically for L1 development and that does not receive close attention in CF. I take as a starting point the authors' claim that “…the results on children's sentence processing indicate that the child parser is essentially the same as the adult one. Although children's sentence processing tends to be slower overall than adults', there is no reason to believe that their processing system is fundamentally different from adult native speakers' processing system…” Although I agree that the child system is not qualitatively different from that of adults, the question is what happens when interpretation of a particular structure can be obtained, in principle, by two different systems, for example, on the basis of structural (“narrow syntax”) or discourse information. Consider Examples 1 and 2.
The difference between these two types of wh- questions is well known. Example 2 represents the so-called D-linked wh- question, shown to exhibit different properties from non-D-linked wh- questions such as Example 1 (Cinque, 1990; Pesetsky, 1987; Rizzi, 1990). Moreover, there is a difference in processing of these questions: De Vincenzi (1991) and Shapiro (2000) demonstrated experimentally with native adult speakers that establishing a dependency between the trace, and the wh- phrase takes more processing resources for D-linked than for non-D-linked wh- phrases. Avrutin (2000) also showed that 4- to 6-year-old children are, overall, at chance in interpreting sentences of type Example 2. When presented with a picture where a tiger is chasing a lion while being chased, at the same time, by another lion, children chose the correct lion only 48% of the time, pointing to the lion chasing the tiger 52% of the time. No such problem was observed in sentences with non-D-linked wh- questions as in Example 1 (results for the same children were above chance: 86% of the time correct). In line with CF's claim, that “…children have more difficulty than adults accessing different knowledge sources and evaluating different types of information in parallel…” Avrutin (2000) suggests that coordination of syntactic and discourse information overloads child processing capacity, which results in a chance performance.1
Interestingly, Hickok and Avrutin (1995) report similar findings for agrammatic Broca's aphasics. The authors suggested a knowledge-based account arguing for the selective deficit of binding versus government chains, an account that I no longer consider tenable.
In fact, recent research on dependencies in natural languages allows us to make more precise claims about the source of errors. Basing his research mostly on pronominal anaphora, Reuland (2001) developed a theory where dependencies can be established at different levels (e.g., lexical, syntactic, semantic, discourse), and that these levels form a hierarchy in such a way that the possibility of a “cheaper” dependency blocks the application of a more “expensive” one. Whenever a lexical dependency is, in principle, an option, it will prevent establishing a syntactic one; the possibility to establish dependency relation in “narrow syntax” will block semantic and discourse dependencies, and so forth. Considering wh- traces to be pronominal elements, and extending Reuland's approach to wh- questions, we can conclude that there is an important difference in how a trace can be interpreted in Examples 1 and 2. Non-D-linked wh- phrases can be linked to their traces only via syntactic dependencies, while D-linked ones can, in principle, enter a discourse dependency. For adult speakers the syntactic dependency is the chosen one in both cases: in Example 1 because there is no other option available, and in Example 2 because it occupies a more economical place in Reuland's hierarchy. Suppose, however, that the syntactic dependency is not the cheapest option available to children, at least at a younger age. In this case, the interpretation of D-linked wh- questions in Example 2 may involve a competition of two equally possible sources of interpretation: narrow syntax and discourse, the former involving establishing a syntactic chain, and the latter involving linking the trace (a pronominal element) to the referent introduced into discourse as part of the D-linked wh- phrase. As a result of such competition, children will demonstrate an overall chance performance, as reported in Avrutin (2000).
Similar claims have been made with regard to children's interpretation of pronouns in exceptional case marking (ECM) clauses. In a number of cross-linguistic experiments, Ruigendijk et al. (2006; see more in Ruigendijk, Avrutin, & Vasic, 2005) present evidence that Dutch-, Russian-, and Spanish-speaking children often interpret the pronoun in the ECM clause as coreferent with the matrix subject (the translation of the relevant experimental condition is given in Example 3).
First the man and the boy ate, and then the boy saw him skating
Children interpreted the pronoun equivalent to “him” as coreferent with “the boy” approximately 50% of the time, significantly more often than the equivalent pronoun in a simple transitive clause, or a reflexive pronoun (equivalent to English “himself”) in the ECM sentence. The authors provide analyses consistent with Reuland's economy hierarchy: in adult language the pronoun cannot be dependent on the matrix subject because semantic- and discourse-level (extrasyntactic) dependencies are blocked due to the potential availability of the syntactic dependency, which in adult hierarchy occupies a more privileged, more economical position. The syntactic dependency, although structurally possible, is disallowed due to the feature specification of the pronoun. The authors argue, however, that the syntactic dependency is not the cheapest one for children at the relevant age (4 years, 0 months [4;0]–6;0), and that their errors are due to the competition between the (unacceptable) syntax-based interpretation and the (acceptable) semantic one.
The weak, unreliable nature of the “narrow syntax” in children is argued to be the source of the optional omission of such functional categories as articles and tense.2
In case of production, of course, children are significantly younger, approximately between 2;0 and 2;8.
It is important that, and in full agreement with CF's claim, the nature of the child processing system does not seem to differ from adults'; however, the ability to fully rely on “narrow syntax” as a source of interpretation may develop in time. Narrow syntax becomes fully operational only when it becomes fully efficient, which is related, among other things, to the maturation of the child's processing capacity. In this sense, it would be interesting to see whether L2 adult learners, whose biological development is not different from adult native speakers, will demonstrate different performance in a wh- question experiment (reported by CF) when non-D-linked wh- questions are replaced with D-linked ones.
To summarize, as important as it is to distinguish between the learner's knowledge of the target language, and his/her ability to use this knowledge, it is also important to realize that the use of a specific processing mechanism requires the availability of sufficient resources. When resources are not sufficient, for example, due to the incomplete maturation of relevant brain structures, the output of the processing system may be different from that of a normal adult native speaker because of reliance on other, competing sources of interpretation or encoding information. This appears to be the case for structures that rely crucially on the use of narrow syntax. By itself, this is not really surprising; after all, syntax is a very useful thing to have, but only when it is fully usable.
