The Symbolic Species (book review)

The Symbolic Species, The co-evolution of language and the brain: Terrence W. Deacon, Norton, 1997, ISBN-0-393-03838-6, 527pp including text, notes, bibliography, and index.

“Terrence W. Deacon is a world-renowned researcher in neuroscience and evolutionary anthropology. He conducts research in laboratories at Boston University, where he is associate professor of biological anthropology, and McLean Hospital at Harvard Medical School. In addition to his research in brain evolution and development, he has played a significant role in the innovation of neural transplantation techniques for the treatment of human brain disorders. He lives wtih his wife and two children in Concord, Massachusetts.”
– dust jacket bio

Deacon is a heavyweight and it shows.

This is a very meaty and dense book, drawing on everything from semiotics, to philosophy, to modern evolutionary theory, to paleontogy, to neurobiology. The objective is to understand and characterize the evolution of the human brain and language. Some of the main points developed in the book include:

Some animals are capable of learning and using symbols, i.e., references which point to other references. Symbolic learning is hard and unnatural, requiring “social support” and extensive neocortex resources. Humans have evolved brains that are good at symbolic learning. This has been done by expansion and modification of pre-existing brain architecture. Brains adaptively modify their connective architecture; it is misleading to think of them as having hard-wired functions. Chomsky’s thesis of a hard-wired universal grammar module is wrong. Languages and grammar have evolved to match the architecture of human brains, in particular to be easy for children to learn them by “good guessing” of language rules. Extensive use of symbolic thinking probably started with Homo habilis but not in the form of language as we know it. Humans are an evolutionary freak in the sense that special circumstances pushed us into using symbols. Once on that path selection feedback pushed us into being over-determined symbol users. Symbolic communication evolved steadily over the last two million years from being primarily gestural to being principally vocalized with sophisticated vocallization depending on specific adaptations.

I have a few caveats. He dismisses “mind modules” a bit too cavalierly. His refutation of Chomsky’s UGM is convincing; likewise simple notions of hard-wired bits of brain implementing various functions are untenable. However functional areas of ability are effected quasi-independently and would have been selected for. I suspect that the simple iconic/indexical/ symbol differentiation of levels of reference aren’t enough. It is notable that there is almost nothing on “tool-intelligence” in terms of brain architecture and its demands on different parts of the brain. This is a neglected area of brain research, I suspect.

His analysis of the reproductive strategy issues that faced Homo habilis is good as far as it goes. The problems are clear. The notion that it was solved by early pair-bonding is quite dubious. There are more plausible alternatives such as female-dominated groups (e.g., rhesus macaques and bonobo chimps) with “sex-for-meat”. There is some evidence that the Neanderthals had this social structure.

He glosses over the paleontological record of tool usage, being more concerned with avoiding the trap of equating primitive tools with primitive minds. There is a real problem here which he simply ignores. The problem with infant care and transport is treated lightly which is depressingly customary.

Chapter notes

Chapter 1:
The main point raised in chapter 1 is that there are no simple languages (languages with a stripped vocabulary and grammar) outside of the human race. This raises the question of why there is this gap and how it could have been bridged evolutionarily.

Chapters 2,3:
These chapters are on structure of language. Chapter 2 raises basic questions to be asked about the nature of reference. Chapter 3 discusses the nature of symbols with Saussere and Pierce as a starting point. The main focus is on the hierarchic character of references. Three major levels are distinguished, icons (the signifier resembles the signified), indicial (the signifier points to a non-symbolic signified), and symbols (the signifier points to other signifiers).

Iconic and indexed references are unstructured, one-to-one, and can be learned with associative learning subject to extinction. Symbolic references are structured, hierarchic, and not readily learned with simple associative learning.

Only humans use symbolic references. However chimpanzees can be taught to use them. Experiments are cited in which mature chimpanzees learn simple symbolic use (verb/noun structure, simple classes).

The issue is raised: what is the nature of the cognitivie barrier and how did it get breeched.

Chapter 4:
Chapter 4 starts by blowing Chomsky out of the water. It begins with Chomsky’s observation that language is too hard to be achieved with associative learning; the structure is too deep. The critical argument that general learning is sufficient to account for language learning is presented and refuted.

“There is another possibility that has been entirely overlooked and it is the point of the remainder of this chapter to explore it. I think Chomsky and his followers have articulated a central conundrum about language learning, but they offer an answer that inverts cause and effect. They assert that the source of prior support for language acquisition must originate within the brain, on the unstated assumption that there is no other possible source. But there is another alternative: that the extra support for language learning is vested neither in the brain of the child nor in the brains of parents or teachers, but outside brains, in the language itself.”

The idea that language is “user friendly” in that it is tailored to be easy for children to learn is presented.

“Chomsky argues that much of the child’s knowledge of grammar and syntax is not learned in the way that words are. I agree. It is discovered, though not by introspection of rules already available in the brain. On the surface, it simply appears that children have an uncanny ability to make ‘lucky guesses’ about grammar and syntax; they spontaneously anticipate the ways words work together. I think this appearance of lucky coincidence accurately captures what happens, though it is not luck that is responsible. The rules underlying language are acquired by trial-and-error learning; but there is a very high proportion of lucky guesses.”

Languages evolve to make it easy for children to learn them, i.e., so that childrens guesses are likely to be good guesses. Languages evolve very fast compared to biological evolution.

Work with Bonobo chimp named Kanzi is cited. Kanzi understands lexigrams and English, has a mastery of simple grammar. He was not trained directly. His mother was trained in language experiments which she did poorly in. Kanzi learned about language as an infant by being around his mother while she was being taught; he picked it up naturally without being trained in the same way human children learn.

Neural net experiments in learning grammatical structure are cited. Deep learning of grammatical structure fails if learning is too good initially. The problem is the usual one of hanging up on sub-optimal peaks. The answer is to make early learning “noisy” – technique is similar to simulated annealing (not mentioned).

Chapters 5,6:
Chapters 5 and 6 go into considerable detail on the relative size of brains in different species and how and what to measure. Encephalization Quotient (EQ) is misleading if taken as single number. Primates have relatively high EQ achieved by lowering body growth rate relative to brain growth rate; humans get their extra brain size by growing their brains for a longer period of time in addition to low body growth rate. Much stuff about hox genes and homeobox regions. (Decent explanation if you’ve never been through it.) Most of the extra growth in humans is in the neocortex. Good stuff on differential development rates; superficial neotony considered and dismissed in a paragraph.

Chapter 7:
Evolution of brains doesn’t need fortuitous and intricate mutations nor bridge functiosn; functional systemic adaptation is used. As brains mature they adapt to the body. Heavy reliance on overproduction of neurons and connections followed by culling and neuronal death. Most neural connectivity is worked out “ad hoc”.

Brain tissue isn’t all that different from species to species. Experiments cited in which embryonic pig brain tissue is transplanted to rats. It works fine just as though it were rat brain tissue.

Brains are structured by Kaufman-like self organization in response to molecular gradients, proximity, and timing. Brain development doesn’t require pre-established harmony to keep up with body changes.

“This suggests that the difference between human and nonhuman brains may be far more complex and multi-faceted than simply an increase in extra neurons over and above the average primate or mammal trend.”

Chapter 8:
This is a heavy chapter on brains, vocalization, and speech. It starts with a nice bit about Hoover, the talking seal at the Boston aquarium.

Beathing/swallowing involves the mid-brain (lots of physiology here). Vocalization is a complex function of this part of the brain. There is a lot of hard-wired vocalization.

Birds are better than mammals at speech because of flight related modications to the breathing apparatus. Mammals manipulate the larynx, birds the syrinx. Visceral control in mamals is replaced by skilled motor control in birds. Cetaceans use blowholes, still another mechanism, to produce controlled vocalization.

Most mammals run facial control on auto-pilot. Primates have much more voluntary motor control of face. Lots of stuff about humans uniquely vocalizing on exhalation. Infants babble a lot because it is easy for them to do.

Chapter ends with claim language started with Homo habilis. Unsubstatiated at this point except for brain size arguments. Language would have been different – much simpler, heavy on consonants, light on vowels – for physiological reasons.

Chapter 9:
Another heavy chapter on brain function. Lots of stuff on prefrontal function mapping and various effects of prefrontal damage. Some forms impact viewpoint switching.

“Or to pose the question more specifically: Is there something about language that requires a predisposition for working with difficult conditional associative relationships, maintaining items in working memory under highly distractive conditions, or using negative information to shift associative strategies from concrete stimulus drive links to abstract associations?”

Prefrontal cortex does not process symbols. Neurological tokens (sensory image chains) are handled in the posterior temporal cortex. Symbols are distributed, references are patterns of virtual links. The role of the prefrontal cortex is the construction of the distributed symbolic architecture.

Williams Syndrome: overdeveloped symbolic manipulation (expanded prefrontal) with weak indexical associations (posterior temporal damage) – sounds like post-modernist syndrome.

Quick discussion of “theory of mind” – thinks premature. (ToM: ability to picture what someone else is thinking from their viewpoint.)

Chapter 10:
This chapter investigates where language is located in the brain. There is a lot of physiological data, PET scans, et cetera and descriptions of different regions. Language processing is distributed hierarchically outward from motor control areas associated with phoneme recognition.

The fundamental insights are (a) language processing is not functionally distributed either by linguistic function or by neurological function; rather it reflects idiosyncracies of brain architecture, (b) linguistic function is not fixed in location – it arises adaptively, (c) our brains have the same general plan as ape brains and, indeed, vertebrate brains in general, and (d) the function of grammar is to convert the serial structure of language into chunks which can be handled by the brains hierarchic parallel processing. He argues that language analysis is not domain specific whereas indexical functions are. Some symbols are stripped of all but the vaguest symbolic content in order to serve as switches – these are isolated like indexical functions but they are not grammar. Symbols are distributed.

There is some neat stuff about first learning specific associations and then, as indirect associations are built up, unlearning the initial specific associations. Symbolic recoding eliminates the exponential explosion implicit in indexical association.

Interesting: The impact of damage to Broca’s area depends on the native language. Different languages have different processing demands and brains adapt to them. Italian brains are different from English brains.

Poo-poo’s pop theories of brain lateralization. Suggests mostly adaptive. Inportant but seriously misreprented in pop discussion. Cute: interpreters use both halves of the brain, one for each language.

Chapter 11:
Starts off with claim that brains expanded in response to language beginning with Homo habilis. Good discussion of Baldwinian evolution, genetic assimilation. Strong evolutionary arguments that there has been brain evolution to support language but not grammar. Arguments revolve around required invariance of impact on brain. Language adaptations: symbolic cognition – prefrontal expansion, language mimicry. Dubious argument that only language (symbolic cognition) requires such heavy computational support.

There must have been sharp selection for brain size. Nothing anywhere in all of this about brain adaptations for tool use.

Suggestion that early language would have been multimodal – spoken language requires specific adaptations. Extensive discussion of speech adaptations. Suggests early speech was embedded in gestures and exaggerated intonation with heavy use of ritualized markers.

Chapter ends with discussion of extended discussion of incompleteness of paleontological record with respect to artifacts (no preservation of any except stone tools until very late). Point made and remade that “primitive” tools don’t necessarily indicate a primitive mind. There is no recognition here that so called primitive tools, even those of Homo habilis, require considerable intelligence – intelligence of a kind that chimpanzees (and almost certainly A. afarensis) does not have. Chimps cannot learn to make simple tools. It requires more than trial and error learning.

Chapter 12:
This chapter speculates on origins and character of use of symbolism. It points out that reproduction strategies make heavy use of information. Thesis is advanced that change in required strategy force shift in communicative strategy.

Extended discussion of why human reproductive strategies are unstable and unworkable without symbolic communication. Hominids were not polygonous – mild dimorphism establishes that. Teeth suggest principal diet of tubers.

Early hominids shifted to regular inclusion of meat in diet. Consequence is that males are meat gatherers; means was needed for reliable sharing of meat with females with infants. Extended discussion of parenting strategies among different species. Dubious argument for pair-bonding/ quasi-marriage. Arguments for symbolic communication needed for reciprocal altruism – again dubious.

Extended discussion of probable early symbolic communication. Not language as we know it. Heavy on ritual activity which is important in establishing symbolic references. Extensive use of gestures, equivalents of dance.

Chapter 13:
Some animals have modest symbolic learning capability. Humans have evolved a major restructuring of the brain to enchance symbolic learning.

“In evolutionary terms, it would be accurate to say that the genetic basis for symbol-learning abilities has been driven to ‘fixation’. … Language acquistion had to become fail-safe. After 2 million years it has clearly reached this status.

The simplest way to make something fail-safe is to design it far beyond the basic requirements. To ensure that a physical structure is totally safe, for example, it must be designed to handle loads that are far greater than those it would normally be called upon to support… Human brain structure is an exaggerated reflection – a caricature almost – of the special demands imposed by symbol learning. We are not just adapted for symbol learning, but for fail-safe symbol learning.”

Extended discussion of emotional communication, differences between human and animal minds, origin of ethics, and “theory of mind”.

Chapter 14:
The final chapter is a discussion of “bigger” issues. There is a discussion about living in a symbolic universe, the compulsion to use symbolic thinking, a discussion of consciousness, and the some of the issues involved in the possibility of artificial minds.