The role of metalinguistic ability in L2 development is seriously underestimated. It may be seen both (1) as a means of initiating or boosting the flow of primary linguistic data and (2) as a generator of substitute knowledge (derived, but epistemologically distinct from domain-specific knowledge) that may compete with or compensate for perceived gaps in the learners current underlying competence. It cannot serve as a simple means of distinguishing the rival theoretical positions.

Epstein et al.'s “full access to Universal Grammar” position is conceptually and empirically problematic. Its shortcomings are illustrated through a brief discussion of the following issues: (1) initial versus final states of grammatical knowledge in a second language, (2) knowledge of gradience of grainmaticality, (3) optionality and retention in non-native grammars, and (4) the empirical measurement of syntactic knowledge.

The most compelling evidence for Epstein et al.'s central thesis that adult second language acquisition is constrained by the innate cognitive structures that constrain native language acquisition would be evidence of poverty of the stimulus. Although there are studies that point to such evidence, Epstein et al.'s primary form of argumentation, targetlike performance by second-language acquiring adults, is much less convincing.

This commentary addresses two pervasive misconceptions which emerge in Epstein et al.'s target article: (1) that study of second language acquisition (SLA) began in the mid-twentieth century; (2) that SLA has only recently become able to contribute to linguistic theory. There is abundant historical counterevidence; I argue that (1) and (2) obscure the legitimacy of Epstein et al.'s “full access” hypothesis.

This target article is mostly a presentation of experimental research devoted to the larger issue of the role of Universal Grammar in second language learning. Deliberately excluding the aspects of human cognition that makes second language (L2) so variant, Epstein et al. focus on what the learners may know and how they come to know it. This is the aspect of Epstein et al.'s work which is more limiting, and potentially more interesting.

Our results support the idea that adults have access to the principles and parameters of Universal Grammar (UG), contrary to Epstein et al.'s misrepresentation of our work as involving partial access to UG. For both LI and L2 acquisition, functional projections appear to develop in a gradual fashion, but in L2 acquisition there is partial transfer in that the lowest projection (VP) is transferred from the speaker's LI.

Epstein, Flynn, and Martohardjono's presentation of the principal approaches to UG access in L2 acquisition is misleading; they have neglected the possibility that the L1 grammar forms the learner's initial representation of the L2, with subsequent modifications constrained by UG. Furthermore, their experimental data are open to several interpretations and are consistent with a number of different positions in the field.

The target article advanced the null, unified and widely misinterpreted generative hypothesis regarding second language (L2) acquisition. Postulating that UG (Universal Grammar) constrains L2 knowledge growth does not entail identical developmental trajectories for L2 and first language (LI) acquisition; nor does it preclude a role for the L1. In embracing this hypothesis, we maintain a distinction between competence and performance. Those who conflate the two repeat fundamental and by no means unprecedented misconstruals of the generative enterprise, and more specifically, of the empirical content of the null hypothesis regarding L2 linguistic knowledge growth. We hope to have identified certain common goals, the adoption of which might constitute a firm foundation for continued productive interdisciplinary development of contemporary theoretical and experimental L2 acquisition research.

Jeannerod (1994) advocates studying motor images to understand the representation of action. We identify an unsettled issue that complicates the use of motor images to study the representation of action and present some evidence for a clear absence of equivalence between motor imagery and movement preparation. We then elaborate and emphasize the relevance of the mental practice literature, which was introduced in the first round of commentary. Finally, we suggest two methods that will allow motor imagery theorists to address these points.

By way of commentary on Jeannerod (1994), we propose that (1) intention, response imagery, and actual response processes carry equal weight in inferring from one process to another, (2) memory networks control intention, which interacts with imagery-based processing to control response imagery, and (3) response imagery will demonstrate learning effects better when imagery reconstruction and elaboration are emphasized and neutral retention tests are utilized.

Jeannerod's target article and a number of commentators stressed the necessity of distinguishing between motor representation and perceptual images. A patient with a lesion of the central somatosensory pathways allowed us to test for this distinction. The outcome suggests that motor images may not be visual and that the “pragmatic” framework proposed by Jeannerod for vision can apply to other sensory modalities.

Two issues have been raised. The first concerns whether consciousness is attached to a given type of action. It is argued that purposive actions are represented before being executed and that motor representations can either remain implicit or become explicit according to the task. The second issue concerns whether or not mental imagery of action is independent from action. Recent evidence showing the commonality of neural mechanisms for motor imagery and action, respectively, goes against the idea of independent processes.

We raise three issues concerning the Eichenbaum, Otto & Cohen (1994) model. (1) We argue against the strict division of labor that Eichenbaum et al. attribute to neocortical and limbic regions. (2) We raise the possibility that the anterior and posterior portions of the hippocampus may be important for different types of information processing. (3) We argue that, rather than reflecting relational processing, different neural responses to “match” and “nonmatch” trials may relate to different required spatial responses.

Eichenbaum et al.'s (1994a) theory suffers from a lack of ecological validation. It is not at all clear why the hypothesized faculties would have evolved and what their adaptive value would be. I argue that hippocampal function can only be understood if the animal is seen in its natural context.

Why is consciousness associated with recovery of memories that are initially dependent on the hippocampal system? A hypothesis is proposed that the medial temporal lobe/hippocampal complex (MTL/H) receives as its input only information that is consciously apprehended. By a process termed “cohesion,” the MTL/H binds into a memory trace those neural elements that mediated the conscious experience so that effectively, “consciousness” is an integral part of the memory trace. It is the phenomenological records of events (Conway 1992), integrated consciousness-content packets, that are recovered when memory traces are retrieved.

Eichenbaum et al. (1994a) hypothesized that perceptually distinct items and the relations among them are processed sequentially by the parahippocampal region and the hippocampal formation, respectively. Predictions based solely on their model's sequential-processing feature might prove easier to disconfirm than those based on its representational features. Two such predictions are discussed: (1) double dissociations should be impossible following hippocampal vs. parahippocampal lesions, and (2) hippocampal lesions should not exacerbate impairments that follow complete parahippocampal lesions.

Various apparently incompatible theories of hippocampal function have been proposed but integration is now needed. It is argued that the involvement of the hippocampus is most clearly seen when the animal needs to extrapolate beyond current sensory information. Such control can involve both the initiation of behaviour in the absence of appropriate sensory input and the inhibition of behaviour that might otherwise be triggered by current sensory input.

Continuing commentary raised several issues concerning our proposal that the hippocampus, parahippocampal region, and cortical association areas mediate different aspects of memory function. Recent relevant findings strengthen our argument that neocortical areas and the parahippocampal region maintain persistent encodings of specific single items and that the hippocampus mediates representations of the relations among these items. The reciprocally and closely interconnected structures that compose the hippocampal memory system work interactively to support flexible memory expression that is relevant to the natural behavior of animals and to conscious recollection in humans.

If selection at the group level is to be considered more than a mere possibility, it is important to find phenomena that are best explained at this level of selection. I argue that human religious phenomena provide evidence for the selection of a “pious gene” at the group level, which results in a human tendency to believe in a transcendental reality that encourages behavioral conformity to collective as opposed to individual interest.

Group selection may be defined either broadly or narrowly. Narrowly defined group selection may involve either selection for altruism or group selection between alternative evolutionarily stable states. The last variety of group selection is likely to have been particularly important in human evolution.