Jackendoff's position with respect to categories (for lexical items and larger constituents) is unclear. Positing categories is (1) implausible in several respects; (2) it makes the binding problem in language seem more massive than it actually is; and (3) it makes it difficult to explain language acquisition. Waiting for connectionism to fulfill its promise, a different track is sketched which is residually symbolic, exemplarist, and analogy-based.

The model of parallel architecture for language presented by Jackendoff is a kind of stratificational model in the spirit of Sydney Lamb. It differs from the more usual stratificationalism most importantly in its clear commitment to nativism, though the variety of nativism is greatly modified from what is more usual among Chomskyans. The revised model presents a potential for fruitful discussion with proponents of stratificationalism, and the potential for enrichment via a relational implementation.

Jackendoff's Foundations of Language: Brain, Meaning, Grammar, Evolution (2002) achieves a major shift in the focus and methods of Generative Linguistics (GL). Yet some of the original restrictive features of GL, cognitivism and Cartesianism in particular, remain intact in the new work and take on a more extreme form with the addition of a phenomenalist ontology.

The nature of the relationship between language and thought has been quite elusive. We believe that its understanding is crucially dependent on the available notions of language and thought. Foundations of Language offers an unusually clear and complete account of both, providing a fruitful and much needed framework for future research. No doubt it will help us think better about these elusive complexities.

Jackendoff's account of relating linguistic structure and brain structure is too restricted in concentrating on formal features of computational requirements, neglecting the achievements of various types of neuroscientific modelling. My own approaches to neuronal models of syntactic organization show how these requirements could be met. The book's lack of discussion of a sound philosophy of the relation is briefly mentioned.

Jackendoff's Foundations of Language: Brain, Meaning, Grammar, Evolution attempts to reconnect generative linguistics to the rest of cognitive science. However, by minimally acknowledging decades of work in cognitive linguistics, treating dynamical systems approaches somewhat dismissively, and clinging to certain fundamental dogma while revising others, he clearly risks satisfying no one by almost pleasing everyone.

In designing stratified models of human language, understanding notions of logical consequence and validity of inference require separating the aspects of meaning that vary between models from logical constants. Modelling meaning requires choices regarding the primitives, where the Fregean program is still offering us the fundamental insights on the role of truth, judgement, and grasping or sharing of thoughts.

Despite its computational elegancy, Jackendoff's proposal to reconcile competing approaches by postulating a parallel architecture for phonological, syntactic, and semantic modules is disappointing. We argue that it is a pragmatic version of the leading module which Jackendoff would probably prefer, but which he does not explicitly acknowledge. This internal conflict leads to several shortcomings and even distortions of information presented in the book.

This commentary discusses the division of labor between syntax and phonology, starting with the parallel model of grammar developed by Jackendoff. It is proposed that linear, left-to-right order of linguistic items is not represented in syntax, but in phonology. Syntax concerns the abstract relations of categories alone. All components of grammar contribute to linear order, by means of the interface rules.

Jackendoff's scenario of the evolution of language is a major contribution towards a more rigorous theory of the origins of language, because it is theoretically constrained by a testable theory of modern language. However, the theoretical constraints from evolutionary theory are not really recognized in his work. We hope that Jackendoff's lead will be followed by intensive cooperation between linguistic theorists and evolutionary modellers.

The commentaries show the wide variety of incommensurable viewpoints on language that Foundations of Language attempts to integrate. In order to achieve a more comprehensive framework that preserves genuine insights coming from all sides, everyone will have to give a little.

High temporal resolution event-related brain potential and electroencephalographic coherence studies of the neural substrate of short-term storage in working memory indicate that the sustained coactivation of both prefrontal cortex and the posterior cortical systems that participate in the initial perception and comprehension of the retained information are involved in its storage. These studies further show that short-term storage mechanisms involve an increase in neural synchrony between prefrontal cortex and posterior cortex and the enhanced activation of long-term memory representations of material held in short-term memory. This activation begins during the encoding/comprehension phase and evidently is prolonged into the retention phase by attentional drive from prefrontal cortex control systems. A parsimonious interpretation of these findings is that the long-term memory systems associated with the posterior cortical processors provide the necessary representational basis for working memory, with the property of short-term memory decay being primarily due to the posterior system. In this view, there is no reason to posit specialized neural systems whose functions are limited to those of short-term storage buffers. Prefrontal cortex provides the attentional pointer system for maintaining activation in the appropriate posterior processing systems. Short-term memory capacity and phenomena such as displacement of information in short-term memory are determined by limitations on the number of pointers that can be sustained by the prefrontal control systems.

One of the fundamental questions raised by Ruchkin, Grafman, Cameron, and Berndt's (Ruchkin et al.'s) interpretation of no distinct specialized neural networks for short-term storage buffers and long-term memory systems, is that of the link between perception and memory processes. In this framework, we take the opportunity in this commentary to discuss a specific working memory task involving percept formation, temporary retention, auditory imagery, and the attention-based maintenance of information, that is, the verbal transformation effect.

Ruchkin et al.'s theoretical conclusions reflect two venerable fallacies. They confound an experimental paradigm with a theoretical concept, and they assume that features of the paradigm that are most readily detected by their methods provide an adequate account of the operation of the theoretical system. This results in a simplistic theory that does not do justice to the richness of the available data.

In attempting to integrate the authors' proposed model with results from analogous human event-related potential (ERP) research, we found difficulties with: (1) its apparent disregard for supraordinate representations at posterior multimodal association cortices, (2) its failure to address contextual task effects, and (3) its strict architectural dichotomy between memory storage and control functions.

The present commentary agrees with many of the points made by Ruchkin et al., but brings up several important differences in assumptions. These assumptions have to do with the nature of the capacity limit in working memory and the possible bases of working-memory activation.

Ruchkin et al. use brain-activity data from healthy subjects to assess the physiological validity of a cognitive working memory model and to propose modifications. The conclusions drawn from this data are interesting and plausible, but they have limitations. Much of what is known about the neural mechanisms of working memory comes from single neuron recordings in animals, and it is currently not fully understood how these translate to scalp recordings of EEG.

In this commentary, the formation of “pre-iconic” visual-prime persistence is described in the context of prime-specific, independent-component activation at prefrontal and posterior EEG-recording sites. Although this activity subserves neural systems that are near identical to those described by Ruchkin and colleagues, we consider priming to be a dynamic process, identified with patterns of coherence and temporal structure of very high precision.

With reference to Ruchkins et al.'s framework, this commentary briefly considers the history of working memory, and whether, heuristically, this is a useful concept. A neuropsychologically motivated critique is offered, specifically with regard to the recent trend for working-memory researchers to conceptualise this capacity more as a process than as a set of distinct task-specific stores.

The notion that the working-memory system is not to be located in the prefrontal cortex, but rather constituted by the interplay between temporal and frontal areas, is of some attraction. However, at least for the domain of sentence comprehension, this perspective is promoted on the basis of sparse data. For this domain, the authors not only missed out on the chance to systematically integrate event-related brain potential (ERP) and neuroimaging data when interpreting their own findings on semantic aspects of working memory, but also neglected syntactic aspects of working memory and computation altogether.