Clahsen and Felser (CF) review ground-breaking work comparing selected types of language processing in monolingual children and adults, on the one hand, and in monolingual first language (L1) adults and adult second language (L2) learners, on the other. They argue that children behave essentially like adults, but that adult L2 learners, even high-proficiency ones, do not. Thus, there is a principled difference to be made among types of learners; there is continuity of mechanism and process to be observed in monolingual development but L2 acquisition exhibits certain fundamental differences. In particular, L2 learners construct shallow syntactic structures (essentially failing to compute trace chains) when processing long-distance filler-gap dependencies. According to the shallow structure hypothesis (SSH), learners immediately interpret incoming words in a minimal semantic representation by assigning thematic roles to argument expressions and associating modifiers to their hosts. They are not mapping detailed and complete syntactic representations onto semantic representations.
Although I view these studies as an essential adjunct to standard L2 acquisition (SLA) research and applaud CF for undertaking in a systematic way the study of sentence processing, I see a fundamental methodological weakness in that the populations from which samples have been selected have not been carefully defined. Psycholinguists need to provide basic comparative data on bilinguals and monolinguals before the consequences of their results to L2 learning can be properly assessed. It is commonplace in SLA research, it is true, to compare learner behavior to that of monolinguals, but this practice is questionable in language processing research where the effects of bilingualism itself on parsing must be researched. We must not assume, prior to investigation, that the population of high-proficiency steady-state bilinguals can be equated with that of monolinguals simply because particular competence tests reveal them to be indistinguishable (Grosjean, 1989). Having two languages in one's head will mean, at least in certain circumstances, that both languages are activated and accessed during language processing because there is no other way to explain unique properties of bilingual language behavior such as nonce borrowings and code switching, or the ability to identify the source language of a given string (Grosjean, 1997). At the same time, we must not assume, prior to investigation, that the population of steady-state “fossilized” L2 users (to use a handy term of Vivian Cook's), whatever their linguistic competence might be, can be equated with that of balanced bilinguals, defined here as those whose acquisition is simultaneous rather than sequential, who have reached levels of proficiency and language use that make them indistinguishable from monolinguals (except for the kinds of bilingual behavior just mentioned) and who use each of their languages on a daily basis (cf. Meisel, 2001, pp. 12–14). Such individuals will serve to provide necessary comparison points for the larger group of other sorts of L2 user.1
These individuals are those whose input is severely constrained because they have acquired their L2 in the classroom (meaning that they may have little understanding of how to interpret certain linguistic registers and genre, cannot draw appropriate inferences from indirect speech acts, and otherwise lack relevant sociocultural knowledge necessary to linguistic performance); those whose performance is lopsided in that they read extensively but have little exposure to spoken language and hardly ever speak the L2 (meaning that they have never developed appropriate processing procedures specific to these modalities); those who use the L2 on a daily basis but in highly stereotyped and ritualized circumstances such as service encounters in banks, restaurants, and shops (meaning that they have little exposure to complex language and infrequent vocabulary); and so forth.
At the very least, the problem of establishing criteria for identifying and labeling language learners (those whose systems are unstable and changing), steady-state L2 users and bilinguals must be addressed. There is no consensus currently even on appropriate terminology and one man's “learner” is another man's “bilingual,” something that makes interpreting results of individual studies difficult. Once a clear picture is obtained of the consequences of bilingualism for language processing of the phenomena discussed here, we will be better able to interpret the significance of the results obtained with groups of high- or low-proficiency learners.
The authors apparently see a direct link between the results of their research and SLA because they attempt to motivate their studies by mentioning an apparent acquisition paradox, namely, that comprehending sentences of a language presupposes prior knowledge of the grammar of that language, whereas building a grammar presupposes the existence of mechanisms for processing linguistic input. The paradox is indeed only apparent because language acquisition is, even in monolinguals, a long-term affair. The hypothesis that lexical and grammatical acquisition is incremental is well documented. It is no news that learners first build mental representations of some things, representations that then permit them to process or parse additional representations of the same things. At some subsequent point in time, in ways that are entirely unclear, the established grammatical and conceptual representations, which are also subserving existing processing procedures, permit learners to build novel representations of different things, which then, in turn, permit them to parse those things. Consider in this regard the well-known facts about the early interpretation of passives, or bound anaphora in English-speaking children. The challenge for theories of acquisition is, precisely, to discover and describe what comes first and what comes after, and to try to make precise how, in the absence of explicit representations of some bit of knowledge, a learner can come to represent it. How parsing and learning interact in specific cases is thus anything but clear (see Carroll, 2001, for discussion) and this paper does nothing to elucidate the matter, whatever its merits in shedding light on parsing.
If the SSH is borne out in further studies, much will depend on giving it substance. CF readily acknowledge that others have proposed comparable ideas, and there is considerable variation among the proposals identified as to mechanics and functional architecture. It is striking, therefore, that one of the most worked out versions of this claim—the late assignment of syntax theory (LAST) of Townsend and Bever (2001)—is not cited or discussed. The LAST proposal is especially interesting for the current discussion because it proposes that all syntactic processing proceeds along two lines in parallel: a shallow analysis and a deeper analysis by synthesis of a complete and fully specified syntactic representation. The shallow analysis consists of an analysis of canonical patterns and semantic roles, including moving wh- expressions back to their source position, the resulting “pseudosyntactic” representation being immediately linked to semantic representations. The pseudosyntax is associatively acquired based on relevant input and is sensitive to the input's frequency. The deeper analysis by synthesis takes place independently of the semantics up to the end of the clause, computing a derivation on a cyclical basis. At that point, a comparison of the synthesized syntactic representation and the input string occurs. When there is a match, the computed meaning is stored and the syntactic details can fade from working memory.
Townsend and Bever (2001) suggest that shallow processing might be impaired in aphasias. The CF studies suggest that their subjects may not be accessing the full information in rule-based derivations. The time course of the two types of analysis is different, and this is where an investigation of bilinguals versus L2 users, and L2 users versus L2 learners might prove to be especially enlightening. For if it is generally acknowledged that L2 learners process their L2 more slowly, the exact cause of this delay is not known. It may be that learners are slower precisely because they, as learners, are constructing processing procedures in real time. Conversely, it may be that L2 users process the L2 more slowly because the L1 processing procedures or lexicon are activated and compete with L2 processing procedures, competition serving in all L2 users to slow down processing. Conversely, it might be the case that certain L1-based procedures “win-out” in an initial competition, because they are more automated and harder to suppress, but an L2-based reparsing has to occur if the syntax and the semantics are to match. This is what we might suspect occurs with L2 users with highly differentiated capacities (highly proficient readers but not highly proficient listeners). Comparing well-defined groups should help us sort out these possibilities. Moreover, they should tell us if a slowed capacity creates qualitative differences in processing because an alternative representation from which a semantic representation can be built “wins out.”
The SSH is an interesting one and should lead not only to much-needed replications of the same phenomena with different groups whose learning profiles and linguistic knowledge are thoroughly investigated and properly documented, but also to studies of novel phenomena. Because if L2 learners are restricted to shallow structure processing, there should be profound differences in their abilities to comprehend those subtle semantic distinctions (such as scope ambiguities) that hinge on the fine detail of hierarchical representations. Research studies on the syntax–semantics interface in SLA have been rather limited to date in number, in the phenomena examined, and in the language pairs involved. The Essex research will hopefully inspire more interest in this area while providing an explicit processing framework within which to couch results.
Such individuals will serve to provide necessary comparison points for the larger group of other sorts of L2 user.
