查看文章

There is no more natural, yet rich and complex, behavior than the human conversation. Two (or more) minds and bodies meet and engage myriad cognitive processes (attention, perception, memory, language, decision-making, etc.), unfolding on multiple timescales, which are interdependent both within the individual and across interlocutors. Studying the interlocutor in isolation or breaking the flow of conversation into unconnected pieces (trials) limits what can be learned about this rich behavior. Yet, until recently, language use was primarily studied in exactly this way.

Maselli et al.[1] are absolutely right that the future of cognitive and neural sciences must continue to move toward ecologically valid methods for data collection coupled with computational models that provide insight into those cognitive and neural processes. Their examples from problem solving, joint action and robotics are illustrative and compelling. Yet, language processing research plays a smallish role in their discussion despite a rich and storied history of gradual evolution toward ecologically valid methods coupled with computational models. We offer here a brief review of that evolution and speculate about its future trajectory. The value of capturing language as it occurs “in the wild” has long been clear to language researchers. Linguists, sociologists, and anthropologists created corpora first from existing texts [eg, newspapers and books; 2], then from recordings of interactions (eg, phone calls between strangers, parents talking to their children [3–5]). By applying computational analyses to these naturally occurring text-based data, researchers have uncovered many patterns …