Suppose that a scientist wants to study a striking animal behavior; say, the courtship display of the stickleback fish, or the cooperative agriculture of leafcutter ants. She will, of course, ultimately want to know the underlying mechanisms of these behaviors: How do they work? How did they evolve? What can we learn from them? But no student of animal behavior would dream of asking these questions without first systematically discovering the facts; beginning with extensive field observation in the wild, then moving to experiments and modeling in the lab. Why, then, have linguists emphatically denied any value in directly observing linguistic behavior?
The culprit is a bad idea: that a science of language should be concerned only with competence (the mental capacity for producing sentences), and never with performance (what happens when we actually talk). Here is its decidedly dualist reasoning: When the idealized language patterns tucked away in the mind get 'externalized' in communication, they are filtered and shaped by contingencies such as motor constraints, attention and memory limitations, errors of execution, local conventions, and more. As a result, it is argued, performance bears little useful relation to the pre-defined underlying object of study: competence. Students of linguistics have been taught not to waste their time with the worldly facts of performance.
This idea belies an unaccountably narrow view of what language is. It has diverted linguists' attention from many substantial questions, each with deep implications. Just a few examples: Without looking at performance, we wouldn't see the systematic and ingenious ways in which people handle the constant speech errors, hesitations, and misfires of conversation, along with the social delicacies of navigating these bouts of turbulence. Without looking at performance, we would not be witnessing the emerging breakthroughs from statistical research on newly-available large language corpora, with results suggesting that we can infer competence from experience with performance. Nor, finally, would linguistics have a causal account of how languages evolve historically: In the cycle of language transmission going from public (someone speaks) to private (someone's mental state is affected) and back to public (that person speaks), and so on indefinitely, both the private domain of competence and the public domain of performance are equally indispensable.
Influential traditions in the discipline of linguistics have embraced an idea that makes little sense, given the fact that language is, after all, just another striking animal behavior. Instead, the science of language should begin with fieldwork observation, for performance is ultimately our only evidence for competence. Perhaps the most unfortunate outcome of this idea is that generations of linguists who have eschewed the study of performance now have nothing to say about the essentially social function of language, nor about those aspects of social agency, cooperation, and social accountability that universally define our species' unique communicative capacity.