Language:
A new kind of evolutionary system
| Simon Kirby | |
Language Evolution and Computation Research Unit Theoretical and Applied Linguistics, Edinburgh |
| Eight key transition events in the history of life on Earth | |
| (Maynard Smith & Szathmáry 1995): | |
| Replicating molecules → Populations of molecules | |
| Independent replicators → Chromosomes | |
| RNA → DNA | |
| Prokaryotes → Eukaryotes | |
| Asexual clones → Sexual populations | |
| Protists → Animals, plants and fungi | |
| Solitary individuals → Colonies | |
| Primate societies → Human societies (Language) | |
The growth of evolutionary linguistics
| Not just the biologists who are interested in language evolution… | ||
| Pinker & Bloom (1990): Landmark paper linking linguistic and evolutionary theory. | ||
| Decade before: 96 articles | ||
| Decade after: 1095 articles | ||
| Evolutionary thinking in mainstream linguistics. | ||
| Jakendoff (2002), Foundations of Language: Brain, Meaning, Grammar, Evolution | ||
| Linguistic theory in the “big” science journals: Nature, Science, etc. | ||
| Computer science, robotics and AI communities actively researching language evolution. | ||
| Survey two approaches to the why question: | ||
| Standard nativism | ||
| Evolutionary nativism | ||
| Introduce a new approach: the Iterated Learning Model | ||
| Present two simulations of the ILM | ||
| Argue that language itself is an evolutionary system | ||
| Language is unique because our brains are unique. | |
| We (and no other species) are born with a specialised innate cognitive mechanism for learning language. | |
| Language is the way it is because our biology constrains it to be that way. | |
| It’s the job of linguistics to deduce the structure of the language acquisition device: Universal Grammar. | |
| (And that’s it!) |
| Why is the Language Acquisition Device the way it is? | |
| The evolutionary psychology position: (Pinker & Bloom 1990) | |
| Language is the way it is because natural selection favoured individuals who were able to learn languages that were useful for communication. | |
| Some problems with the evolutionary nativism position: | ||
| Doesn’t really explain why language is unique | ||
| Not specific about how natural selection is going to work | ||
| Is everything in language optimally tailored? | ||
| Are there alternative mechanisms? | ||
| A new hypothesis: | ||
Where does the primary linguistic data come from?
The Iterated Learning
Model
(a framework for computational simulation)
What will the agents talk about?
| Need some simple but structured “world”. | |
| Simple predicate logic: | |
| Agents can string random characters together to form utterances. | |
| Learners try and form a grammar that is consistent with the primary linguistic data they hear. | ||
| Fundamental principle: learning is compression. | ||
| Two processes on hearing a meaning-signal pair: | ||
| A rule is added to the grammar that is specific to that pair. | ||
| Search for ways of “compressing” the grammar. Are there pairs of rules that can be subsumed under a single rule? Are there duplicate rules? | ||
| Compression process uncovers any generalisations in the data. | ||
| Start with one learner and one adult speaker neither of which have grammars. | |
| Choose a meaning at random. | |
| Get speaker to produce signal for that meaning (may need to “invent” random string). | |
| Give meaning-signal pair to learner. | |
| Repeat 2-4 one hundred and fifty times. | |
| Delete speaker. | |
| Make learner be the new speaker. | |
| Introduce a new learner (with no initial grammar) | |
| Repeat 2-8 thousands of times. |
| Initially, speakers have no language, so “invent” random strings of characters. | |
| A protolanguage emerges for some meanings, but no structure. These are holistic expressions: | |
| ldg “Mary admires John” | |
| xkq “Mary loves John” | |
| gj “Mary admires Gavin” | |
| axk “John admires Gavin” | |
| gb “John knows that Mary knows that John admires Gavin” | |
Results 1b: many generations later…
| gj h f
tej m John Mary admires “Mary admires John” |
|
| gj h f
tej wp John Mary loves “Mary loves John” |
|
| gj qp f
tej m Gavin Mary admires “Mary admires Gavin” |
|
| gj qp f h m Gavin John admires “John admires Gavin” |
|
| i h u i
tej u gj
qp f h m John knows Mary knows Gavin John admires “John knows that Mary knows that John admires Gavin” |
Quantitative results: languages evolve
| There is no biological evolution in the ILM. | |
| There isn’t even any communication; no notion of function in model at all. | |
| So, why are structured languages evolving? | |
| Hypothesis: | |
| Languages themselves are evolving to the conditions of the ILM in order that they are learnable. | |
| Only rules that are generalisable from limited exposure are stable. | |
| The poverty of the stimulus ensures that holistic expressions cannot survive. |
| Languages are not completely regular. | |
| Languages are not completely stable. | |
| In the previous simulation, languages evolve to a completely regular fixed-point. | |
| Why would holistic expressions survive? | |
| Why do expressions change? | |
| Completely accurate transmission of signal implausible. Might this lead to fixed end points? | ||
| Include a least effort principle: | ||
| Speakers always use the shortest string possible for a given meaning. | ||
| Speakers occasionally drop letters in production. | ||
| Simplified meaning-space: 5x5
“paradigm” i.e., each meaning is a coordinate in 5x5 space |
||
Results 2a: early protolanguage stage
Results 2b: later stages, regulars
Results 2c: occasional, but short-lived irregulars
| Top ten verbs of English by frequency: | ||
| be, have, do, say, make, go, take, come, see, get… | ||
| was, had, did, said, made, went, took, came, saw, got… | ||
| Add frequency biases in meaning space. (modelled on Zipf’s law) | ||
| Meanings in top left of table get spoken more often | ||
Results 3: frequent=irregular infrequent=regular
Language is an adaptive system in its own right
| (At least) two adaptive problems for language: | ||
| Must be learnable even under “poverty of the stimulus” conditions | ||
| Must be produced by speakers employing least-effort principles | ||
| The solution is a language that is compositional where it matters (where learning data is likely to be sparse), and short where it matters (where utterances need to be produced frequently). | ||
| Specific language universals: | ||
| Word-order universals. Why do languages tend to be consistently left- or right-branching? | ||
| Universal constraints on question formation, relative clauses, anaphora etc. | ||
| What about creolisation? | ||
| Need more sophisticated models of population dynamics. | ||
| Where do the meanings come from? | ||
| Models of meaning-formation and grounding during learning. | ||
| Why is language specific to humans? | ||
| Explore conditions for emergence of syntactic structure. | ||
| Model the biological evolution of mechanisms required for iterated learning itself. | ||
| The transition to language is the transition to a new kind of evolutionary system. | ||
| The LAD does not directly determine the structure of language. | ||
| Explains some language structure without appealing to | ||
| hard innate constraints | ||
| communicative function | ||
| The “poverty of the stimulus” is not a syntactic learnability problem. It is required for the emergence of syntax | ||
| Take home message: | ||
| Rather than looking at the way we have adapted to language, we should look more at how language adapts to us. | ||