Clahsen and Felser (CF) analyze the performance of monolingual children and adult second language (L2) learners in off-line and on-line tasks and compare their performance with that of adult monolinguals. They conclude that child first language (L1) processing is basically the same as adult L1 processing (the contiguity assumption), with differences in performance being due to cognitive developmental limitations. They argue that differences in L2 performance, however, are more qualitative and not explained by shortage of working memory (WM) resources, differences in processing speed, transfer of L1 processing routines, or incomplete acquisition of the target grammar. They propose a shallow structure hypothesis (SSH) to explain the differences reported in sentence processing. According to this, the syntactic representations computed by L2 learners during comprehension are shallower and less detailed than those computed by native speakers and involve more direct form-function mappings.

The main question they are addressing, whether L2 processing is fundamentally different from L1 processing, has been repeatedly asked from different perspectives (e.g., Dehaene et al., 1997; Hahne, 2001; Hahne & Friederici, 2001; Hasegawa, Carpenter, & Just, 2002; Kim, Relkin, Lee, & Hirsch, 1997; Perani et al., 1998; Sebastian-Gallés, Echeverría, & Bosh, 2005; Weber-Fox & Neville, 1996), and several hypotheses have been formulated to account for the differences (e.g., Ullman, 2001). What is new in this interesting contribution is the attempt to clarify specific aspects of L1 and L2 morphological and syntactic processing and to account for them. Their proposal of the SSH is intriguing, timely, and should lead to further research. However, it needs a further degree of specification and clearly more empirical data.

Although the comparison of child L1 processing with that of L2 adults is interesting and illustrative, perhaps it should not be pushed too far, because there are a number of important differences other than language learning to consider. In that sense, the comparison of early and late bilinguals of the same age, matched in proficiency, would be a good approach to consider. Early bilinguals are not actually discussed in the paper, despite the relevance of this group to completing the picture the authors are trying to build of L1/L2 processing differences. Do proficient early bilinguals use “superficial” or “deep” syntactic strategies for both their languages, or structure-driven strategies for the (temporarily?) dominant language~and lexical–semantic strategies for the other? What would the SSH predict in the case of these “balanced” bilinguals? To answer that question it is maybe necessary to know why the sentence parsing options available in the L2 are “restricted” to shallow strategies and under what circumstances. What would prevent adult L2 learners from eventually achieving nativelike parsing strategies? Why are they able to use every other type of information, including morphosyntactic information, and yet not syntactic structural cues? If the authors believe that this is due to developmental factors (critical periods/age of acquisition effects), can they account for why there is a critical period for (some aspects of) sentence parsing and not for morphosyntactic processing, without being as “vague” as the Ullman model?

Another concern is that, because the results from the different studies on ambiguity resolution in relative clause attachment (in relation to L1 transfer, etc.) are, as they admit, inconclusive, the authors are really basing their conclusions about the “striking” differences in L2 processing strategies, on very little data. This is not to say, of course, that their hypothesis is incorrect. As the authors state, this kind of shallow processing strategy or “good enough” representation is probably part of the L1 repertoire, “an option available to the human language comprehension system in principle” (see Sanford & Sturt, 2002).

Thus, it seems intuitively logical that for L2 adults, even after long periods of exposure, if a “shallow” strategy is available, less costly but effective, then this is what will be used, given the higher cognitive demands of processing in the L2. It could, then, be an early interlanguage feature of L2 sentence processing that continues to be effective, and so employed, even at advanced learner stages. It could even be that reliance on this type of processing routine interferes with or impedes acquisition of more structure-driven parsing strategies. However, more data is needed from L2 learners from different backgrounds (e.g., immersion vs. academic learning) and different levels of proficiency, as well as different comprehension modalities (auditory as well as visual) before it can be claimed that adult L2 learners are “restricted” to this type of sentence-processing strategy. In addition, are the authors suggesting that all syntactic representations computed are less detailed? Perhaps there are characteristics of the particular structures studied that lend themselves to this type of strategy. Again, more data from other experiments with different structures is necessary before claims can be made about “restrictions.” Because only a very small number of structures in the three domains (morphosyntax, ambiguity resolution, and syntactic dependencies) were tested, it remains to be seen in which domains L2 performance is qualitatively worse than native performance and how general is the decrease in performance across the domain.

It is interesting that the authors appear to be taking it for granted that their highly proficient L2 adult participants have reached a fixed, immutable stage of L2 competence beyond which they cannot progress and that whatever processing strategies the participants were using in these studies will remain unchanged, presumably because, as seen from the low number of errors, these strategies are effective for successful L2 comprehension. However, this need not necessarily be the case. Even when L2 late learners demonstrate high levels of L2 knowledge and competence in some measures (particularly off-line measures) there could still be room for “improvement” or change in terms of greater processing automaticity, or perhaps even changes in the parsing options, as a result of longer exposure to the L2, or more intensive experience of the particular language structures in question. One example of this is that some of our recent data seem to suggest that very proficient adult late bilinguals who have spent around 20 years immersed in the L2 environment, process agreement of certain morphosyntactic features (e.g., number agreement) in a similar way to natives (Gillon-Dowens, Barber-Friend, Vergara, & Carreiras, 2004).

As to the question of the “confounding factors” that might account for the differences in adult L2 syntactic processing, again more research is needed before these factors of WM limitations, differences of processing speed and “incomplete” acquisition (whether due to L1 transfer or lack of L2 automaticity) can be dismissed as not significantly contributing to L2 processing differences. Although each of these factors might not be individually responsible for these differences, the interaction and combined effects of these limitations will surely account for at least part of the differences found. For example, looking at the question of the higher cognitive demands and the limitations of WM resources in the L2, if the SSH is correct, and describes the kind of processing strategy adult L2 learners use, WM limitations may be one of the reasons adult L2 learners fall back on this type of “good enough” or shallow representation. Obviously, WM resources will be more taxed in the L2, especially in the case of these long-distance syntactic dependencies. According to the authors, however, if WM limitations were a main problem then “we would expect L2 learners to prioritize on grammatical information in the same way as children do.” This does not necessarily have to be the case, however, as L2 acquisition does not have to mirror child L1 acquisition. Adults, among other differences, already have well-developed, complex lexical–semantic networks before learning the L2 so there is no reason why their processing strategies should initially be of the “structure-based, least-effort” type, as is the case of children, as reported here. Adult L2 learners, through increased exposure to the L2, may develop more nativelike, structure-based parsing heuristics, as WM resources are released by faster lexical access and greater automaticity at a morphosyntactic processing level.

A final point is that the authors argue against the Ullman proposal because of its vagueness. We may agree with that, but the proposal they put forward—shallower L2 sentence processing (but not morpohological processing)—is not very well specified either. Specifically, which components are processed in this way? Does the processor carry out a general shallow processing and so some components are not processed properly? Do some language components reach automaticity easier and are therefore less prone to be processed in a shallow way? It may be the case that there is still not enough empirical evidence to present a more detailed proposal, but presumably the authors have predictions about future studies.

In sum, the SSH is an interesting proposal that hopefully will be refined by future data from the research in this field that it will undoubtedly promote.