Human uniqueness, learned symbols and recursive thought.

Recent Developments in the Evolution of Language

James R Hurford,
Language Evolution and Computation Research Unit,
School of Philosophy, Psychology and Language Sciences,
University of Edinburgh.

Note: In Cognitive Systems, 2006, 7(1): 23-32. Where there is any difference, the printed version is the authoritative version.


The last quarter of the 20th century saw a surge in research in the evolution of language, and this activity continues to grow and extend its influence in the present century. This article is a personal review of some conclusions that can be deemed to have been established in that period. Many of these modern conclusions had ancient precursors as speculative hypotheses with little empirical backing. Modern empirical research in a range of fields has driven foundations deeper, and careful theoretical work has begun to weave a more consistent network of ideas across disciplines. Many mysteries remain, but some clear outlines of the evolutionary bases of humans' most distinctive capacity have begun to emerge. Often the clearer outlines have revealed more complex problems than was vaguely suspected earlier. Three propositions have been selected here, and each will be briefly discussed in a separate section. The three propositions are:

1. `Language' is not a Single Monolithic Behaviour

`Language' needs to be broken down into many components, for each of which a different evolutionary story is appropriate. Laypeople, and even many academic nonlinguists, tend to think of language as a single unified faculty. Expressions such as `the evolution of language', `the language instinct' and `language is what distinguishes humans from non-humans' reinforce this view. Within language evolution studies, works such as Lieberman's (1984) The Biology and Evolution of Language and Dunbar's (1998) Grooming, Gossip and the Evolution of Language, despite their many virtues, have fallen into a the trap of suggesting a single source for a complex faculty. Neither of these books, for example, gives any account of the syntactic complexity of languages. To a gullible reader, they can be taken as claiming that once some essential component was in place, be it the lowering of the larynx or the realization that symbolic behaviour could serve to groom many group-members at once, all the rest of linguistic behaviour followed as a quite natural consequence. An example of this approach of a `single magic bullet' to language evolution can also be seen in the title of an article by Bickerton, ``Catastrophic evolution: the case for a single step from protolanguage to full human language'' (Bickerton, 1998). In recent years, this view of the `magic bullet' has given way to a more complex view which recognizes that there are necessarily many interweaving stories to tell. Jackendoff's (2002) book replaces Bickerton's single automatic cascade with a scenario of eight stages, none simply following automatically from the previous, though sometimes enabled by it -- and this only deals with the evolution of syntax.

Linguists are deeply conscious of the distinctions between the various subsystems of a language. The make-up of most university curricula in Linguistics attests to this, with separate courses on Pragmatics, Semantics, Morphology, Syntax, Phonology, and Phonetics. The professional specializations of linguists also reflect these divisions: very few syntacticians or semanticists have anything to do with phonetics or phonology, very few pragmaticists and discourse analysts get involved in the intricacies of syntactic or morphological theory. As the principles of pragmatics are entirely different from those of, say, syntax or phonology, it would be correspondingly inconceivable to a specialist in one of these fields that the same evolutionary story could be told about any two of them. Since professional linguists have been notoriously reluctant to get involved in theorizing about language evolution, it has been possible for accounts of the `magic bullet' type to pass without much protest.

Hauser et al. (2002) make a useful, and long overdue, distinction between the faculty of language in the narrow sense (FLN) and the faculty of language in the broad sense (FLB). FLN is that bundle of features that distinguishes human language from the behaviour of other species. For decades now, many linguists, and particularly those in the generative tradition, have confined their attention to the most characteristic aspects of language, focussing on structural properties of morphosyntactic and phonological systems. Hauser et al. speculate that FLN may contain no more than the human capacity for recursion, and perhaps not even that. This bold conjecture would place even phonology outside FLN, implying that aspects of phonological organization, such as the structuring of syllables, have counterparts in animal vocalizations. Naturally, the conjecture has been challenged, with other linguists, in particular Pinker and Jackendoff (2005), pitching in to claim that there is much more that is specific to the human language faculty than just recursion. Whatever conclusions are ultimately reached, the FLN/FLB distinction serves the useful purpose of requiring, at the very least, that we tell separate stories about the evolution of FLN and the evolution of FLB. Theories about the evolution of FLB-minus-FLN have to account for behaviours in non-humans homologous with aspects of human linguistic behaviour. Theories of the evolution of FLN, on the other hand, are more problematic, in that the relevant data only come from one species.

Theories of the evolution of language in the broad sense (FLB) involve those domains where a language system interacts with non-linguistic systems. These can be identified as the domains of pragmatics, semantics and phonetics. One consequence of an evolutionary approach to language is that it forces a generalization of definitions of the linguist's traditional components of linguistic theory. Definitions of these fields suitable for evolution studies necessarily differ from those usually adopted in Linguistics, in making no mention of language. Language evolved out of non-language, and these definitions are appropriate for envisaging pre-linguistic foundations upon which phonetics, semantics and pragmatics could plausibly be built, pre-phonetics, pre-semantics and pre-pragmatics, if you will.

It is convenient to define the evolution of phonetics as the evolution of the hardware used in spoken language. This postulates a hardware/software division between phonetics and phonology. Put another way, the evolution of phonetics involves the biological evolution of the human auditory system up to the cochlea and the auditory nerve, and the biological evolution of the vocal tract. These are peripheral perceptual and motor systems. The evolution of phonology, on the other hand, is a story of how higher-level systems arose for coordinating the complex inputs and outputs of these peripheral systems into discrete phonological units for purposes of communication. While all these stories surely interweave, they are also surely partly separate. The human auditory system has much in common with those of birds and other mammals, whereas the human vocal tract is more specifically distinctive and adapted. The evolution of human phonological systems is at least in part a matter of self-organization subject to cultural pressures of communication within an overall biological envelope. Prosodic systems and inventories of phonemes do not evolve biologically. Recent works (e.g. de Boer, 2000a, 2000b; Oudeyer, 2006) have considerably elaborated on the initial ideas about self-organization in phonology by Lindblom and others (see Lindblom et al., 1984).

At the other end of language from phonetics stand pragmatics and semantics. Semantics, in a sense suitable for discussing the evolution of language, is the system of relations between the world outside the organism and internal representations of that world. This accommodates the essential extensionalism of linguistic semantics, while not being dependent on the existence of language. At some stage in evolution, animals began to have mentally organized (proto-)concepts of objects, events and situations in their environment. Higher animals closely related to humans certainly have systematic representations of selected aspects of their environment, or they could not survive. Let us call such representations pre-semantics.

Pragmatics could conceivably be defined as the domain of an organism's overall practical behaviour in relation to every relevant thing in its environment. But with a view to the goal of accounting for the origins of linguistic pragmatics, it is convenient to settle for a narrower, specifically social, definition, according to which pragmatics is the domain of an organism's behaviour toward its conspecifics. Let us call this pre-pragmatics? Conceivably, some extremely non-social animal could survive with minimal or zero pre-pragmatics. Animals, such as certain fish, which deposit their eggs or sperm without encountering their `mates' may well have virtually no behaviours specially tuned to conspecifics. They may eat each other or ignore each other, based on the same criteria as their interactions with all other species or indeed all other objects. The only glimmer (if that) of pre-pragmatic behaviour in such a fish is the male's selective depositing of sperm on the eggs of it own species.

Conversely, there can be pre-pragmatic communicative behaviour which does not involve pre-semantics. This is communication between conspecifics conveying no information about the sender's internal representation of entities other than the two animals involved, the sender and the receiver. In linguistic terms, such communication is non-referential, with the same flavour, mutatis mutandis, as pure non-propositional speech acts in human language, such as greeting, threatening and submitting.

A crucial step in the evolution of language was the bringing together of pre-semantics and pre-pragmatics. Propositional communication involves an organism `opening up' its internal representations of the world to a conspecific, as enabled by developing conventions of naming and reference. This is no trivial step. Humans are the most communicative of species, and it remains an evolutionary puzzle why we are so generous in revealing our internal states to others. The various extant theories of altruism (Hamilton, 1963, 1964; Trivers, 1971; Axelrod et al., 2004) begin to approach this problem.

2. Animals have Rich Conceptual Systems

A common assumption is that, because animals do not signal to us what they are thinking, they cannot be thinking very much. But of course, this conclusion does not follow. For all animals in the wild, we still have very little idea of the content of the signals they exchange among themselves.

So far, it has been possible to recognize pure pre-pragmatic communicative acts, such as the submission gestures of baboons, the threatening charges of male gorillas, and the often elaborate courtship rituals of birds. These are purely pre-pragmatic because they cannot be said to be about any third entity; they just concern the relations between the sender and the receiver. When a human farmer threatens a trespasser to get off his land, there may be some shared concept of the relevant patch of land present in the minds of both farmer and trespasser. But when a male gorilla charges an intruder, as far as we know there is no need to postulate any such shared concept of a third entity in the gorilla or the intruder. Conceivably it could be argued that there is some common understanding between the gorilla and the intruder about a mutually acceptable distance from the origin of the charge, in that the intruder knows to run, and the gorilla gives up after a certain distance. But the facts can equally well be explained by a waning of attention on the part of the charging male.

Semantic signaling has also been observed in the wild. This is signaling conveying information about something other than the sender and the receiver. Predator-specific alarm calls, as seen in many species of monkeys and chickens are an obvious example. Such cases reinforce the view that the animals partition their environment into distinct categories. But so far there is no evidence from observation of behaviour in the wild that such semantic signaling ever conveys any structured information. For instance, though a vervet can signal LEOPARD, it never, as far as we know, signals anything as complex as LEOPARD BEHIND YOU or LEOPARD IN THAT TREE. But the absence of signals conveying complex thoughts should not lead us to conclude that the animals are incapable, privately, of entertaining such structured thoughts.

A claim has been made (Hurford, 2003) that the separation of function between ventral and dorsal pathways in the brains of humans, non-human primates, and even rats and cats can be correlated with the logical distinction between a (one-place) predicate and its argument. The basic idea is that the dorsal cerebral mechanism is responsible simply for locating and attending to any arbitrary salient object in the environment; this corresponds to the delivery of a logical individual variable. The other mechanism, the ventral, is responsible for making a judgement about the properties of the object, such as its colour, or shape, or some more complex property, such as which face it is; this corresponds to the assignment of a predicate to the individual variable. The overall resulting brain activity can thus be interpreted as corresponding to a logical representation of the form PREDICATE(x). This rather abstract proposal assumes that mental representations are physically instantiated in patterns of neural activity. Asymmetries and functional separations in neural activity are candidates for matching up with the categories of logic, the other discipline dealing with the mental. Each such proposed matching needs to be argued for on its own merits; many constructs invented by logicians may turn out to be neurologically unjustified. Hurford's claim is that a correlate of fundamental logical structure, the asymmetric combination of ontologically distinct terms, predicates and their arguments, exists in animals without language, all mammals, as far as we know. Such basic representations are not as complex as would be conveyed by sentences such as There's a leopard in that tree. But LEOPARD(x) does manage to capture the attribution of a property to a deictically indicated object, as in That's a leopard! or the assertion of the existence in the situation of utterance of an object satisfying some predicate, as in There's a leopard (round here)! This proposal is a direct challenge to philosophers, such as Davidson (2001) who deny propositional form to creatures without language. Animals with a dorsal/ventral separation like that found in macaques would seem to be capable of privately entertaining a thought equivalent of PREDICATE(x) form without necessarily ever expressing this thought in a public way.

The evidence from experiments with captive animals indicates that they can in fact entertain some quite complex thoughts. An increasing amount of research has begun to shed light on just how rich the conceptual lives of non-human animals are, or at least can be. The most spectacular case is that of Alex the African grey parrot, as reported in many articles by Irene Pepperberg and summarized in her book The Alex Studies (Pepperberg, 2000). At a relatively simple level, Alex can identify colours, shapes, materials, and low numerosities. For example, on being shown an object and asked what colour it is, he typically responds with the right colour word, such as red or blue. It is important to note that even this apparently simple task goes beyond First Order Predicate Logic. Categorizing an object as, say, blue, involves a first-order predication, applying a predicate to an individual object, for example BLUE(x), where `x' denotes the arbitrary object attended to. But the question put to Alex contains the word colour, and so Alex has to be capable of selecting which class of first-order predicates are suitable for his response. He is clearly capable of correctly applying a second-order predicate to a first-order predicate, as in COLOUR(BLUE). Even more impressively, on being shown an array of objects and asked `What's same?' (or `What's different?'), Alex can correctly pick out the relevant second-order predicate, e.g. SHAPE, if the objects are the same (or different) in shape. In the case where no two objects in the array have the same shape, colour or material, the answer to `What's same?', usually correctly given by Alex, is `None'. This involves scanning all the objects in the array and comparing all their relevant attributes, keeping the interim results of comparison in memory until all attributes of all objects have been registered, and then delivering the appropriate response. Again, this involves at least second-order predication, and either negation or universal quantification (or both). By any standards, these are relatively abstract concepts. It seems that Alex is generalizing. That is, the questions he is answering correctly are not necessarily questions whose specific answers have been inculcated by rote-learning.

Alex is not as well motivated to learn as a human child, and must often be cajoled to perform. The breakthrough that Irene Pepperberg has achieved with Alex is similar to that achieved by Sue Savage-Rumbaugh with the bonobo Kanzi in its advance on pure behaviourist stimulus-response techniques (see, for example, Savage-Rumbaugh et al. 1998). In both cases training was more similar to a human child's exposure to language than reward-punishment regimes. Alex was trained indirectly by being made witness to exchanges between a trainer and a third party, a `rival'; he watched the rival's behaviour and then was able to duplicate it. Kanzi, also, was not explicitly trained, but picked up his first symbols through watching unsuccessful attempts to train his mother by more direct methods.

A recent report of research on dolphins and macaques concludes that 11[T]here is a strong isomorphism between the uncertainty-monitoring capacities of humans and animals. Indeed, the results show that animals have functional features of or parallels to human metacognition and human conscious cognition.'' (Smith and Washburn, 2005:19) `Uncertainty-monitoring' refers to the capacity to recognize how sure or unsure one is in making a judgment. In other words, it has been shown that these higher animals are not only able to register a `Yes' or `No' answer to a challenge put to them by a researcher, but also to register an `I don't know' response. They know whether or not they know something. This reveals a degree of self-awareness, or metacognition, previously unsuspected in non-humans.

In conclusion of this section: animals may have quite complex representations of the world, they are just not naturally disposed to communicate these internal intentional states.

3. Primates are not Necessarily the Closest to Human-Like Capacities

Until recently, it has seemed natural to expect that the density of features of language-like behaviour in animals would decline smoothly as a function of their genetic distance from humans. It was assumed that chimpanzees would be closest to humans in language-readiness, followed by the other great apes, then other primates, and then other mammals, and that birds would be expected to be a long way from humans in any mental capacity underpinning language. The previous section described the competence of Alex, a parrot. Alex has truly been a surprise. He may be a uniquely smart parrot, but his accomplishments, and those of other bird subjects, have put into question the assumed simple inverse correlation between language-readiness and genetic distance from humans. A recent book (Rogers and Kaplan, 2004), with the suggestive subtitle Are Primates Superior to Non-Primates? focuses on this question. The lead article in that collection (Emory and Clayton, 2004) concludes ``Although avian and primate brains differ significantly in size and structure, similar principles of organisation are evident. We suggest that birds and primates reflect a case of divergent evolution in relation to neuroanatomy, but convergent evolution in relation to mental processes.'' (Emory and Clayton, 2004:4)

Birds, unlike mammals, do not have larynxes, but have syrinxes instead. But many species of birds can use their syrinxes to imitate sounds made by their own or other species or even non-animal sounds, like the sound of a creaky door. In parallel with Emory and Clayton's conclusion about avian and primate brains, quoted above, birds and mammals reflect a case of divergent physiological evolution, but convergent evolution in relation to vocal function.

All birds and mammals can make vocal sounds, but not all can do vocal imitation. Vocal imitation is the capacity to acquire new controlled sound-producing behaviours by using a homolog or analog of the human vocal tract. Among mammals, our closest relatives, the great apes, are notoriously poor at vocal imitation. Many birds easily outperform them. But some more distantly related mammals have vocal imitation capacities superior to those of apes. A recently discovered case is that of elephants, reported by Poole et al. (2005). They observed an elephant imitating the sound of a lorry, and elephants are known to make individual learned signals, apparently for purposes of contacting and bonding with other individuals. Vocal learning is also well attested in whales and dolphins (Janik and Slater, 1997; McCowan and Reiss, 1997). Vocal learning has been observed in some species of bats. ``Female greater spear-nosed bats, Phyllostomus hastatus, live in stable groups of unrelated bats and use loud, broadband calls to coordinate foraging movements of social group mates. Bats benefit from group foraging. Calls differ between female social groups and cave colonies, and playback experiments demonstrate that bats perceive these acoustic differences. ... [T]he group distinctive structure of calls arises through vocal learning. Females change call structure when group composition changes, resulting in increased similarity among new social group mates. Comparisons of transfers with age-matched half-sibs indicate that call changes are not simply due to maturation, the physical environment or heredity.'' (Boughman, 1998:227). Across bird species, the capacity for vocal learning is also curiously patchy; only three groups of birds exhibit it -- parrots, songbirds and hummingbirds (for hummingbirds, see Jarvis et al. (2000) and references there.)

A further example of the lack of correlation between a capacity necessary for language and genetic closeness to humans is found in the pre-pragmatic abilities of domestic dogs, who can follow pointing signals better than chimpanzees. ``Dogs are more skillful than great apes at a number of tasks in which they must read human communicative signals indicating the location of hidden food. In this study, we found that wolves who were raised by humans do not show these same skills, whereas domestic dog puppies only a few weeks old, even those that have had little human contact, do show these skills.'' (Hare et al., 2002:1634)

Basic tool use is not directly connected to language, but, like some pre-adaptations for language, tool use is another feature of behaviour that is now being found in species quite genetically distant from humans. The fashioning of sticks by chimpanzees to fish for termites is well known. Dolphins have now been shown to use sponges for foraging, and it is forcefully argued that this behaviour is culturally, rather than genetically, transmitted (Krützen et al., 2005). Interestingly, and perhaps problematically for the claim of cultural transmission, this behaviour is only transmitted down the female line, from mothers to daughters. Among birds, New Caledonian Crows have been found to be capable of making tools (Kenward et al., 2005). As with behaviours contributing to langage-readiness, we see in tool use the same kind of patchwork distribution over species, not smoothly correlated with genetic distance from humans.

To summarize this section, various necessary but not sufficient features of language-readiness can be found distributed around a number of species not as closely related to humans as the great apes. Only humans have accumulated a sufficient set of such features, enabling them to become the only truly language-ready species.