Are Our Children Tur[n]ing into Linguistic Coiners?

 Connecting Pedagogy, Social Cognition, Language Acquisition, and Artificial Intelligence

What we know about the world and about each other, even as infants, is so much more than what we can directly experience. In a Symbolic Systems lecture about pedagogy and social cognition, for example, we learned that very young infants about the age of 2 or 3 years old do not have the ability to lie, even when knowing that lying would benefit their situation. However, after a certain age, children begin to conduct “primary lying”, which develops in plausibility as they develop their cognitive abilities (Talwar & Lee). How do children learn to lie and manipulate at such a young age, especially when they’re not taught in school?

Likewise, the myriad words we use to describe elements of our world are more varied and fluid than the concrete lexicon of words taught to us in an elementary-school reading class; words like “funner” or “funnest” come to us naturally despite their ungrammatical nature and the fact that we were never taught them in a classroom setting. We automatically apply these linguistic rules and learn to coin new syntactical elements without even learning how to.

In Clark and Hecht’s Learning to Coin Agent and Instrument Nouns, this productive application of linguistic knowledge is exemplified by an experiment where children are asked questions about agents, either asking them to name an agent (subject noun) or their action (verb base). For example, they would be given a picture of something and told, “I have a picture of a person called a biter. What does a biter do? A biter is someone that ---,” or, conversely, “I have a picture of someone who bites. What could we call someone who bites? Someone who bites is called a ---,” and expected to fill in the blanks. Developmental linguistic ability was on display as a mean 78% and a mean 48% of children were able to apply the –er affix to agent and instrument nouns, respectively, yet 36% more and 30% more were able to apply the affix to agent and instrument nouns respectively in the age bracket of (5;3-6;0) than in (3;0-3;8) years. Evidently, Clark and Hecht’s paper suggests “children and adults alike…[learn to] fill gaps in their vocabulary by constructing new word forms to carry meanings for which conventional forms have not yet been Iearnt or happen not to exist.” (Clark & Hecht 22). Thus, we can understand from this that this amazing feat of noun coining is not something that is explicitly learned in school, yet still learned through social interaction and acquisition of linguistic abilities through experience. Like the ability to lie, or manipulate others, the ability to constantly manipulate language to express ourselves is an implicitly learned skill that is evolved and grown through social transmission of information in a social environment.

We now know these feats of explicitly unlearned human development are possible, but how is it possible that we develop these extensive abilities of applied social cognition and productive control of language formation without even learning them? Neuroscience, psychology, and symbolic systems researchers all point towards the development of Theory of Mind, of the understanding of oneself and of others as mental beings with their own mental states, learned by social understanding of others’ intentions and attention. These precursors to Theory of Mind development show a sequential acquisition of social cognition, just as abilities in linguistic formation are formed sequentially, based on Clark and Hecht’s evidence. Using learned principles of semantic transparency, allowing “[children] to choose known elements for forming new labels for categories”, productivity, drawing “their attention to more specialized devices for expressing such meanings as ‘agent’ and ‘instrument’”, and conventionality, causing them to “drop their unconventional means [for expressing a particular meaning] in favor of a conventional one,” (Clark & Hecht 7). These principles are likely learned implicitly through linguistic interactions with others in society at an early age in their linguistic development when their linguistic abilities are probably most malleable. It is here, in the development of linguistic abilities such as “spontaneous coinages”, that we see the psychological processes of social cognition, or learning from our experiences in a social environment, efficiently at work.

The connections between pedagogy, language acquisition, and social cognition have immense implications outside of the realms of linguistics and cognitive neuroscience. The notion that linguistic abilities are sequentially learned and developed through social interaction completely shifts the focus of AI and human-computer interaction researchers. Now, instead of creating a hard-coded machine programmed to implement certain functions, these researchers may be inclined to look into creating an AI with a malleable, child-like “mind” that develops its linguistic and cognitive abilities through interaction in society. As Turing said, and as both Talwar and Lee’s and Clark and Hecht’s research seem to implicate, “instead of trying to produce a program to simulate the adult mind, why not rather try to ... simulate the child’s?”