Volume 24, Issue 4

In 2005 (vol. 22, Issue 2–4) The Linguistic Review explored the topic of linguistics' role in present day cognitive science. Prominent members in the various fields of cognitive science, including philosophers, psychologists, ethologists and linguists, were brought together to give their views on this topic. As there needed to be some centralizing point, the journal chose to use Ray Jackendoff's work The Foundations of Language (2002) as a jumping off point for all the contributors, a work that explicitly aims at linking linguistic theories of the mind to researchers who approach this vast topic from other angles.

The special issue of The Linguistic Review on “The Role of Linguistics in Cognitive Science” presents a variety of viewpoints that complement or contrast with the perspective offered in Foundations of Language (Jackendoff 2002a). The present article is a response to the special issue. It discusses what it would mean to integrate linguistics into cognitive science, then shows how the parallel architecture proposed in Foundations seeks to accomplish this goal by altering certain fundamental assumptions of generative grammar. It defends this approach against criticisms both from mainstream generative grammar and from a variety of broader attacks on the generative enterprise, and it reflects on the nature of Universal Grammar. It then shows how the parallel architecture applies directly to processing and defends this construal against various critiques. Finally, it contrasts views in the special issue with that of Foundations with respect to what is unique about language among cognitive capacities, and it conjectures about the course of the evolution of the language faculty.

For over ten years now Jackendoff has argued that linguistic theory as carried out by “mainstream generative grammarians”, particularly under the rubric of minimalism, is misguided. The arguments have remained remarkably constant, and have been repeated on numerous occasions. The present article indicates why we remain unconvinced by the strength of his arguments.

Usage-based accounts of language-learning ought to predict that, in the absence of linguistic input, children will not communicate in language-like ways. But this prediction is not borne out by the data. Deaf children whose hearing losses prevent them from acquiring the spoken language that surrounds them, and whose hearing parents have not exposed them to a conventional sign language, invent gesture systems, called homesigns, that display many of the properties found in natural language. Children thus have biases to structure their communication in language-like ways, biases that reflect their cognitive skills. But why do the deaf children recruit this particular set of cognitive skills, and not others, to their homesign systems? In other words, what determines the biases children bring to language-learning? The answer is clearly not linguistic input.

In this note, I clarify the point of my paper “The Nature of Semantics: On Jackendoff's Arguments” (NS) in light of Ray Jackendoff's comments in his “Linguistics in Cognitive Science: The State of the Art.” Along the way, I amplify my remarks on unification.

The hermetic, exclusionary stance of formal linguistic research towards relevant data condemns the linguistic enterprise to irrelevance. The procedures and principles of evolutionary biology are excluded from discussions of the evolution of language. Stories are instead crafted to fit speculation. Although human speech is a central element of our linguistic capacity, the findings of studies of speech are thought to be outside the domain of linguistic research. Well attested facts such as the formant frequency relationships that characterize vowels are ignored. Phonological studies continue to use binary coded tongue positions that neither describe the relationships that hold between vowels, or predict the direction of sound changes. Linguists generally continue to hold to the premise that a domain specific “faculty” of the brain exists, contrary to the findings of neurophysiologic studies that demonstrate linkages between the neural bases of motor control, cognition and language. A “Universal Grammar” continues to be a core property of linguistic theory, despite evidence to the contrary from genetic, behavioral and these neurophysiologic studies.

Jackendoff and other linguists have acknowledged that there is gradience in language but have tended to treat gradient phenomena as separate from the core of language, which is viewed as fully productive and compositional. This perspective suffuses Jackendoff's (2007) response to our position paper (Bybee and McClelland 2005). We argue that gradience is an inherent feature of language representation, processing, and learning, and that natural language exhibits all degrees of gradience. Contrary to Jackendoff's assertions, we do not reject the possibility of innate constraints on language, feeling only that the jury is out on the nature and specificity of such constraints. We address a number of questions Jackendoff raises about the process of grammaticalization, drawing on extant literature of which he appears to be unaware. We also address Jackendoff's views on the prospect that connectionist models can address core aspects of language processing and representation. Here again extant literature of which Jackendoff seems unaware addresses all four of his general objections to connectionist approaches.

The debate about the place of linguistic theory in cognitive science encouraged by The Linguistic Review is a good example of communication between different research communities. In this follow-up paper we (1) clarify our theoretical and methodological positions, (2) propose a theoretical model for language production similar to Jackendoff's Parallel Architecture , and (2) discuss emerging empirical evidence for this model. Our data suggest that perceptual, semantic, and syntactic information becomes available to the speaker in parallel providing competing production cues. Main architectural parameters of the proposed model are similar to Parallel Architecture, but we put a greater focus on the interface between language-specific and general cognitive domains. We view such interface as a regular mapping mechanism between the grammatical constraints imposed by the language system and the perceptual, semantic, and grammatical priming parameters available in the communicative environment.

Progress can be made in the exploration of the evolutionary origins of language by recognizing that the language capacity in humans is comprised of many and varied interlocking pieces of a biologically based mosaic. To understand how language can be both species-specific and evolved from capacities in ancestral species it is necessary to recognize that distinct aspects of linguistic cognition probably followed distinct evolutionary paths. This paper presents reasons to support Jackendoff's notion of linguistic layers and reiterates the value of a theory of conceptual structure that involves tiers, especially distinguishing the thematic tier, involving Paths and their component parts, derived from spatial structure, and the action tier, involving Actors, derived from social cognition.

Editor's note: TLR wishes to increase the number of book reviews (either short reviews or review articles). Reviews offered to TLR could be on books listed below, or any other linguistic book. Please contact the Review Editor for details.