As a postgraduate in Edinburgh in the late 1990s I wound getting engaged in the question of the evolutionary origins of language. At first I thought this was a bit of baroque fun, but then I realised it was a suitable core question for anyone seriously interested in cognitive science or cognitive technology. Because there is little question that language benefits cognition and therefore computation, both by facilitating reasoning through more efficient representations, and by allowing the transmission of solutions or "good tricks" so that one mind can build on the progress of another.
The problem with language then is that only one species has it, which means one of two questions needs to be answered:
A surprising number of researchers assumed that language (or shared cognition) is not useful, under the false belief that evolution, being based on "the survival of the fittest" cannot explain altruism. In fact, cooperative behaviour is ubiquitous in nature, and has been increasingly well understood in an evolutionary framework since the 1960s. The key thing to understand is that there are no single-gene organisms. One gene cannot reproduce, and reproduction is key to life. But organisms share genes, we share something like 87% of our genes with fruit flies, because the hard problem of encoding life and getting it working really has turned out to be all that cellular and genetic reproduction stuff.
To the extent that we share genes with others, we are related to them. And this is (to a first approximation) how cooperation evolves. It makes sense for you to invest a cost if that cost provides a benefit to others great enough that it compensates for your cost. Well, assuming the others have exactly the same genes that you have, because the genes are what are encoding your behaviour and get this kind of thing going. But what if you only share some of your genes? Then the benefit has to be greater than the cost, to the proportion of that relatedness. However, notice that you might benefit a bunch of others with a single act, in that case you can multiply that number in too.
So I will answer question 1, not question 2, because I think shared computation / cognition is broadly useful. Why don't other species have language then? Well, remember that I said cognition was broadly useful. Cognition is a risky strategy for choosing behaviour, it involves trying new things and making mistakes. There are a lot of tradeoffs to what extent a species relies on cognition rather than genetically encoded, predefined behaviour. For one, thinking seems to take time, so it is mostly only a strategy in relatively long-lived species, with relatively long childhoods.
Second, remember that I talked about the problems of encoding the instructions for life. Evolution is an unsupervised algorithm, a bunch of hacks tripped over by nature. If a really good solution is found, it can be lost in the next generation by the same hacks that found it. So genomes that reliably produce good behaviour require massively redundant encodings of the good tricks. Basically, they require differentiated representations, where some parts of the genome are more or less likely to be allowed by the mechanisms of reproduction to improvise.
What scientists are increasingly agreeing is that humans represent something called a "dual replicator system". That is, not only do we evolve as organisms, but our culture – the information we transmit between each other that also influences our behaviour – also evolves. These two learning systems influence each other.
Having said all this, my theory is almost boringly simple. Our one special thing comes from being the only existing species that have two relatively less special things at the same time. Very few species of animals live long enough and remember enough to be able to "record" (remember) the huge amount of information required for a second replicator system. Many of these species are apes. Of the apes, we are the only ones who have also evolved vocal imitation, though many other species that are not apes have evolved that too. We know we are basically chimpanzees. I'm convinced by the theory that we were chimpanzees that moved out into the savannah, where there was a lot less food than the jungle. Because there was less food, we needed larger territories, and because we needed larger physical territories, we needed two parents to provide for our children, not one like most mammals. Though notice birds also have this problem, because they have high metabolisms and can't carry much. There are also tiny South American primates called callitrichids, like marmosets and tamarins, that also live like birds. Both birds and callitrichids often have two parents and some older children help take care of younger children. Like people do.
Somewhere while we were organising ourselves out on the savannah, vocal imitation got stumbled upon by evolution and found to be useful. Maybe it was just for territorial calls, like it is for birds. But once we had that incredibly rich representation, with volume, precise times, sequences of pitch and formants, we had enough information getting transmitted around that evolution took off again, and a second replicator system was born.
This is not to say that vocal imitation is the only thing that makes humans special. Of course not. Neither is language. But language facilitates all the other amazing things, the highly plastic agriculture, the vast assortment of tools. Living in a context of more information sharing afforded a greater probability of selection for social and cooperative strategies and the many traits that support them, like advanced theory of mind and meta cognition.
Most of this has been written up and published a few times. Like I said, I've been working on the problem since about 1998. Probably it makes most sense to work backwards through the papers that talk about this if you want to read more than what is above:
Update from August 2016: We now have empirical evidence that I'm right about some of this, see my recent blogpost: Semantics derived automatically from language corpora necessarily contain human biases
The problem with language then is that only one species has it, which means one of two questions needs to be answered:
- If it's useful, why don't other species have it?
- If it's not useful, why does ours?
A surprising number of researchers assumed that language (or shared cognition) is not useful, under the false belief that evolution, being based on "the survival of the fittest" cannot explain altruism. In fact, cooperative behaviour is ubiquitous in nature, and has been increasingly well understood in an evolutionary framework since the 1960s. The key thing to understand is that there are no single-gene organisms. One gene cannot reproduce, and reproduction is key to life. But organisms share genes, we share something like 87% of our genes with fruit flies, because the hard problem of encoding life and getting it working really has turned out to be all that cellular and genetic reproduction stuff.
To the extent that we share genes with others, we are related to them. And this is (to a first approximation) how cooperation evolves. It makes sense for you to invest a cost if that cost provides a benefit to others great enough that it compensates for your cost. Well, assuming the others have exactly the same genes that you have, because the genes are what are encoding your behaviour and get this kind of thing going. But what if you only share some of your genes? Then the benefit has to be greater than the cost, to the proportion of that relatedness. However, notice that you might benefit a bunch of others with a single act, in that case you can multiply that number in too.
So I will answer question 1, not question 2, because I think shared computation / cognition is broadly useful. Why don't other species have language then? Well, remember that I said cognition was broadly useful. Cognition is a risky strategy for choosing behaviour, it involves trying new things and making mistakes. There are a lot of tradeoffs to what extent a species relies on cognition rather than genetically encoded, predefined behaviour. For one, thinking seems to take time, so it is mostly only a strategy in relatively long-lived species, with relatively long childhoods.
Second, remember that I talked about the problems of encoding the instructions for life. Evolution is an unsupervised algorithm, a bunch of hacks tripped over by nature. If a really good solution is found, it can be lost in the next generation by the same hacks that found it. So genomes that reliably produce good behaviour require massively redundant encodings of the good tricks. Basically, they require differentiated representations, where some parts of the genome are more or less likely to be allowed by the mechanisms of reproduction to improvise.
What scientists are increasingly agreeing is that humans represent something called a "dual replicator system". That is, not only do we evolve as organisms, but our culture – the information we transmit between each other that also influences our behaviour – also evolves. These two learning systems influence each other.
Having said all this, my theory is almost boringly simple. Our one special thing comes from being the only existing species that have two relatively less special things at the same time. Very few species of animals live long enough and remember enough to be able to "record" (remember) the huge amount of information required for a second replicator system. Many of these species are apes. Of the apes, we are the only ones who have also evolved vocal imitation, though many other species that are not apes have evolved that too. We know we are basically chimpanzees. I'm convinced by the theory that we were chimpanzees that moved out into the savannah, where there was a lot less food than the jungle. Because there was less food, we needed larger territories, and because we needed larger physical territories, we needed two parents to provide for our children, not one like most mammals. Though notice birds also have this problem, because they have high metabolisms and can't carry much. There are also tiny South American primates called callitrichids, like marmosets and tamarins, that also live like birds. Both birds and callitrichids often have two parents and some older children help take care of younger children. Like people do.
Somewhere while we were organising ourselves out on the savannah, vocal imitation got stumbled upon by evolution and found to be useful. Maybe it was just for territorial calls, like it is for birds. But once we had that incredibly rich representation, with volume, precise times, sequences of pitch and formants, we had enough information getting transmitted around that evolution took off again, and a second replicator system was born.
This is not to say that vocal imitation is the only thing that makes humans special. Of course not. Neither is language. But language facilitates all the other amazing things, the highly plastic agriculture, the vast assortment of tools. Living in a context of more information sharing afforded a greater probability of selection for social and cooperative strategies and the many traits that support them, like advanced theory of mind and meta cognition.
Most of this has been written up and published a few times. Like I said, I've been working on the problem since about 1998. Probably it makes most sense to work backwards through the papers that talk about this if you want to read more than what is above:
- Joanna J. Bryson, ``Cultural Ratcheting Results Primarily from Semantic Compression''. The Proceedings of Evolution of Language 2010, Smith, Schouwstra, de Boer & Smith (eds.) pp. 50-57.
- Joanna J. Bryson ``Representations Underlying Social Learning and Cultural Evolution'', Interaction Studies, 10(1):77-100, March 2009.
- Joanna J. Bryson ``Embodiment vs. Memetics'', Mind & Society, 7(1):77-94, June 2008.
Update from August 2016: We now have empirical evidence that I'm right about some of this, see my recent blogpost: Semantics derived automatically from language corpora necessarily contain human biases
Comments