Looking at how language became so complex
When it comes to language, one could say ‘you speak as you learn’. “The fundamental structural features of language are a consequence of how a language is learned,” explains Kenny Smith, a professor at the University of Edinburgh. “Because language is passed down from generation to generation, the mistakes and little modifications we make eventually become part of the linguistic system.” During prior research, which received funding from the EU, Smith helped develop modelling and experimental techniques for studying how languages evolve. Although these kinds of models have done wonders in terms of demonstrating the evolution of languages, they did so by using simple languages. “The kinds of languages people previously looked at have been very simple,” adds Smith. “However, because real human languages are enormously complex, we needed to take our modelling capabilities to the next level.” With the support of the EU-funded ELC project, Smith has expanded methods pioneered at the University of Edinburgh’s Centre for Language Evolution to address two key questions relating to language complexity. “First, we wanted to know how linguistic complexity influences language learning,” he explains. “Second, we looked at where any conditions were particularly conducive to developing complexity in a language.”
Individual experiments and computational modelling
To answer these questions, the ELC project, which received funding from the European Research Council, used a combination of individual experiments with human participants and computational modelling. “We did some lovely modelling work on language evolution in complex heterogeneous populations that I am really excited about, and a large experimental study on the evolution of irregularity that is very cool,” notes Smith. Out of this work, the project came to several interesting conclusions. For example, when looking at whether a particular kind of organisation in inflectional paradigms facilitates learning, researchers found that its impact was much less than originally thought. “This opens up an interesting set of questions that I thought had been neatly tidied up and shows that there’s no substitute for testing these sorts of claims empirically,” says Smith. Another line of research focused on whether people copied an optimally informative or over-informative partner in interaction. “If I’m trying to get you to give me the cup and not the pencil, I can say ‘pass me the cup’, I don’t have to say, ‘pass me the red cup’,” explains Smith.
Mimicking your interlocutor
While there’s a growing body of work that emphasises communicative efficiency, where people should focus on being optimally informative and avoid wasted effort, researchers found that that’s not necessarily the case. Instead, they discovered that people tend to copy whatever their interlocutor does. In other words, if your partner behaves in a redundant manner, you’ll follow suit. “This demonstrates the need to integrate these social effects into our theories about efficiency and optimality in language design,” remarks Smith. “How do languages end up being so well-designed, with complexity in exactly the right places, if so much depends on us just talking like the people we are talking to?”
Showing what can be done
Together, these findings show how one can take big ideas in evolutionary linguistics and turn them into specific claims that can be experimentally tested using models and experiments. “There are a fairly limited number of projects that have tried to turn theories about the impact of learning or interaction into hypotheses that can be tested experimentally,” concludes Smith. “Our work may not be the last word on these topics, but I think it shows what can be done.”
Keywords
ELC, language, linguistics, computational modelling, evolutionary linguistics, language learning
