‘What is human?’ as a question is something that has eternally fascinated thinkers and writers. Put less poetically, what we are looking for is some way of defining what that sense of humanity’s uniqueness that we all feel actually means: a specification of what it is about us that makes us special.
Until recently, the religious answer, that we were special because we were made in God’s image, was considered adequate, but it could not survive the twin hammer-blows of the Darwinian revolution and increasing knowledge of just how sophisticated animal behaviour can be. In the twentieth century it became briefly popular to define humanity in purely biological terms, but the consequences of that trend were so terrible that such a definition is no longer admissible (and in addition, as I shall show below, it would also be entirely meaningless).
And so we are left with a tricky question. As we learn more about natural animal behaviour, one by one the old uniquely defining characteristics of humanity turn out to be shared with the most unlikely species.
Writers have contributed to this discussion with suggestions that humanity is a property of more than just homo sapiens sapiens. Philip K Dick, in his story ‘Human Is’, concludes that an alien is more human than the man it is masquerading as, because it, unlike him, is capable of displaying caritas. Likewise, Stanislaw Lem, in his satirical story ‘The Great Washing Machine Tragedy’, elegantly demonstrates how it is completely impossible to find a definition of humanity that would not extend to a self-aware robot. Similarly a recurring theme in ‘Star Trek’ has been the question of where the line is to be drawn between (say) human and machine.
It should not surprise readers that I will come down firmly in the Dick – Lem camp, and not with those frightened individuals who seek to maintain the myth that there is something about homo sapiens sapiens that is uniquely special.
The goal is to find if there is something about homo sapiens sapiens that makes us unique. Something that sets us apart from all other species, confirming the comforting idea that we humans are hot stuff, and are somehow different from other animals. If it turns out that there is no such criterion, the secondary goal is to find some criterion that is usable (in the sense that we can actually apply it) and whose consequences, though they may involve defining humanity rather more widely than we would have expected, are not obviously stupid.
There are basically two approaches to definition: either it is something about what we are, or something about what we do. This can be rephrased as saying that humanity is either biologically or behaviourally determined. We now examine those two approaches.
A biological definition is somehow dissatisfactory, as finding our uniquely human attribute in some strand of DNA as opposed to some aspect of our behaviour does seem to rather miss the point. However, there is a more insidious problem than just being a bit boring. If humanity is defined as ‘is a member of homo sapiens sapiens’ then this is formally identical to the criterion ‘is a member of sturnus vulgaris’, which defines the common starling. And while it is clearly true that each species is unique in the sense that it is different to all other species, this is at best a triusm and goes no way at all towards finding a way to mark out one species (ours) as being unique in the stronger sense set out above. And so a purely biological definition cannot be used to set us apart from all other species.
The problem of finding a clear criterion
Therefore we have to look to behaviour. What behaviour or groups of behaviours are characteristic of homo sapiens sapiens but of no other species? This question proves to be singularly hard to answer.
Back when things were simple, humanity was distinguished by any number of criteria. So there was man the toolmaker, the problem solver, the long-term planner, the self-aware, the possessor of culture, and so on and so forth.
And then Jane Goodall went to Tanzania and saw chimpanzees using tools. Since which evidence of tool use among non-human animals has blossomed, so now even some birds are known to deliberately make and then use tools. As for being a problem solver, there is now very strong evidence for problem solving in apes, elephants, cetacea and birds. Most embarrassingly of all it turns out that portia fimbriata (a spider) is capable of solving complex problem and modifying its behaviour to deal with new challenges. So some of the feats we are most proud of can be replicated by an animal that doesn’t even have a brain (in our sense).
What used to be considered a defining characteristic of humanity is a theory of self, that is to say that humans are capable of ascribing identity and self-awareness to others than themselves, and of differentiating the world as they know it from the world as others know it. Well, now it has been demonstrated that so do chimpanzees, ceteaca and elephants.
Finally culture. One key aspect of culture is individuals voluntarily co-operating for the greater good to achieve something they could not do on their own, and sharing the reward. That was meant to be uniquely human. Chimpanzees do it. Dolphins do it (indeed, some of them go one stage further and enter into partnerships with humans). So do some spiders. So, on to the next aspect: altruism. Surely that is uniquely human? Well no: it is well attested in elephants and apes, and there are well documented cases of cetacea taking action to protect humans from predators. And finally, the vertical transmission of skills and knowledge down generations is clearly demonstrated in cetacea, where hunting skills tend to be possessed by specific groups, and we even see clear evidence of adults teaching young. Indeed, some whales even have vocal ‘dialects’ that can be used to distinguish local groups from one another.
So, basically, it is annoyingly hard to find any behavioural characteristic that we proudly think distinguishes us from the animals that is not, in fact, shared by some other, often embarrassingly simple, animal. We have, it appears, only one card left in our hands.
That is to say, language. It seems, through all the blizzard of evidence of animal capability, that our linguistic capabilities remain a constant. In fact the situation may not be as clear as all that, but this is our last best option, so let us examine it.
First, a definition of language is in order, because if we are too loose we run the risk of being able to say that (say) cats have language, while if we are too rigorous we might rule ourselves out. So I shall define a language to be a system of sentences which can be generated from a finite set of words using a finite set of grammatical rules, such that I can form sentences of arbitrary complexity.
Let’s explore this a bit. First notice that I have nothing to say about the semantics of language, that is to say whether the sentences I have constructed mean anything. A good reason for this is that it is useful if our definition can include languages that we cannot translate. That is to say, if I come across an unknown language, I want to be able to say whether it is or is not language without having to wait until it becomes translatable, and that means that the definition can involve only syntax.
The only other aspect of the definition that may cause surprise is the stipulation that I can form sentences of arbitrary complexity. This is, in fact, one of the key defining features of human language. For example, what differentiates our language from, say, the vocal signals of a prairie dog, is that we are not limited to a fixed expressive range consisting of distinct units. In spite of, like the prairie dog, having a finite number of basic units (in our case, words) we can combine them to form new expressive units, to the extent that it is often said (with only mild exaggeration) that every human utterance creates a new sentence, never before spoken.
So, the proposed criterion for humanity is this: if we define language in this way, then only humans are possessed of language. Clearly every known human natural language complies with the definition, so the question now is, does it work to rule out animal communication systems?
Are we really alone?
So, does the criterion work? At first sight it seems promising. Many animals have communication systems based on sound, light, scent or posture, but in nearly all cases these are collections of simple gestural signals along the lines of warnings, threats and invitations. None of these constitutes a language.
The only cases that need special attention are elephants, chimpanzees and cetacea. Elephants can be dealt with very quickly: so little is known about their communications (which have only recently been discovered) that we simply cannot say whether they form a language or not.
Chimpanzees have a rich vocabulary of vocal and gestural signals, there is no evidence of any grammar, and it does seem that the signals are discrete units in themselves, rather than being part of a greater whole. Moreover, experiments in which efforts were made to teach basic language to chimpanzees (and other apes) have always foundered on the problem of grammar. It seems that our cousins can handle basic subject-verb composition and simple predication of a property of a thing, but cannot comprehend the recursive structures that allow human languages to build sentences of arbitrary complexity. So they have a concept of grammar, but it is too limited to generate more than a finite number of simple sentences.
Problems develop when we look at cetacea, and here I will concentrate mostly on one species: -the bottle-nosed dolphin – largely because it has been at the centre of most language-related research. The starting point for this discussion has to be the extraordinary work done in Hawaii, where experimenters developed a genuine language, with a limited vocabulary and a grammar capable of generating sentences of arbitrary complexity, and taught it to dolphins. The dolphins understood complete sentences, and were able to carry out specified tasks, and even answer questions. So clearly the dolphin is capable of language.
Now the question is, do dolphins use that capability in the wild? Here there are no clear conclusions, but there are some suggestive facts. The prime point I will make is that dolphin ‘chatter’ is immensely complex, but in a special way. It is neither a simple collection of calls, but nor is it the ever-varying but ultimately stereotypical song of hump-backed whales. Rather, there are distinct elements that can be recognised and that recur, but the context and pattern of those elements is always changing, so, in essence, all utterances are unique. This is, as we noted above, a crucial feature of human language, and is highly suggestive of the hypothesis that these dolphins do, in fact, have language according to our definition.
For various reasons, researchers seem to fall over in their haste to claim that dolphins do not have language. This often reads most peculiarly, as papers in which clear evidence of linguistic capability is presented then end with a ritual, and unsubstantiated, assertion to the effect that language remains man’s alone. There are, I believe, two reasons for this: one good, one bad.
The good reason is simple enough. In the 1960s (when else?) there was a regrettable phase when some rather over-excited researchers made wild claims for dolphin intelligence and linguistic capabilities, including even the assertion that they could understand human speech! This was, of course, debunked, but it is still the case that the ‘Flipper’ version of dolphins holds sway in the public consciousness. Therefore, ever since, researchers have been eager to not make overly excited claims. As a result, they seem to often err too far in the other direction, and to try to find ways of explaining away clearly intelligent behaviour as simply reflexive activity or mimicry.
Thus when one population started a strange fashion of wearing a sponge on their rostrum, researchers, desperate to avoid admitting that dolphins might have the concept of fad or fashion, came up with somewhat wild-eyed theories about how this might give some advantage in hunting, even though no evidence for this hypothesis has been forthcoming. Likewise, it is apparently inadmissible to say that dolphins’ extraordinary appetite for sex can arise because they enjoy it – clearly recreational sex is a human-only thing. In dolphins it is all about expressions of dominance.
This leads on to the bad reason, which is, I believe, fear. Basically, we want to be uniquely special, set apart from all animals, to find a scientific justification for holding dominion over them. And yet, one by one, we have seen the differences between us and animals vanish. Language and culture (and fashion and recreational sex are cultural activities) are the last bastions of our human uniqueness. If we admit, as a cold-blooded look at the evidence suggests we should, that even these characteristics are not unique to homo sapiens sapiens, what then?
Towards a definition
In spite of these concerns, it is clear that language is the best bet for a definition of humanity. I’ll now examine some language-based definitions. First one that successfully isolates homo sapiens sapiens, but which is philosophically untenable, then one that goes too far in the other direction, and has the problem of being utterly unusable, and finally an attempt at a synthesis of the two, which concludes with a definition of humanity that is wider than just homo sapiens but which has the advantage of being both clear and easy to apply.
The first definition is simplicity itself: we define humanity as the community of beings which possess language and with which we are able to communicate using natural language. So we are now, for the first time, introducing semantics into the discussion: it is not enough just to possess language, we must have shared semantics.
So, can we go home now? No. This definition has some interesting problems. At the moment we cannot converse (much) with bottle-nosed dolphins, though it is reasonably clear that they have some form of language. So, for now, they are not human. Now suppose some Champollion of the future deciphers delphine; at that point the dolphins abruptly become human. That is somewhat unexpected: one would expect the scope of what is considered to be human to remain reasonably constant over time.
Things get worse. The definition shares the fundamental weakness of the biological definition, because it basically says that a being is human if it talks like homo sapiens sapiens, so instead of the definition telling us that we are human, we tell it that we are human and then use ourselves as the yardstick.
Then it goes from bad to grotesque. Imagine we discover a previously uncontacted tribe, who speak a wholly new language (something that has happened often enough in the past, and may happen again). By this definition they are not human. But they are homo sapiens sapiens too, so one could equally well say that according to the definition, they are human and we are not. There is nothing in the definition to break this deadlock.
And so, this definition, though promising, breaks down because its definition relies on just too many imprecise notions.
So, let us go to the and propose the following: the criterion for humanity is linguistic capability. So humans include homo sapiens sapiens, dolphins and (for all we know) aliens from planet Zog. This is perhaps a broader definition than we would like, but it is at least nicely unambiguous.
Or is it? I now need to introduce the concept of an epistemic barrier, of which I have written at length elsewhere. In brief, an epistemic barrier is an insurmountable barrier to communication or comprehension, so while (on current evidence) we can translate between all human languages, the hypothetical alien from planet Zog’s language (if it were on the other side to a barrier) would be incomprehensible to us. And this is not just a matter of our not having tried hard enough or of not having found the trick. The barrier is systemic: we cannot ever understand the language, because to do so would require us to be the alien and not ourselves.
One of the key features of an epistemic barrier is this. Languages that are not separated from us by a barrier can be identified as languages and (potentially) translated. However, for languages on the other side of a barrier, not only can we not translate them, we cannot recognise them: they are essentially indistinguishable from noise.
This is the sentence of death for this definition. If we cannot identify these separated languages, then we cannot tell who is language capable and who is not. Therefore the definition is unusable, and so valueless.
The via media
There is a potential compromise definition which comes from the flip side of the point about epistemic barriers: any language that is not separated from me by an epistemic barrier should be identifiable as a language, by means of detection of a functional grammar, even if we cannot translate it. This is why I made my definition of language above purely in terms of syntax.
So define humanity as the property of being capable of using a language from which we are not separated by an epistemic barrier. Consider the problems with the earlier definitions. Defining something as human depends on linguistic capability, not on understanding, and so we don’t get the problem of species suddenly becoming human. Also we don’t have the undiscovered tribe problem: we may not speak their language (yet), but we know that it is a language. Finally, the solution to the epistemic barrier problem is built in to the definition. A being, like the alien from planet Zog, that speaks a language on the other side of a barrier is inhuman, which is as it should be.
The only possible problem is that there still is a tacit assumption that our bubble of languages is somehow special, but we can even remove that by adapting a two-tier classification. Thus all language-capable beings are sapient, and they are divided (by epistemic barriers) into linguistic groups, one of which happens to be the one we presumptuously call the ‘human’ group, though it could equally well be called ‘delphine’, or (better) more neutrally, ‘Terran’.
So we have a well-formed definition. It includes us and rules out the alien from planet Zog, which is good. Rather unnervingly, evidence suggests that, though it rules out other apes, it does not rule out some dolphins, but if that is the price for having a sound definition, so be it. It seems this our days of sapient uniqueness are over.
Appendix: the definition of ‘language’
For those who are interested, I present here a rigorous definition of ‘language’ in place of the rather vague definition used in the main body of the essay. Note that (a) this is not necessary in order to gain an understanding of the main argument, and (b) the definition requires some mathematical sophistication (specifically, comfort with the notion of infinite cardinals). Therefore, if you are happy to take on trust that the definition can be made rigorous, should cease reading now.
Definition of a language
Recall that I informally defined a language as a formal system which has a finite vocabulary, a finitely expressed grammar and which is capable of expressing sentences of arbitrary complexity. The first two parts of this definition can be made rigorous, but for the third we need either a way of defining complexity, or an alternative criterion. It is this latter approach that I take here.
So, here is the definition:
A language is a formal system consisting of:
- A finite vocabulary of symbols
- A grammar consisting of finitely specifiable collection of rules for building statements from the symbols
such that the class of statements constructed from the vocabulary with the grammar is not denumerable.
In the case of a natural language the symbols are words, the grammar is the rules defining the syntax of the language, and the statements are sentences. At this point note the following fact: this definition of a language is solely concerned with syntax. Statements in such a language will be syntactically correct, but need not be semantically correct. For our purposes in this essay, this will suffice.
Consider the definition in light of what we expect of a natural language. It has three parts:
A finite vocabulary: natural languages have a finite number of words. A language with an infinite number of words could not be comprehended, so we would end up using a finite subset of the vocabulary, hence reducing to the finite case.
A finitely specifiable grammar: the rules determining the grammar of the language can be written down as a finite statement in some metalanguage (we will generally use a mix of formal logic and English). Again, a language whose grammar is not finitely specifiable cannot be wholly comprehensible, and, indeed, cannot have arisen naturally, for how could the rules have ever been defined? And so, again, we would end up using only a finitely specifiable fragment, and so we are back in the finite case.
Uncountably many statements: from these finite resources I can build not just infinitely many, but uncountably many statements. This is not immediately obvious. I therefore present below a general argument relating denumerability to bounds on the complexity of statements in the language. After that I give three examples, working up from set theory, through first-order logic, to natural language, showing that in each case the language in question complies with my definition.
Complexity = non-denumerability
The purpose of this argument is to motivate the relationship between my original requirement of being able to express statements of arbitrary complexity, and the new requirement of the language being non-denumerable. For the purposes of this discussion I shall consider the complexity of a statement to be the number of parameters required to specify it uniquely within the language.
Suppose that I have a language in which the class of statements is denumerable. A simple example is the language with a vocabulary of one symbol (say ‘*’) and the simple rule that ‘*’ may or may not be followed by another ‘*’, so the statements are strings of ‘*’ of arbitrary length, and can therefore be encoded as integers. As every statement in the language corresponds with a single integer, all of the statements have the same complexity, and so the language cannot express statements of arbitrary complexity.
Now consider the language which has two symbols ‘*‘ and ‘_‘ and the rules that the first symbol of a statement must be ‘*’, that ‘*‘ may be followed by ‘*’, ‘_‘ or nothing and that ‘_‘ must be followed by ‘*’. Then statements consist of strings of ‘*’ separated by underscores, and so correspond to arbitrary lists of integers. The class of such statements has cardinality equal to the continuum (as the set is a strict superset of the power-set of the integers). This language can express statements of arbitrary complexity, because I can impose no upper limit on the length of the list of integers required to specify a statement.
Generalising to arbitrary non-denumerable languages: if such a language has an upper bound on the complexity of its statements, then its statements can be placed in a one-to-one correspondence with n-ples of integers for n < N for some fixed N. This is impossible, therefore the complexity of statements in the language is unbounded.
Therefore the requirement for a language to be able to express statements of arbitrary complexity is equivalent to non-denumerability. As non-denumerability is a precise criterion (unlike the rather vague ‘complexity’) it is to be preferred.
Example: set theory
In any standard version of set theory I can start from a vocabulary consisting of the empty set and some connectives, and, with the aid of a finitely specifiable number of axioms, build a language whose statements are the proper class of sets.
It can be shown that the statements include a countably infinite set. But now I can construct the power-set of that set, and this is non-denumerable. But every element of the power-set is itself a set, and so is a statement in the language. Therefore the language is non-denumerable.
Example: first-order logic
Now consider first-order logic, in which I have a vocabulary consisting of a finite number of predicates, a finite number of unbound variable markers, standard connectives (quantifiers, logical operations, etc) and brackets. Moreover, the standard axioms for first-order logic are finitely specifiable.
Here the key observation is the fact that I can build new predicates from old. For example, suppose that my language has one predicate x → Px. I can immediately build predicates
(x1,…,xn) → Px1 & … & Pxn
for all n > 0, and so the language contains a countable infinity of predicates. I construct statements by selecting collections of predicates and binding their variables using the connectives. Therefore the cardinality of the class of statements is at least that of the power set of the set of predicates, which is the continuum. Therefore the class of sentences is non-denumerable.
So, first-order logic constitutes a language according to the definition, which is exactly what we would expect, given that it abstracts key features of natural language.
Observe that in first-order logic I can use it to build predicates of arbitrary complexity. Given a one-variable predicate, I can construct a more complex one-variable predicates from it, e.g.
x → Px ⇒ x → P’x ≡ (∀y) (Qxy ⊃ Py)
(where Q is some two-variable predicate). Now take the definition of P‘ and write P‘ in place of P, to get a new predicate
x → P’’x ≡ (∀y) (Qxy ⊃ P’y) = (∀y) (Qxy ⊃ (∀z)(Qyz ⊃ Pz))
I can repeat this process indefinitely, creating an infinite tower of predicates of arbitrarily increasing complexity. Therefore, given any sentence containing the predicate Px, I can substitute for x → Px a predicate of arbitrary complexity to obtain a sentence of arbitrary complexity.
Example: natural language
Natural languages have finite (though very large) vocabularies, and a finitely expressible grammar. A critical defining feature of all known natural languages is the property known as embedding. This is the linguistic version of the ability to substitute predicates of arbitrary complexity for simple predicates in first-order logic.
A clause embeds in a sentence if it can be substituted for an individual word in that sentence in a syntactically correct way. So, given any sentence in a natural language we can replace any adverb with an adverbial subordinate clause, e.g:
he ate quickly ⇒ he ate as though he feared the food may be snatched from him
And similarly for nouns, verbs and adjectives.
As with first-order logic, by using repeated embedding I can form an infinite hierarchy of arbitrarily complex clauses. But now, a sentence is a collection of clauses joined together with connectives. Therefore the class of statements has at least the cardinality of the power set of the integers, and so is not denumerable. Therefore the definition is effective.