Evolving Honest Communication Systems: Kin Selection and "Mother Tongues"

Evolving Honest Communication Systems:
Kin Selection and "Mother Tongues"
W. Tecumseh Fitch
Harvard University
(To appear in: The Evolution of Communication Systems: A Comparative Approach (Ed. by Oller, D. K. & Griebel, U.).
Cambridge, Massachusetts: MIT Press.
In a famous passage, J B S Haldane (1955) conveyed the seed of the idea of kin selection when he
acknowledged the selective advantage of saving, at risk to his own life, drowning brothers or cousins,
but not more distant relatives. In an odd turn for so insightful a biologist, he then concluded that it was
highly unlikely for such logic to explain known examples of altruism. Haldane's reasoning was simple
and logical: in a large population the average relatedness would be much smaller than the 1 in ten risk
of drowning, and it would indeed be unprofitable, genetically speaking, to jump to the aid of a
randomly-chosen individual. What is peculiar is that Haldane overlooked the fact that a "gift" of
altruism, if bestowed selectively to closely-related kin, could easily be selected for, thus leaving it to
Hamilton (1964) to comprehend and formalize inclusive-fitness theory, and make the most significant
contribution to evolutionary theory since Darwin. Indeed, such "altruistic" acts need only satisfy
Hamilton's famous equality Br > C, (the Benefit to kin, as diluted by their fraction of relatedness r,
must exceed the Cost to self), for selection to favor the action. Perhaps Haldane's reluctance to
acknowledge this possibility was influenced by his own rather unusual experience of twice saving a
drowning individual, when he gave no conscious thought to relatedness. The unsavory implications of
doing otherwise were resurrected by the term "nepotism" associated with early experimental work on
kin selection in the 1970's (e.g. Sherman, 1977). Perhaps distaste for nepotism partially accounts for
the fact that kin selection has only slowly begun to be integrated into mainstream theory on the
evolution of communication (Grafen, 1979; Maynard Smith, 1978; Maynard Smith, 1991; Maynard
Smith, 1994; Johnstone & Grafen, 1992a; Godfray, 1991; Lachmann & Bergstrom, 1998; Bergstrom &
Lachmann, 1998b)
My aim in this chapter is to help move this process of integration forward, particularly in the context
of the evolution of human language, where the intersection between kin selection and communication
theory appears to have considerable implications. I will refer to systems of communication that have
evolved in a context of kin selection as "mother tongues", and will argue that such systems have very
attractive theoretical properties for the evolution of rich communication systems like spoken language.
Namely, mother tongues can be selected for accurate or "honest" communication (because senders and
receivers sometimes have each others' best "genetic interests" at heart), and for semantic complexity
(exchange of detailed information being thus valuable by increasing inclusive fitness, to a theoretical
limit set only by the complexity of senders' and receivers' mental structures). I will suggest that these
dual virtues allow mother tongues to evade the evolutionary traps of constant Machiavellian deceit, or
wasteful Zahavian handicaps, that can bedevil communication systems among nonkin.

---

Furthermore, in some cases, selection on mother tongues will favor complexity for its own sake, allowing
the formation of "kin dialects" which increase the accuracy of kin recognition. This observation, when
applied to human language, may account for the otherwise very curious fact that language seems more
complex than necessary for communication, in the sense that our ability to recognize regional, class or
other dialects far exceeds the needs of semantic communication. Great English authors like Vladimir
Nabokov or Joseph Conrad (born in Russia and Poland respectively), despite enormous grasp and fluid
command of the English language, still spoke with an accent and sounded detectably "foreign" to native
English speakers. Highly sophisticated concepts are exchanged between native and non-native
speakers on a daily basis, throughout the world, but these speakers nonetheless perceive each others'
speech as distinctly different. These everyday facts are inexplicable if human language evolved solely
for purposes of semantic communication: why should our phonological system accurately reproduce a
greater complexity than needed to efficiently transmit propositional information? I will suggest that
this can be easily accounted for if the existence of such dialectal variation increases the ability of
distant kin to recognize each other, and thus allows selective transmission of valuable information
among kin.
In this paper I provide a brief introduction to some relevant evolutionary theory (including kin
selection and theory on the evolution of communication) and consider some of the selective forces
previously proposed to underlie the evolution of human language. Then I will advance the argument
for mother tongues, reviewing known examples of kin-specific communication in nature and relating
them to the evolution of extended parental and sibling care in birds and mammals. The massive
transmission of information that occurs during the extended learning periods of childhood in many
higher vertebrates can strongly select for vocal systems which aid this exchange even slightly. These
are the conditions, I argue, that lead to many of the honest low-cost signals believed to represent
mother tongues in animals, including squirrel alarm calls, primate grunts, the purrs of cats and other
carnivores, ultrasonic calls of rats, signature whistles in dolphins, and others. Despite their ubiquity,
such systems appear to lack the structural complexity that would be necessary to convey arbitrarily
complex thought, as human language does. Paradoxically, the most structurally complex
communication systems in nature, other than human language, are probably the learned passerine bird
and humpback whale "songs", which as far as we know convey no propositional information at all. I
propose that it was the co-occurrence of mother-tongue selection with additional forces selecting for
structural complexity that provided the apparently unique evolutionary conditions within which
language evolved, and suggest that one of these additional forces was the use of dialects to identify
distant and previously unknown kin. Beyond a certain critical mass of communicators, the large shared
pool of information made available to a widely-extended kin group becomes a highly-adaptive
resource, and almost irresistibly selected for. The mother tongue hypothesis thus bypasses many of the
problems currently plaguing neo-Darwinian theories of the evolution of language, as well as
identifying numerous relevant points of contact between human language and communication systems in
other animals.

---

Background: Kin Selection, Signaling Theory and the Evolution of Language
Altruism & Kin Selection: The notion of kin selection and the theory of inclusive fitness represent the
th
most important contribution to evolutionary theory in the 20 century, and provides an explanation for
"altruistic" phenomena, such as sociality in insects, that deeply troubled Darwin. The central notion,
as already intuited by (Haldane, 1955), was that fitness is not simply a factor of an individual's
survival, nor its success at producing offspring, but also the reproductive success of all those relatives
who share its genes (Hamilton, 1964b; Hamilton, 1964a). To the extent that an individual's actions aid
relative's survival or reproduction, at a minor cost to itself, they will increase its overall inclusive
fitness. This (in retrospect rather intuitive) concept of inclusive fitness has important implications for
the evolution of social behavior.
Hamilton realized that, from a gene's perspective, the critical question is not the survival of a
particular mortal body in which it finds itself, but in the propagation of copies of itself in any body.
Hamilton further realized that any mechanism whatsoever that allowed altruists to "donate" their
resources selectively to kin would be powerfully selected for, as long as the exchange met the condition
t h a t
C < Br
where C is the cost to the donator, B is the benefit to the receiver and r is the fractional coefficient of
relatedness (between 0 for unrelated individuals and 1 for clones or identical twins), which acts to
"discount" the benefit accrued by the recipient. Any act for which the benefit to the recipient,
discounted by its relatedness, exceeds the cost to the donator, shall increase the donator's inclusive
fitness, and should be selected for. This reasoning provided an immediate solution to the problem of the
"altruism" of female honeybees, ants and other eusocial insects, who cooperate to raise their sisters,
and protect them at the cost of their own lives, while bearing no young themselves. This problem was a
serious worry for Darwin, who saw honeybee altruistic behavior as directly contrary to his theory of
natural selection. In the framework of inclusive fitness, eusociality follows directly from a genetic
peculiarity of Hymenopteran insects, which results in sisters being more closely related to one another
than they are to their own offspring (Wilson, 1975).
A second theoretical explanation for apparently altruistic acts does not require kinship. This is known
as "reciprocal altruism" (Trivers, 1971), in which two unrelated individuals mutually benefit by taking
turns exchanging resources at times when the cost to the donor is low and the benefit to the receiver is
high. Such behavior requires a rather special set of circumstances, and empirical demonstrations of
reciprocal altruism in nature are rare at best. (Note that the claim of (Packer, 1977) of reciprocal
altruism in baboons has repeatedly failed to replicate (Bercovitch, 1988), and the exchange of blood by
vampire bats studied by (Wilkinson, 1984) was almost entirely among kin). Thus despite the ubiquity
of reciprocity in human society (the entire global economy can be thought of as a huge, institutionally-
regulated system of reciprocal altruism), the available evidence suggests that such systems are very

---

rare in animals, in sharp contrast to kin-selected systems which are very common. Despite the obvious
fact that modern human languages are used by non-kin to reciprocally share information, contemporary
language is always used quite intensively among kin and there is no a priori reason to assume a
preponderance of non-kin communication in early stages of language evolution. Kin communication,
with ubiquitous nonhuman examples known, may provide an easy evolutionary route to reciprocal
sharing among non-kin, but the converse is not true. Thus, in what follows, I will focus on kin selection
and make little mention of reciprocity among non-kin.
Honest Signaling Theory: "Honest" signals are those which accurately (though not necessarily
perfectly) convey some information about some relevant quality of the signaler (e.g. its species, sex,
size, condition, etc.) (Dawkins & Guilford, 1991). Like many terms in modern ethology, this one should
be interpreted in the technical sense just given, not by the layperson's model of human honesty, which
includes assumptions of self-knowledge, intention to communicate and various other complications.
Thus, if a newborn baby cries only when requiring care, this is an "honest" signal, despite the fact that
few would attribute honesty, in the ordinary sense, to newborns. Below I will dispense with the
quotation marks around the term, with the understanding that "honest" is being used in its technical
sense throughout.
A long tradition in ethology assumed that much communication evolved to facilitate honest
communication, particularly among kin (Dawkins & Krebs, 1978; Hinde, 1981). This assumption came
under strong attack in several seminal papers by Amotz Zahavi (Zahavi, 1975; Zahavi, 1977), which
spurred a large and still growing literature on the theory of honest signaling. The central claim of
Zahavi's attack, echoed less forcefully in (Dawkins & Krebs, 1978), is that such honesty cannot be
simply assumed. In fact, natural selection should in many cases favor animals which are deceitful
(again in a non-cognitive sense) if dishonest behavior leads to personal advantage and increased
reproductive success. This is easily understood when we consider the advantages obtained by liars and
cheats in human sociey if they escape detection and punishment. By highlighting the readiness with
which Machiavellian deceit can destabilize honest signaling systems, this perspective essentially
turned the tables, suggesting that the real question for theorists interested in the evolution of
communication is why any signaling systems are honest (if indeed they are). Zahavi's proposed
solution to this question, the "handicap principle," embodies a rather non-intuitive claim: that honest
signals are possible only when the signaler pays a high cost when emitting the signal. According to
Zahavi (Zahavi, 1993), such costs are necessary if a signal is to stay honest and remain in circulation
over evolutionary time. Despite early critiques of this idea from a mathematical viewpoint (e.g.
(Maynard Smith, 1976)), the handicap principle received theoretical support from a complex
mathematical model of signaling introduced by Grafen (Grafen, 1990a; Grafen, 1990b) and has since
generated a weighty volume of theoretical work, along with a less impressive body of empirical work.
Recently some results of early theoretical work were discovered to be dependent on errors in the original
papers (Siller, 1998). It is far beyond the scope of this paper to review all of this literature, much of

---

which is highly technical, (see Johnstone, this volume, for a more detailed introduction). Thus I confine
myself here to three major themes that have emerged from this literature.
Any signal obviously bears some cost (even if only the time wasted not doing something else)(Maynard
Smith & Harper, 1995). Unless the handicap principle requires more stringent conditions on signaling
costs than their simple existence, it reduces to this obvious and uninformative fact. Empirical
demonstrations of a cost of signal production thus provide no support for the handicap principle, which
demands additional "strategic" costs over and above those necessary to produce a detectable signal.
When the costs of signals have been measured, highly costly signals appear to be the exception rather
than the rule, contrary to the predictions of the handicap principle. For example, vocal signals in
many vertebrates appear to have surprisingly low metabolic costs (Chappell et al., 1995; Horn et al.,
1995; McCarty, 1996), while still maintaining honesty; the same is true of human speech. Various
signals, such as piloerection ("raising the hackles") or crest erection in birds, appear to have virtually
no physiological costs but are extremely common, (Marler, 1968; Wilson, 1972). Furthermore, signaling
systems evolve within a context of physical and physiological constraints which may make honesty
difficult or impossible to escape (Maynard Smith & Harper, 1995). In such cases honesty is the default
condition and it is dishonesty which demands an adaptive explanation ((Fitch & Hauser, in press)).
An example are formant cues to body size which are present in vocalizations of many birds and
mammals. Formants are the resonances of the vocal tract, and their frequencies depend on the length of
this tube of air in a manner that follows from basic acoustic laws. Because (in most cases) vocal tract
length is itself necessarily correlated with body size, formants provide a "free" cue to body size in any
vertebrate that produces a vocalization (Fitch, 1997; Fitch & Giedd, 1999). Such size information does
not cost anything extra to encode: it is present as a result of the physics of sound production and the
anatomy of the vocal production apparatus. Although selection has acted in some cases to exaggerate
these formant cues by elongating the vocal tract beyond its normal anatomical confines (Fitch & Reby,
2001; Fitch, 1999), it is this deception which is an adaptation, not the original honest signal which
existed, and was stable, by default.
Second, an important limitation of much of the early signaling theory is that it did not include the costs
and benefits to receivers, implicitly assuming the "assessment costs" to be at or near zero. Later work
redressed this oversight, both by focusing on the costs to receivers of eliciting or evaluating signals
(Dawkins & Guilford, 1991), or on the effects of less-than-perfect signal reception (Wiley, 1994;
Johnstone & Grafen, 1992b). These results again weaken the central claim of the handicap theory,
showing that "conventional" signaling systems, which convey some information most of the time, but
tolerate a low level of deceptive signaling, can be more stable evolutionarily than totally honest
systems which exact a high price from both signalers and receivers. A nice example of the cost of
assessment are red deer roaring battles, in which rival males must engage in a potentially exhausting
(and dangerous) roaring display in order to incite a resident male to roar at full strength (Clutton-Brock
& Albon, 1979). The costs paid by receivers are clearly non-negligible in this well-studied case (though
it appears that various physical constraints may provide an additional guarantee of some level of

---

honesty (Fitch & Reby, 2001)). In general threat displays should tolerate some level of bluffing,
because the costs to intruders of "probing" could be severe bodily injury if the signaler is NOT bluffing
(Adams & Caldwell, 1990). Similarly, it may not pay a female choosing a mate to spend weeks
evaluating the quality of her potential mates. Instead the best strategy can be to choose one who looks
or acts like a past good mate, or even simply choose the male that other females are mating with (e.g.
(Losey et al., 1986)). These issues are important, and virtually ensure that evolution will not fill the
world with arbitrarily costly signaling systems. However, while such considerations are important in
understanding the limits of the handicap notion, they are less relevant in the quest for cheap, honest
signaling systems that I am concerned with here. Their main importance in the context of the evolution
of language is that the cost of discovering the truth provides a theoretical model for how a low level of
dishonesty can persist indefinitely in a basically honest system. Such a characterization would appear
to fit human language rather well, and perhaps can account for the obvious fact that, despite its
undeniable (and quite remarkable) capacity for honest transmission of information, language is not
always used honestly.
Finally, the most significant modification of honest signaling theory comes from a consideration of
communication among kin. This issue has been studied in the context of the Sir Philip Sydney game,
introduced by (Maynard Smith, 1991), a stylized theoretical framework in which an individual must
decide whether to donate some indivisible resource to a needy relative. Donation to a relative in this
framework allows a potential inclusive fitness benefit to the donor at an immediate cost to the donor.
Maynard Smith found that an honest signaling system can evolve in the presence of low or zero, cost to
the signaler if sender and receiver interests do not conflict (technically, if the outcomes are ranked
equivalently by both participants). Further models have extended this theory (e.g. (Godfray, 1991;
Johnstone & Grafen, 1992a), also suggesting the possibility of low signaling costs for closely-related
individuals. This work, especially that concerning begging by offspring from their parents, has been
tested empirically as well, and very low physiological costs have been measured in beggin nestling
birds (McCarty, 1996). More recently, an important series of papers have found that when certain
assumptions of earlier models are relaxed, multiple low-cost, or cost-free stable signaling systems can
evolve among kin (Bergstrom & Lachmann, 1998b; Bergstrom & Lachmann, 1998a; Lachmann &
Bergstrom, 1998). Thus given the partially shared genetic interests among kin, cheap or free honest
communication systems can evolve and will be evolutionarily stable.
Animal Mother Tongues: Kin communication in other species
I'll now give a brief and selective overview of kin communication in animals to highlight some of the
empirical work that has been done on this topic. There are a number of interesting phenomena that
could be mentioned but I'll stick to three: food calls as an aid to food learning, nestling begging calls,
and alarm calls. I'll focus on the latter, which has the largest body of empirical work devoted to it.
In contrast to altricial birds which are typically born totally blind and helpless, and are fed by the
parents for days or weeks after birth, many precocial birds such as grouse, ducks and chickens must

---

essentially feed themselves from a very early age. In such species, it is quite common for the young to
follow their mother, providing an opportunity to learn by example from her behavior. The existence of
food calls, emitted when sighting food or feeding on it, provides a nice example of a very simple form of
kin communication that can transmit useful, learned information about what is (or is not) edible.
Simply by feeding on a particular substance, and emitting food calls to attract the attention of her
young, a mother provides an opportunity for her children to benefit from her past experience and thus
bypass a certain amount of trial and error learning. This may be the simplest example of kin
communication that helps transmit learned information, and to be valuable it requires no explicit
"teaching" or sophisticated mental model of offspring knowledge. A similar example is provided by
migrating birds. In many species juveniles accompany their parents on the first long-distance migration
and thus learn safe migration routes and stopping points (Matthews, 1968).
Many birds and mammals emit characteristic calls at the appearance of predators, which alert
conspecifics to the predator and often immediately elicit escape reactions. The existence of such "alarm
calls" in birds and mammals was recognized as a problem by evolutionary theorists four decades ago
(Hamilton, 1964b; Maynard Smith, 1965), and played an important role in the development of kin
selection theory. The problem is simple: from an individually selfish viewpoint, why should an
organism spotting a predator vocalize, and thus call attention to itself, when it could just slink away,
leaving its unsuspecting groupmates to be attacked? Although various "selfish" proposals have been
offered, e.g. that the call deters predators from repeated hunting at that site (Trivers, 1971; Sherman,
1985), these are seen as relatively implausible compared to the alternative: that calls serve to warn
kin and thus increase inclusive fitness. The mathematical prerequisites for this are quite simple
(Hamilton, 1964b; Maynard Smith, 1965; Charnov & Krebs, 1975): the alarm call must be relatively
low-cost and the species must inhabit groups that contain kin. During the 1970's, a wealth of
comparative data were gathered on this topic, mostly from ground-living squirrels, that provide strong
empirical support for this kin-communication hypothesis. In two seminal papers (Dunford, 1977;
Sherman, 1977), different species of ground squirrel, females with kin present were found to be the
predominant alarm callers (males were silent, except as juveniles living with their mothers and
siblings). Males in general, and newcomer or transient females, did little alarm calling. Sherman's
paper further demonstrated a cost to calling: callers were significantly more likely to be killed than
non-callers. Further data on other species substantiated these conclusions (Hoogland, 1983; Smith,
1978; Barash, 1976) with the interesting twist that, in those species where males participate in
parental care and/or live amongst kin, males also call. Although it may be the case that much of the
alarm calling serves to protect an individual's offspring, Sherman (Sherman, 1980) provides data
indicating that offspring are not the only kin "protected" via alarm calls, and therefore that alarm
calls are kin-communication in the wider sense, not just parental care (Shields, 1980).
Of course, once a proclivity for alarm calling in general is established, kin selection might also act to
increase the specificity of the calls, for example to distinguish aerial from ground predators, an
elaboration observed in a wide variety of bird and mammal species (Klump & Shalter, 1984), including

---

nonhuman primates (Seyfarth et al., 1980). The "honesty" of these differentiated signals is easy to
explain in the context of kin selection: to the extent that there are different optimal escape strategies
for different classes of predator, an individual will increase its inclusive fitness by emitting calls
different enough to allow listening kin to adopt the best escape strategy. Note that this does not
require intent to label different predators on the part of the caller: a simple difference in arousal caused
by different predators would suffice, if it leads to discriminable differences in call acoustics (Owings &
Hennessy, 1984; Seyfarth & Cheney, 1997). Screams of surprise might reliably signal less dangerous
predators than screams of terror. Thus one does not have to posit intentional "predator labeling" to
understand the adaptive value of signal elaboration in the context of kin-selected alarm calls. Of
course, the number of differentiated signals that are necessary in this context will be limited by the
number of predators (or more generally, of types of danger that require differing responses). Even in the
case of highly preyed-upon species like vervet monkeys (with 16 known predators, (Cheney &
Seyfarth, 1981)) or Belding's ground squirrels (with 9 known predators, (Sherman, 1985; Sherman,
1977)), alarm calling alone will never lead to an infinitely extensible set of vocalizations. Thus, alarm
calls provide strong selection for honest communication, but not for unbounded generativity.
Lest it be thought that ALL kin communication is harmonious, the work on nestling begging must be
mentioned. Nestling begging is a type of kin communication that has been studied from both theoretical
(e.g. (Godfray, 1991)) and empirical perspectives ((McCarty, 1996; Briskie et al., 1994; Haskell, 1994)).
Begging involves food as a benefit (not just information), and so differs considerably from the above two
examples of mother tongues. The competition between nestmates for food can be extremely fierce in
birds, sometimes leading to siblicide, and thus it would be unsurprising if some "dishonest" begging
occurred. However, the evidence from metabolic rate suggests that begging is not particularly costly
((McCarty, 1996)), and no clear examples of dishonesty are known, even in this highly competitive
situation. The use of acoustic signals to acquire food or other resources from the parents may lead to a
much greater degree of sibling competition and/or parent/offspring conflict than if the young are being
"fed" information, at low physiological cost, by the parents.
To summarize, I have argued that mother tongues provide, in many cases, the preconditions for the
evolution of honest, low-cost (or "cost-free") communication systems. Using the prototypical example of
alarm calls, I showed that such systems are probably common, if not ubiquitous, in kin-group-living
mammals. Thus the theory of mother tongues, by focusing on the genetic common interests of kin, seems
to satisfy one of the primary desiderata for a theory of language evolution: honest communication
without handicaps. I submit that this is an important point in its own right. I also admit that it seems
(to me at least) relatively obvious; hence my surprise that this hypothesis appears to have eascaped
explicit mention by previous theorists, who have typically focused on communication among non-kin
adults when discussing language evolution.
Critical Hurdles in the Evolution of Language

---

The brief overview of signaling theory above shows that, from the viewpoint of natural selection,
honesty is not always the best policy, and that when honesty exists, it demands an explanation. This
has some important implications for the evolution of human language. Spoken language is low-cost (see
below) but has an unparalleled capacity to honestly convey detailed and arbitrarily complex
information. Thus language is quite anomalous from the viewpoint of handicap theory (Zahavi, 1993),
which has led many recent writers to highlight this apparent discrepancy (e.g., (Zahavi, 1993;
Dessalles, 1998; Knight, 1998; Power, 1998). Of course the discrepancy is only troubling to the extent
that the handicap principle itself is true. The main point of the review above is that communication
systems evolving among kin do not have to obey the strictures of handicap theory (whether other
systems do is a separate question). While it is true, as mentioned above, that cheap honesty can evolve
as a result of other constraints, it seems highly unlikely that receiver error or physical constraints (e.g.
on the acoustic cues to body size) could account for the capacity of language to be both highly accurate
and open-ended. Humans are not restricted to discussions of relative body size. Thus it seems worth
considering the thesis of this paper: that human language evolved in a context of communication among
kin.
As many commentators on the problem of language evolution have remarked, language is so different
from the communication systems of other animals that the very comparison may seem strained.
However, it is equally clear that there are fundamental biological similarities between humans and
animals, both in terms of neural functions (there are no new neurotransmitters, or new types of neuron in
humans) and genetics (virtually all of our genes are shared with mice, and the sequence similarity of
these genes is probably around 98% between chimps and humans). Further, many aspects of human
language are built on a foundation shared with other animals (these include, uncontroversially, most
aspects of the vocal production and hearing apparatus, the system of long-term memory that must
underlie the lexicon, and rather complex conceptual structures including memory of places, events and
individuals). An important issue to be faced by any theory of language evolution is thus how to
conceptualize the undeniable differences between language and other communication systems in a
manner that neither trivializes the differences nor neglects the similarities, both of which are
important. To my mind, a natural framework within which to understand language evolution is
comparative, that is, in comparison to the many animal communication systems that are now
reasonably well-understood. A comparison of language and animal communication systems allows us to
identify important differences along with key similarities, both homologies (probably present in our
prelinguistic ancestors) and analogies (repeatedly evolved solutions to some common problem) (Hauser,
1996). This approach highlights difficult evolutionary problems that were solved, somehow, en route
to modern language. These are, in no particular order:
1. Cheap Honesty: an ability and propensity to communicate rich and accurate information, at low cost;
2. Vocal Imitation: an ability to learn and reproduce arbitrary acoustic signals; and
3. Complexity (Discrete infinity): an ability to generate an open-ended system of words & sentences.

---

While the combination of these capabilities appears unique to humans, each of these three basic
capacities has at least some parallels in the animal kingdom. While there are many examples of
either information-poor systems (e.g. most bird songs appear to have little meaning beyond "I'm a male
of species X, other males stay away, females approach") or actively deceitful systems (some bird alarm
calls are used deceptively as frequently as 60% (Møller, 1990; Møller, 1988; Munn, 1986)), there are also
"honest" systems in nature that convey accurate information (e.g. mammal alarm calls) including
information about events that are not perceptually present (e.g. honeybee dance). Vocal imitation is
also well-developed in other species, though not apparently in other primates [Janik, 1997 #2591;
Fitch, 2000 #2659 - most birds have a capacity for vocal learning and some, such as mockingbirds, have
a rich capacity to imitate not only bird calls but also many environmental sounds.
Finally the ability to generate an open-ended and potentially infinite variety of words (phonology)
and sentences (syntax) by recombining smaller units, seen in all human languages, has often seemed
qualitatively different from the capacities of any other animal. At two different levels, modern
human languages can generate infinitely diverse and complex structures: at the level of phonology
(which generates new words from meaningless phonemes) and of syntax (generating new phrases or
sentences). This abstract property follows from the manner in which language generate their structures
by recombination of a finite set of primitives (phonemes/syllables in phonology, or words in syntax).
Any word can be extended by adding various affixes (as in adding "non-" to
"disestablishmentarianism") and any sentence by adding new phrases (as in adding "John believes
that" to "Mary will exceed the expectations of the committee"). Although these generative systems
have important differences (e.g., most syntactic generation is done on line, producing novel utterances,
while the products of phonological and morphological generation are typically memorized in a stored
lexicon), the critical factor in both systems is flexible, open-ended generation of novelty by
recombination of discrete elements (Studdert-Kennedy, 1998; Nowak et al., 1999).
This concept of discrete infinity has occasionally been criticized because, in reality, we produce neither
infinitely long words nor infinitely long sentences. These limitations appear to be a matter of
implementation limitations (of memory, time, breath or whatever) rather than some intrinsic
limitation of the principles of phonology or syntax per se. I agree that such limitations play an
important role in the neural implementation and evolution of language, and do not advocate ignoring
them. However, I don't believe that these facts justify neglecting the basic productivity of language
which is so central to its functioning, and so different from most other communication systems. To avoid
needless argument, I will refer to language as being "highly generative" rather that "infinitely
generative", because the key consideration is the vast difference between human language and other
communication systems, rather than infinity per se. The songs of birds or whales use recombination of
basic units to form larger, more complex units, and there are no obvious limits on the variety of the units
thus formed. Although these larger units appear to be ends in themselves, rather than a vehicle for
transmitting detailed messages, they are richly generative nonetheless (a similar point could be made

---