I was recently asked for my thoughts about this book, and I wrote down some. I think the book is a very important one, and other reviews have raised many relevant issues to think about. But after 6 years, the book is worth revisiting, particularly in the context of what we no know about social learning in primates.
Deacon's position is that the evolution of human minds is mainly about the evolution of language. So for him, explaining the evolution of language (and the brain features that support it) explains much of interest about humans.
From a biological perspective, there are two basic options to explain the evolution of language. Neither has a clear consensus behind it, because we basically have no empirical data that addresses the question. One approach is that of Chomsky, whose position is that the brain has a strongly innate ability to learn language, so much so that the grammars of natural languages are confined to a small range of possibilities. But also intrinsic to Chomsky is the idea that the neural underpinnings of language were not themselves selected for their function in language but instead for some other function. In other words, Chomsky has argued that an intermediate form of language--not fully encompassing the grammatical structures of today's langauges--is not possible. Although this has been criticized as anti-evolutionary, in fact it is supported by some prominent evolutionists, such as Stephen Jay Gould, who views it as likely that other brain functions requiring symbolic logic were the targets of selection, and that language later arose as an artifact of culture.
The second approach is that of Steve Pinker, who basically takes Chomsky at face value--namely, that there is an innate brain capacity for learning natural languages--and claims that language function itself was the target of selection. This explanation has the inconvenient consequence of making it necessary to explain what an intermediate form of language may have been like. It also makes it possible that today's people still vary in their biological capabilities with respect to language, and that selection may still be happening. Large parts of Pinker's books (The Language Instinct, The Blank Slate) are dedicated to explaining why those inferences are unlikely, in his view.
Pinker invokes as his major evolutionary mechanism the Baldwin effect, which is the idea that learning can make evolutionary change more likely by enabling greater flexibility of behavioral response and thereby reducing the costs of genetic variability within populations. In this view, when behaviors like symbol use or language fall within the range of some individuals in the population, the rest of the population may well be able to learn them. As the population changes behaviorally to learn these skills, natural selection can begin to act on the genetic variation that may be related to them, either because the genes underly the behaviors themselves or the ability to learn the behaviors. In Pinker's view, this is how the ability to learn language, evolves, encompassing the creation of a set of innate brain functions that he calls the "Language Acquisition Device" (LAD). The innate character of the LAD is the reason why there are developmental windows for language learning, and why humans tend to learn language-typical sounds, words, and grammatical features according to a stereotypical series of steps.
There are some elements of broad agreement between these two different viewpoints that are important to point out. First, they agree that much of interest about the structure of language is essentially determined by innate features of the brain. They only differ on the source of selection for those innate features. Second, they agree that the essential feature of human language is its grammatical organization. The agreement on grammar as the essential feature follows from a number of empirical observations, including the fact that children learn words well before they learn to form sentences or grammatical phrases, that chimpanzees (and other animals) are capable of understanding words and learning to use symbols in simple combinations, and that humans readily learn pidgins as adults that lack the complex features of natural language grammars.
Deacon differs from this assumption by claiming that symbols are the essential element of human language, far more important in his view than grammar. He argues strongly (and relatively convincingly) that "symbol" use in animals is actually an instance of indexical learning, since those who teach chimpanzees language make sure that the words for things (or signs or computer icons) reliably co-occur along with those things in the chimpanzees' environments. But of course if symbols are learned as indexes by chimpanzees, they are presumably learned that way by prelinguistic children as well, so Deacon needs a way to define symbol that emphasizes how symbols actually differ from indexes. His answer is that symbols are logically connected to other symbols in an interlocking set of relationships. It is this relationship set that characterizes human symbol use, rather than the mere presence of arbitrary signs (Peirce's definition of symbols). In other words, Deacon believes the evolution of language is not to be explained in terms of innate grammatical functions, but instead in terms of the acquisition and manipulation of symbols and symbol-relations. (Deacon does not as far as I can tell use a single term for this idea, but he does talk about the "systemic logic of symbolic reference," I think symbol-relations is a good way to shorten this idea).
One may ask why these symbol-relations are not the same as grammar. Deacon's view appears to be that the symbol-relations do underlie grammar (or more directly, syntactical relations) as human languages use them. His main disagreement is with the notion of Universal Grammar. Chomsky observed that natural languages in humans appear to "choose" their grammatical rules from a finite set following a flowchart-like series of options. Universal Grammar is his reification of these options, and the basic hypothesis of generative linguists has been that the capacity to use Universal Grammar to structure linguistic communication is present in the minds of language speakers. An important finding about Universal Grammar was a proof by E. Mark Gold (1967) that the rules of a language's grammar could not be learned from a finite series of utterances by any purely inductive process. This supported the notion that some knowledge of grammar must be innate--that language learners must be pre-equipped with biases that encouraged certain kinds of assumptions about syntactic relations, or they would never be able to figure out the grammatical rules of their language.
But Deacon argues that Universal Grammar is unnecessary. In his view, innate assumptions are not the only way to create learning biases that enable the acquisition of grammar rules. Biases in learning might instead stem from the constraints that young children typically face in interpreting speech. In his view, children ignore many of the details of syntactic relations in their initial attempts to interpret speech. Using a top-down approach, they focus on those elements that are readily understood and later fill in the details. Deacon references experiments in language interpretation by neural networks as well as the transition from pidgins to creoles in mixed-language cultures to support this viewpoint.
And there are strong evolutionary reasons to doubt the existence of a Universal Grammar. In short, no single language uses all (or even most) of the rules included in Universal Grammar. So the phenotype (behavioral expression) of nearly all the people in any human population must not include many of those rules. If this is true there is certainly no way that natural selection (which can operate only on phenotypes) could result in Universal Grammar being included in most people's brains.
Like Pinker, Deacon also applies the Baldwin effect as his principal evolutionary mechanism (He makes this very clear on page 328, where he begins "Both Pinker and I argue that...). But instead of grammatical knowledge, he argues that the Baldwin effect applies only to those aspects of language that "impose consistent invariant demands on neural processes" (329). In Deacon's view, the essential process encouraged by Baldwinian evolution in ancient hominids was the extension of joint attention and the breaking of mnemonic-indexical connections that are involved in symbol learning.
Deacon places these functions correctly in the prefrontal cortex of the brain, but incorrectly argues that this part of the brain underwent relative expansion during human evolution; in fact chimpanzees probably are the same relative size in terms of prefrontal cortex area as humans, and several other parts of the brain may also be involved. What is essential in terms of human evolution is the overall expansion of the neocortex, and much less so the relative sizes of different parts, although the changes in relative extent in the parietal association areas and some specifically language-related features such as Broca's area may be even more important.
Deacon's viewpoint is very much in line with recent research on learning in primates, particularly as represented by Michael Tomasello. There is an increasing recognition that the social learning processes that underlie the kind of cultural behaviors seen in primates are derived from extended periods of juvenile learning involving more extensive interaction with other individuals, particular mother-offspring interactions. Many of these interactions involve joint attention on objects or other individuals, and the use of such joint attention to enable social learning appears to characterize marmosets, chimpanzees, and other primates in which social learning is important.
On the other hand, there is no strong evidence that this is the primary mode of evolutionary change leading to human symbolic communication or symbolic culture. Deacon has told a story that makes sense, but there is no strong empirical evidence that supports this view as opposed to other possible ideas about language evolution. I think he is closer to the truth than Pinker, but there are several missing elements as well.
UPDATE (2008-07-01): I have e-mail from Terrence Deacon regarding one of the points above: prefrontal cortex area in humans compared to chimpanzees. To paraphrase, he notes that the crucial test on the relative size of prefrontal cortex in humans has not been done, as relative measures have been of frontal cortex, not prefrontal. He writes:
The fact of the matter is that the data which could settle this question are still too skimpy to provide a definitive answer. This must be approached by laborious histological methods. One of my grad students has actually chosen this project for his PhD. So soon we will have a real test. I am betting that my original analysis, based on work by some of the most illustrious neuroanatomists of the last century, will hold up. But before this you can read those papers for yourself and make up you own mind about what has and has not been shown.