BABIES / EMBODIED INTELLIGENCE
Infants understand abstract ideas before they learn words.
Almost all people learn their first language as children, and don’t remember themselves before that point. Actually, language might be why we don’t remember. Here’s a diagram from a 1948 study that compared the development of language with the development of memory:
So, three curves that rise with age: “V” for vocabulary, “R” for recollection, and “A” for the Aussage test, a memory test involving complex verbal recall.
You could, of course, interpret this graph both ways: maybe better memory helps with superior language, maybe superior language helps with memory. Notice though that vocabulary rises first, and improved recollection follows, which would seem to indicate that it is language that allows us to remember our childhood experiences better.
What, then, goes on in the minds of babies who cannot yet talk?
Many human ideas do not exist without words — for example, the idea of a center of gravity, or a limited liability company, or a criminal indictment. And so, for most of us, it is difficult to picture what it is like to have never known any form of language — our inner worlds are full of these completely virtual things that seem as real as physical objects. Almost all people learn their first language as children, and don’t remember themselves before that point — childhood amnesia subsides around the age of 4, by which point children usually speak pretty well. Rare accounts of those who learned their first language in adulthood imply a sharp line separating their life in two completely different halves.
But there clearly is some thought without language. Other animals may lack in limited liability companies, but it would be hard to argue they don’t think. The same holds true for prelinguistic infants: any parent would testify that their child could understand things well before being able to say them.
What exactly infants can understand, though, remains a matter of debate. Just how far can human thought go without words? The question is important not only because we want to know more about our children (although that, too), but because the relationship between language and the mind is one of the most pressing questions of our time — it is at the center of the debate about AI and whether or not it could ever become conscious or malicious.
There are, of course, lots of things that prelingual infants can’t do. For instance, until 18 months of age, most can’t recognize themselves in a mirror. Another classic example is the so called A-not-B task. An experimenter repeatedly hides a toy in location A within the infant’s reach, and the infant successfully retrieves it. Then, the experimenter moves the toy to location B while the infant watches. Most infants under a year old, however, continue to look for the toy in location A, unable to process the thought that the location has changed.
But sometimes, an infant can surprise you. For example, many languages have this odd association between round shapes and “round” sounds. Words that describe smooth, round things also tend to sound kind of round: ball, round, amoeboid; whereas words that describe pointy things often sound sharp and jagged: spike, knife, pincers. You would think that this is cultural: as we learn language, we learn to associate its sounds with meanings, and so words describing sharp objects start sounding sharp. Except that infants have the same associations before they know what knives or pincers are: they think that jagged shapes are congruent with the sound kiki, and round shapes with the sound bouba, as would we all:
It might be rooted in an association between the sound and the shape of the mouth making that sound, and the association itself might either be innate, or learned very shortly after birth. This example shows how studies on prelingual infants can tell us which parts of our cognition are derived from culture, and which ones from biology.
New research by Jean-Rémy Hochmann’s lab in Lyon shows that the cognition of prelingual infants is more complicated than we give them credit. There are apparently signs of grammar before there’s any sign of words: a primordial, foundational grammar of abstract thought. The study, led by Liuba Papeo, showed that infants assign pictures of humans engaged in various interactions (kissing, kicking, helping) to one of two roles — agent or patient, and are surprised when the roles are unexpectedly reversed. This is equivalent to assigning words in a sentence the roles of subject and object.
So, based on this, you could say that the idea of “subject” and “object” is built into our brain on a deeper level than words are. It’s likely that there are many other ideas similarly built in. I think this study would please Noam Chomsky, whose central thesis has always been that what we call “language” is externalization of innate cognition. According to this view, words — what we learn from culture — populate a pre-existing scaffold of thought, which does not depend on culture, but depends instead on genes and evolution.
Language, Chomsky has always said, is not for communication — it is for thinking. I always have a hard time getting this point across to my students. Surely dogs and infants can also think, they say? This study clarifies this point. You could say there’s “communication language” — grammar plus words, which we use to exchange thoughts with others — and “Chomskyan language”, the broader, abstract grammar of cognition that we use internally. Infants don’t yet have “communication language”, but, as this study shows, they have “Chomskyan language”.
Figuring out where exactly the lines are drawn between the two is a key task for neuroscience. It might turn out a large part of the logical structure of our inner worlds is derived not from culture, as one would expect, but from physical, biological factors (such as the shape of the mouth pronouncing kiki or bobo). If this turns out to be true, then language models, which lack this grounding in physical reality, would prove to be fundamentally deficient compared to embodied brains of humans and other animals — as indeed some neuroscientists argue. Maybe the patterns that the physical world produces, and our brain perceives through the medium of our body, are too complex to be captured in language alone.
But it also might be that this embodied part of our cognition is negligible compared to the infinite possibilities of mind building created by language. Maybe you don’t have to observe real-life humans to understand what a subject and an object is. Maybe words alone can create a mind that would match, or even exceed, a human mind. In the movie Arrival (and Ted Chiang’s respective novella) learning an alien language allows a linguist to see time in a different way. Maybe we do learn most of our conception of the world from words — as Sapir-Whorf hardliners suggest — and if so, maybe artificial intelligence can achieve human level without ever having been a baby.
I can see it both ways.
A version of this post was published on my Psychology Today blog.