The computational scheme for learning language that I have outlined over the last few pages provides for all the necessary functional components of this monumental task. Controlled experiments in language acquisition show that babies indeed solve it by leveraging a few strategic assumptions about the nature of the data they face and by applying statistical inference to the problems of reference determination and structure discovery.
31
The recipe for learning language seems to boil down to this, then: scan the speech you hear for recurring structures while monitoring its situational context, note statistically significant matches within the speech stream as well as between speech and the outside world, feed what you already know back into the discovery machine, and soon enough it will not be Greek to you anymore.
A great thing about this scheme is that it is perfectly suitable for complete novices: you don’t have to know anything specific about the language you are immersed in before you start. Beyond that, it only gets better: the farther along you are, and the more words and other constructions you already know, the more good candidates for further analysis suggest themselves. Knowing more also helps you make use of context (which with time feels increasingly less opaque) in figuring out difficult passages: you start by learning what co
a
a stands for, and in just a few years, overhearing a conversation at a party, you surprise yourself by guessing right the meaning of
je ne sais quoi
.
It Takes a Village
Before you start humming
All by myself
, though, consider this: given that language is inherently and fundamentally a social game, is “going it alone” really the best way to learn to play it? Because language is an evolutionary game played by many players, various benefits—both for the memes and for their human hosts—may ensue if listeners and speakers collude to make learning easier and if the structures they trade make collusion worthwhile. Language learning on all levels is indeed boosted by social interactions.
In learning names of objects, for example, the baby’s assumption that a new word labels the most salient object would be more effective if a certain degree of cooperation on the part of the caregiver (which need not be conscious and deliberate) can be relied upon. Indeed, when speaking to a baby, people usually signal that they do so by imparting a special modulation to their voice; they also engage the baby’s attention and draw it (by shifting the gaze) to the object being named.
On their part, babies are far from being passive receptacles of information: they often take the initiative in seeking knowledge about the game they find themselves in. Although babies can learn language by simply being around adults who communicate among themselves, participation in social interactions with adults and peers makes a huge difference. In learning names for things, in particular, the baby’s retention of a verbal label works best if the caregiver offers it in response to the baby’s own vocalization directed toward the object, and if it is delivered within an appropriately short time window.
Speakers also often make it easier for listener-learners to match utterances while seeking after the patterns of constructions and their usage. The basic operations of alignment, comparison, and statistical testing are made more effective by the tendency of successive utterances in natural speech to be variations on a common structural theme. The resulting variation sets—runs of structurally related utterances—are particularly common in child-directed speech, where about one-quarter of all sentences appear within one variation set or another. As an example, here is a two-phrase variation set, taken from an Italian mother’s conversation with her very young child:
Dove sono; dove sono i coniglietti
(“Where are; where are the bunnies”). In contrast to the passage from
The Odyssey
, the matching parts here are very close in time to one another, making the corresponding cue to structure much more prominent.
32
While caregivers are generally unaware that they produce variation sets, this and other social behaviors that help children learn do depend on the personality and mood of the speaker. The effects of reduced cognitive stimulation and social interaction—conditions that are often brought about by poverty or parental depression—are far-reaching and permanent. Plopping children in front of a TV will not prevent them from learning to speak, but these children will not master language as well as their socially engaged peers. Unhappiness thus has a way of perpetuating itself.
33
As children become better speakers themselves, they gradually increase their contribution to dialogue, and with it their participation in generating variation sets. The prevalence in adult conversation of variation sets in which all participants have a say suggests that there is more to this phenomenon than helping children learn language. The striking scope of coordination between conversants is illustrated by this transcript of a snippet of kitchen conversation between two adults, which has been aligned to reveal matching parts (omitting line 2, which consisted of a 0
.
2-second pause, and line 5, where Lenore sneezed):
34
Corpus studies show that this kind of coordination is a rule rather than an exception in naturalistic dialogue, while functional imaging reveals corresponding coordination between the brain dynamics of speakers and listeners.
35
The labyrinth of nested constructions that embodies the knowledge of language is thus best thought of not as confined to a single brain but as spanning entire communities, with space always available for new additions to plug themselves in.
At any given time, the person who happens to be speaking chooses one among the many forking paths that run through the labyrinth. At each fork, the speaker’s choice arises from a tug-of-war among the probabilistically weighted available options. The utterance that takes shape is thus conditioned on a variety of factors: the speaker’s memories (experiential history) and brain dynamics (intent); the constraints imposed by the structure of the labyrinth (grammar); the environmental context within which the conversation is situated; and last but not least, the dynamics of other brains that participate in the exchange, which affect the speaker’s brain most significantly through the medium of language.
By putting in place a general-purpose medium of communication, the memes that comprise language support the emergence and propagation of other memes. These include anecdotes of hunting and gathering, epic poems, pieces of gossip, reports of miracles, propaganda, news, slander, jokes, conspiracy theories, election promises, family and tribal history—in short, all those aspects of human culture that can spread by word of mouth.
The medium of language is symbolic and self-enriching. By distilling memories of use and context into shared patterns of structure and meaning, language empowers those who speak and understand it to build virtual bridges between brains and thus to trade mind states through what must seem almost like telepathy to the uninitiated. A passage brimming with meaning is, for those who can read it, like the friendly alien mother ship in
Close Encounters of the Third Kind
—a humble structure suspended in midair next to Devil’s Tower, slowly revealing to the stunned humans as it turns upside down against the starry Wyoming sky a vast, scintillating constellation of light and sound.
Because language has what seems to be a direct line to the mind’s emotional core, it can be deadly—but so can all the tools that our species ever invented, starting with the flint blade. Best learned with a little help from our friends, language is the most advanced contrivance we have for fashioning virtual worlds and for bringing our friends there. As such, language is an accomplished and exquisite tool for generating happiness—and, of course, for sharing it with the rest of the village.
SYNOPSIS
What had been the boringly biological evolution of our species turned into a wild ride when some of our ancestors, habitually engaged in a constant cognitive arms race against all and sundry, got help from unexpected quarters—a bunch of behavior-turned-replicator memory patterns, or memes. The subsequent co-evolution of humans and their culture has been, and still is, sustained by two key properties of the loose coalition of memes that constitutes language: the categorical or digital nature of phonemes and words, and the constrained manner in which these elements can be composed to form hierarchical constructions.
Such constructions reuse familiar building blocks, yet allow for the expression of complex and potentially novel meanings in the standardized, hence interpretable, form of a sequence of sounds, gestures, or marks in a visual medium. These, in turn, serve as the hints and clues that the brain of a listener or a reader works from in its search for a reasonably well supported guess as to the speaker’s or writer’s intent. The constraints within the grammar of constructions that emerges through hierarchical composition take the form of statistical dependencies. Thus, words are highly probable sequential patterns of phonemes, while larger constructions are likewise defined by some words depending on the choice of others.
Its reliance on categorical representations and on statistical dependencies reveals the organic connection between language and the rest of cognition. (Insofar as every neuron’s output expresses a pattern of conditional probabilities over its inputs, dependency is the universal calculus of computation in the brain, whose utmost goal is to support forethought by mirroring patterns of conditional probabilities linking prior experience to possible futures.) And yet, language is unique among cognitive functions in the degree to which—in the best co-evolutionary tradition—it both helps and is helped by social interactions. As social animals, we revel in group play, which is what language evolves to promote and we evolve to master. Happiness and misery being the two-pronged stimulus with which evolution prods its pack animals, is it any surprise that we can be moved to tears or to laughter by a few well aimed words?
6
Nobody, at Home
The web of cause and effect. Through a scanner, darkly. Because it’s there. Connecting the dots. Flow. Soul music. Being and time and zombies. That which we are.
You who are on the road
Must have a code that you can live by
And so become yourself
—CROSBY, STILLS, NASH, AND YOUNG,
“Teach Your Children” (1970)