HOW TO BLOCK A FASTER PROCESS?

In the first part of the target article CF report data from Clahsen, Hadler, and Weyerts (2004) that show a frequency effect for irregular past tense production and a reversed frequency effect for regular past tense production; that is, high-frequency regular past tense forms were produced slower than low-frequency forms. CF interpret the frequency effect as evidence for the dual mechanism model of inflectional morphology. For the reversed frequency effect in regular production CF adopt an explanation from Pinker (1999). This explanation assumes that high-frequency regulars are lexically stored, low-frequency regulars are not. Pinker suggests that regular production is delayed for high-frequency verbs because it is blocked by their lexical entry. Because there is no lexical entry for low-frequency regular verbs, production of their regular past tense form is not blocked and thus faster.

This explanation demonstrates a weakness of the dual mechanism hypothesis as a processing model. The assumed blocking mechanism clearly lacks serious consideration of the timing of the processes involved. If unblocked regular past tense production is faster than the retrieval of lexically stored past tense forms (which both CF and Pinker seem to assume), then it is hard to see how the retrieval of a stored past tense form can ever block regular production. Retrieval should always be too late, and irregulars should always be regularized. On the other hand, if retrieval were faster than regular production the blocking mechanism could work. However, in that case stored high-frequency regular past tense forms should be produced faster than low-frequency forms synthesized by a rule mechanism. Hence, there could be no reversed frequency effect. A simple way out of this paradox might be a modified dual mechanism model in which it is not the retrieval of the stored past tense forms as such that blocks regular production but some “pointer” to a stored past tense form that is already available as soon as the verb stem is retrieved.

Note, furthermore, that according to Pinker (1999, p. 303), who refers to Beck (1997), the suggested explanation only holds when there are so many irregulars in the stimulus set that participants adopt a special strategy: they always search for a lexical past tense entry. Surprisingly, this means that looking up a past tense entry is not the normal procedure, which again raises the question how the blocking mechanism actually works outside of experimental stimulus sets.

Clearly, what is needed to achieve some progress in our understanding of morphological production is a model that is explicit in terms of the timing of lexical retrieval and regular production and can explain (a) the main effect of regular past tense production being on the whole faster than irregular past tense production that is found in most studies (Beck, 1997; Jaeger et al., 1996; Prasada, Pinker, & Snyder, 1990; Sach, Seitz, & Indefrey, 2004), (b) how overregularization is avoided, and (c) the conditions under which (reversed) frequency effects occur. For the time being I do not see that any of the suggested models satisfactorily deals with all three points.

ARE LOW-SPAN NATIVE SPEAKERS STRIKINGLY DIFFERENT FROM HIGH-SPAN NATIVE SPEAKERS?

In the second part of their target article CF turn to syntactic processing. They provide evidence suggesting that during sentence parsing L2 learners unlike native speakers do not build up structural representations but semantic (“shallow”) representations. In an experiment by Marinis, Roberts, Felser, and Clahsen (2005), for example, native speakers but not advanced L2 speakers of English showed sensitivity to the intermediate gap in sentences like The nurse who the doctor argued ________ that the rude patient had angered ________ is refusing to work late. Based on this finding, CF conclude a “striking” difference between L2 and native speakers. Again, if the main interest is to provide evidence for a dichotomous view of language processing, this result may be considered satisfactory. If we want to understand what listeners in general do when processing sentences it is not. Note that we learn from another experiment (Roberts, Marinis, Felser, & Clahsen, 2004) that low memory-span native speakers did not show sensitivity to the gap in sentences like John saw the peacock to which the small penguin gave the nice birthday present ________ in the garden last weekend. It seems that an unbiased interpretation of the two studies would have to conclude that L2 speakers behave at least like some native speakers. Thus, if we were to classify types of listeners it would be high working-memory span native speakers on one side and low-span native speakers together with L2 speakers on the other. Obviously, what the latter groups have in common is not their language background but much more likely some limitation of their processing resources. CF attempt to argue against such an explanation for L2 speakers by pointing out (a) that reading time differences do not seem to affect sentence processing, and (b) that L1 learners show “overreliance on structural information due to their relatively limited working-memory resources.” Both arguments are hardly convincing. If L2 processing as such imposes a high load on processing resources, then a resulting preference for semantic rather than structural sentence processing should already be found in even the fastest L2 reader, and in this respect no differences to slower readers are to be expected. The second argument transfers findings from children's inflectional behavior to the domain of sentence processing. However, CF themselves suggest that the processing in these two domains may differ when they conclude that advanced L2 speakers seem to employ the same mechanisms as L1 speakers in inflectional processing but not in sentence parsing. The second argument, furthermore, ignores the much more relevant findings of the Roberts et al. (2004) study on sentence processing showing that the memory-span effect in L1 learners was just like in adult speakers, that is, low-span children did not rely on structural information.

In sum, the studies reviewed by CF do not provide evidence that L2 sentence processing is, in principle, different from L1 sentence processing. Nonstructural sentence processing observed in L2 speakers is an option that is also used by native speakers when they have limited processing resources, be it a low working-memory span or more severe limitations in aphasia (Hagoort, Wassenaar, & Brown, 2003). Furthermore, to the extent that structural sentence processing relies on Broca's area, neuroimaging data suggest that the degree of structural processing varies as a function of task demands and complexity of the sentence material (Dapretto & Bookheimer, 1999; Friederici, 2002; Indefrey, Hellwig, Herzog, Seitz, & Hagoort, 2004). To put it simply, native speakers seem to employ their structural sentence processing resources to the degree that is necessary to perform a task at hand or to understand a particular type of sentence. Quite possibly this means that structural processing is not simply on or off, but there are varying types of “shallow” representations with different amounts of syntactic information being represented. What we need to know is for which types of sentences speakers with a given amount of processing resources process which types of morphosyntactic or structural information.