In this article, the authors lay out an impressive body of evidence that supports two main claims. First, they favor the continuity hypothesis, according to which children's parsing mechanisms are essentially the same as adults'. Parsing strategies change little over time, and those changes that occur are attributed to differences in lexical processing efficiency and working memory capacity. Second, they suggest that there are substantial differences in the parsing strategies adopted by native speakers and adult learners of second languages.

Although I do not have extensive experience investigating child language processing, the one study I have conducted on children (8–12 years old; Traxler, 2002) suggested that they, like adults, did not make use of subcategory preferences to make initial parsing decisions in subject–object ambiguities (as in When Mary tripped the table fell over/the policeman stopped to help her up). Whether the subordinate clause verb preferred a direct object or was incompatible with a direct object, children's self-paced reading times showed evidence of processing difficulty at and following the point of syntactic disambiguation (the main-clause verb, in this case). However, the verbs' subcategorization preferences predicted overall difficulty resolving the ambiguity for both children and adults (Traxler, in press). Both of these findings are compatible with the authors' proposal that the first language (L1) parsing mechanism is stable over time.

One claim that may require further justification is that differences in language processing performance are based on the way working memory differences affect parsing per se, rather than other aspects of processing and interpretation (Caplan, Hildebrandt, & Waters, 1994; Caplan & Waters, 1990, 1995, 1999; Traxler, Morris, & Seely, 2002; Traxler, Williams, Blozis, & Morris; 2005; Waters & Caplan, 1992, 1996a, 1996b; Waters, Caplan, & Rochon, 1995). There are a number of reasons to be cautious about the relationship between working memory and parsing performance. First, studies of stroke and Alzheimer disease patients show that people with very low scores (1 or 0) on the sentence span test are able to parse and interpret sentences containing long-distance dependencies, which require readers to hold information in a working memory buffer while inputting and structuring subsequent text (Rochon, Waters, & Caplan, 1994; Waters & Caplan, 1997). Second, the working memory task that has been most widely used in sentence processing research (sentence span; Daneman & Carpenter, 1980) has fairly low test–retest reliability, and instability in classifying participants' working memory capacity makes it more difficult to replicate findings (Waters & Caplan, 2003). Third, many of the studies in which a connection between working memory capacity and some aspect of comprehension performance have been found (including the original Wanner & Maratsos, 1978, study) have relied on dual- or triple-task paradigms, where the ability to maintain task instructions or switch attention between the primary and secondary task are at least as important in determining performance as the actual language processing task. Fourth, studies where a single-task paradigm (e.g., eye-movement monitoring) is used often employ quasiexperimental designs in which a conceptually continuous variable (working memory capacity) is treated as a categorical variable, or use extreme-group designs (e.g., King & Just, 1991; MacDonald, Just, & Carpenter, 1992), which can artificially exaggerate possible between-group differences. Fifth, associations between working memory capacity and processing time are sometimes found in sentences that can be interpreted incrementally, and that should not impose substantial working memory demands (e.g., Pearlmutter & MacDonald, 1995), and are sometimes not found in studies involving sentences, such as object relatives, that are thought to impose substantial working memory load (Traxler et al., 2005). Sixth, working memory capacity covaries with a number of other potentially important variables such as lexical decoding skill, print exposure, and ability to shift attention (Engle, 2002; La Pointe & Engle, 1990). Thus, working memory capacity may be serving as a proxy for unassessed variables in studies where a significant association is found between working memory and comprehension performance. Unless these other potentially important variables are measured, and unless their contribution to performance is evaluated simultaneously with working memory capacity, it is not entirely safe to conclude that working memory differences are the source of observed differences in performance.

There is also a great deal of work that needs to be done to elevate minimalist parsing strategies, such as good enough heuristics (Ferreira, Bailey, & Ferraro, 2002) or Townsend and Bever's (1999) pseudosyntax approach, to the status of fully articulated accounts of grammatical representation and processing. It is not clear, for example, when processing difficulty or failure to assign a correct standard interpretation reflect the workings of a system that does not take the time to generate fully specified syntactic representations for complex constructions because it is relying on sketchy or partial syntactic trees, or a system that is merely subject to random error. It is clear, however, that comprehenders can fall back on more rudimentary interpretive strategies, such as assigning prototypical theta roles to entities named in a complex sentence, rather than making syntactically driven assignments. What we need are better accounts of precisely when minimalist strategies will be applied (i.e., whether they apply widely across a broad range of sentence types, or whether they serve as a backup system when the syntax is too complex or when the standard interpretation is deficient) and whether individual differences in lexical knowledge and cognitive resources (such as working memory, sensitivity to semantic cues, tendency to perseverate, or the ability to manage attention flexibly) affect the degree to which people employ minimalist parsing strategies.