Clahsen and Felser (CF) present a thought-provoking article that is likely to have a strong impact on the field, in particular, on developmental psycholinguistics and second language (L2) acquisition research. Unlike the majority of previous work on language acquisition that focused on “competence,” that is, the knowledge basis underlying grammar, CF emphasize the need to approach language acquisition with psycholinguistic measures of processing. Based primarily on behavioral and electrophysiological on-line data, they argue that language acquisition in early first language (L1) and late L2 follows different patterns.

Some of the main issues CF address are as follows:

1. Whereas children use basically the same “syntax first” mechanisms in processing grammar as adult native speakers (“continuity hypothesis”), late (adult) language learners “underuse” syntactic information and instead rely on lexical–semantic and pragmatic information (“shallow syntax” hypothesis).
2. However, in contrast to what others apparently claim (e.g., Ullman, 2001), the rule-based computational system (procedural system) is assumed to be available to late L2 learners at least to some extent, in addition to the mental lexicon (stored knowledge, declarative system).
3. CF distinguish between morphological processing, which may be similar in early and late learners and involve the rule system, and syntactic processing, which is subject to the above mentioned differences.
4. The authors do not see evidence for transfer between L1 and L2, so it should not matter which L1 is available during the initial state of late L2 learning.
5. Future research should focus particularly on the largely neglected processing mechanisms of language learners at early stages.

I am in agreement with many of these points, although some appear incomplete and may require modifications. In some cases, even the evidence referred to in their article does not necessarily support their claims.

First, I find that Lück, Hahne, Friederici, and Clahsen's (2001) plural production data showing much worse performance accuracy for regular (51.8%) than irregular (98%) nouns is counterintuitive if one assumes that the children have acquired at least some rule knowledge, as suggested by the frontal negativities in event-related potentials (ERPs; found in 8- to 12-year-olds), that are taken to reflect “rule-based morphological processing” (see CF's Table 2). The authors tentatively attribute the poor performance to the potentially lesser degree of familiarity with loan words. However, CF's Table 1 suggests that familiarity (measured as frequency) of regulars should not have an impact at all. In fact, the rule-computed plural -s should be the default by definition, even for entirely novel words, and should therefore result in better than chance level performance in any case. Moreover, this finding is in apparent contrast to the high overregularization rates reported for children's production of past participles.

Second, one of CF's claims is that L1 should not have an impact on the late acquisition of L2. However, apart from considerable evidence from behavioral data, some recent ERP data seem to support L1–L2 transfer as well. (a) The cited ERP study on grammatical gender processing in Dutch by Sabourin (2003) reported that only German learners displayed a “nativelike” ERP response to gender violations, whereas Romance and English learners whose L1 gender system is more different from Dutch did not. In addition to this supportive transfer, negative transfer was found as well in cases where German and Dutch differ. (b) A recent ERP study by Tokowicz and MacWhinney (2005) demonstrates further support for negative transfer: English learners of Spanish displayed P600 effects for violations of features that either similarly exist in their L1 (subject–verb number agreement) or are not represented in L1 at all (gender agreement), but not for features that follow different distributions in L1 and L2 (determiner number agreement). (c) Even one of the authors' own ERP studies seems to support transfer effects. Hahne, Müller, and Clahsen (2003) found that adult Russian learners of German displayed a nativelike LAN-P600 pattern for overregularizations of German past participles (which are similar to Russian participles), but not of German noun plurals (which strongly differ from Russian plurals).

Third, a main point, particularly in apparent contrast to Ullman's declarative/procedural (DP) model, is that CF posit at least partial accessibility of the (procedural) rule system in late L2 learners; however, only for “local” morphology and not (or much less so) for phrase structural rules in syntax. I believe it is premature to draw such a clear distinction. Morphology can be viewed as local, perhaps, even as intralexical, only in the domain of morpho phonology, on which most of the authors' work has focused. In this domain, rule application is, in fact, limited to the local composition or decomposition of polymorphemic words. According to CF, this is the limited level at which rule-based processing in late L2 learners may still be functional, whereas the “ability to compute grammatical representations at the sentence level is reduced.” However, the regular/irregular distinction in morphophonology is just one aspect of inflectional morphology, whose main function (i.e., the morphosyntactic marking of tense, number, case, or grammatical gender, etc.) crucially depends on structural relations among sentence constituents. Thus, in the case of morpho syntactic agreement violations (e.g., subject–verb agreement), inflectional morphology cannot be processed locally at the word level but crucially involves phrase structural representation. As the studies cited above demonstrate (Sabourin, 2003; Tokowicz & MacWhinney, 2005), this is also a domain in which nativelike processing in late L2 learners may be possible and in which transfer from L1 may occur. Similarly, the domain of syntax also comprises a number of principles and parameters that may cause distinct patterns of difficulty in language learners. In contrast to the behavioral data presented by CF, there is at least some evidence that late L2 learners can achieve nativelike competence and performance levels (e.g., White & Genesee, 1996, for subjacency and the empty category principle in late French-speaking learners of English).¹

Intriguingly, one of the ERP studies CF referred to found nativelike ERP patterns including the early anterior negativity in proficient late L2 learners of an artificial language when they were presented with phrase structural violations (Friederici, Steinhauer, & Pfeifer, 2001). Thus, proficiency appears to influence the neurocognition of late-acquired L2 beyond morphophonology and may indeed lead to nativelike processing even in subdomains of syntax. Unlike for morphophonology, these proficiency-dependent changes may occur only late, and at very high levels of proficiency. They may be observable only in nativelike end states of exceptionally successful L2 learners. That is, even the “high end of the proficiency scale” on which most previous L2 studies have focused according to CF may not usually be sufficient to capture these changes, unless a subject screening for nativelike proficiency has been performed (White & Genesee, 1996). Even after several decades of exposure to the target L2, the actual level of ultimate attainment conceivably depends on many factors. Among these, the age of first exposure, but also the structure of L1, the relative use of L2 compared to L1 (see also Birdsong, this issue), and a variety of sociocultural factors may play a role. CF's article presents a valuable account for the vast majority of averagely successful late L2 learners near or at end state. I completely agree with the authors that one area of future research should focus on the largely neglected earlier stages of language acquisition. In addition, however, it might be worthwhile to also look at the top end of nativelike attainment in late L2 learners. Only data from this population will shed light on the critical period hypothesis, that is, on the question of whether maturational constraints ultimately prevent late language learners from involving the same neurocognitive mechanisms as native speakers. Friederici et al.'s (2002) data suggest that changes in syntactic processing during L2 acquisition may turn out to be as dynamic as those observed for morphological processing. The combined employment of behavioral and physiological on-line measures is likely to answer some of these questions.