The key structural characteristics of language are the result of cultural adaptation. There is no straightforward link between genes and language structure, linguist and computation scholar Simon Kirby told participants in the Evolang conference in Barcelona this morning (Thursday, March 13, 2008). The conference is already shaping up as a sharp quarrel between the orthodox disciples of Noam Chomsky’s nativist ideas of an innate language structure (e.g., see Stop Your Yacking) and energized heretics who explain the origins of language structure differently. Kirby’s presentation added more fuel to the insurgents’ arguments.
Kirby’s method uses models that are complex enough to track language evolution in three dimensions:
- in an individual,
- in a culture, and
- biologically.
Modeling skeptics frequently complain that for all the complexity of the interactions that models can handle, the “languages” they model are very simple. Modeling, Kirby explained, can take various forms based on computational interactions, mathematical computation, and lab experiments using human subjects. He argues that modeling produces so many surprises that an "armchair approach" can no longer serve.
Kirby reported on a forthcoming article about lab experiments. In it, one person was presented with an artificial language linking pictures with randomly created two-syllable words (e.g., kunige). The subject tries to learn the pictures and words and find ways to use those words express meanings. Next, another subject is brought in and to try and learn the “language” that the first participant had mastered. Chomsky famously argued that passing along language in this way cannot work because of the “poverty of the stimulus,” that is to say there is not enough information in the output to explain any underlying structure used to express meaning. And yet Kirby found that after ten “generations” of this passing along of output the subjects are using a language that has become structured to optimize both learnability and expressivity.
This experiment, Kirby said, shows that adaptive structure does not imply natural selection. Instead, there is a cultural co-evolution of learnable features and expressive structures. Only weak innate biases are required, and they can be weakened even further by the process of co-evolution. Kirby added that this weakening of genetic biases fits in with the observations about tonal languages reported on this blog (Was the First Language Tonal?).
Kirby did not reject the complete nativist package. He accepts the notion that language rests on a “substrate,” or genetically-based brain power, that is used in expressing meaning. Nevertheless, he insisted that the innate features are pre-adaptations that enable “iterated learning,” i.e., the process in which the output of one person’s learning becomes the input for the next person’s learning.
Iterated learning is not limited to humans. Birds learn complex songs through iterated learning, although their songs do not include a semantic component. Instead, they seem to demonstrate fitness. Other primates do not show iterated learning of complex, sequential signals, but they do seem to be able to draw inferences about one another’s intentions.
Kirby concludes that humans are unique in combining the biological prerequisites for learning complex sound sequences and inferences about intentions. Language began by using the capacity to learn complex sequences of sounds to signal intentions. Once begun, cultural evolution took over and quickly optimized signal learnability with expressivity. The key structural characteristics of human language co-evolve in this manner and the nativist assumptions about innate structures that overcome the poverty of the stimulus are unnecessary.
Comments