Erin S. İşbilen
Human language is extraordinarily complex. The average adult knows hundreds of thousands of words that can be combined in an infinite number of ways. But how do we learn this immense amount of information? How do we blossom from novices, either in childhood or as second language learners, into skilled speakers who use language with such automaticity and ease?
The question of how individuals learn language is a problem that has puzzled researchers for decades. Scientists and philosophers alike have devoted long and decorated careers to exploring this subject, often unearthing as many questions as answers. Yet, young children can figure out how to use language remarkably quickly very early during development, tackling obstacles that scientists are still learning to describe. How is it that children can learn language with such apparent facility despite its overwhelming complexity?
For many years, the dominant view of language evolution was that the brain had adapted to learn language. The idea was that humans were born with a genetically encoded blueprint for learning linguistic material. While the animal kingdom hosts an immense array of communication systems, such as birdsong, whale song, visual signaling in cuttlefish, and seismic communication in elephants, human language stands out in its dazzling complexity and diversity. There are over 7,000 human languages currently spoken around the planet, many of which sport dramatically different features. For example, in tonal languages like Mandarin, the tone in which a word is spoken can lend the same wordform (e.g., ma) dramatically different meanings. Straits Salish, a language spoken in British Columbia and its surrounding regions, is said to have no strong distinction between nouns and verbs. Given these considerations, it is easy to understand why the view of the human brain evolving for language prevailed as the reigning hypothesis for so long. To our current knowledge, there is no other animal communication system that hosts as much variation in how individuals signal to one another within a single species. Some scholars thus reasoned that through the course of human evolution, language may have become built into the human brain through the cultivation of language-specific mechanisms and neural regions. The idea was that language had become encoded in our very genome.
However, despite the apparent promise of this account, the last quarter of a century of research has discovered that there is actually very little evidence for the genetic endowment of language and language-specific brain regions. To the contrary, language is one of our most widely distributed cognitive abilities: it draws on neural circuitry involved in many other cognitive functions. For example, Broca’s area—a brain region located around the left temple—may appear to be language-specific. Damage to this area leads to dramatic language impairments, particularly in an individual’s ability to process and produce sequences of words. The speech of such individuals makes sense semantically, but the order in which the words are spoken is often ungrammatical. While central to language, Broca’s area damage is associated with difficulties in processing sequences of information in general, whether they are linguistic or not (for example, sequences of shapes appearing on a screen one after another). This implies that Broca’s area is involved in the processing of many different types of information, making it unlikely that it evolved for language specifically. This makes sense when we consider the rate of biological evolution versus the rate of language evolution. Biological evolution is very slow, unfolding over generations of learners. By contrast, language changes very rapidly within our very lifetime—we can all think of words and phrases that exist now that did not exist only a few years before. This difference in the rate of evolution poses a fundamental challenge to biological accounts of language evolution. By the time the brain might have evolved to learn a specific structure, that structure would have likely changed. Language is a moving target, with a life cycle and evolutionary trajectory of its own.
In recent years, the theoretical emphasis in cognitive science has shifted away from asking how the brain evolved to learn language to asking how language evolved to be learnable by the human brain. In addition to the differences in the rate of change between language and biology, this question makes sense in the broader context of evolution: while humans can survive without language, language cannot survive without its human hosts. Language therefore faces a far stronger pressure to fit our communicative needs and cognitive limitations. Indeed, even Charles Darwin likened language change to a sort of survival of the fittest in his 1871 book, The Descent of Man. Darwin argued that words and grammatical structures that are easy for us to learn will proliferate, and their continued use will enable them to “survive” by being passed onto the next generation. By contrast, linguistic structures that are not easy for us to learn and process will be used less and less and eventually go extinct. It is therefore more likely that language culturally evolved to fit our existing neural mechanisms and learning processes, rather than requiring dedicated brain regions of their own.
If language evolved to be learned by the human brain by piggybacking on pre-existing mental mechanisms that process a wide variety of materials, what then are some of these mechanisms that determine which linguistic structures live or die? One critical factor is the limitations of our working memory—or, our ability to encode and remember information on a moment to moment basis. To illustrate, speech is extremely fast, with most speakers producing 10 to 15 speech sounds per second. But if we contrast this with our working memory limit for sound sequences, most individuals are only capable of maintaining 3 to 5 linguistic items in memory at a time. The same limitations extend to sign language, which poses very similar processing challenges as auditory speech. To complicate matters even further, both spoken and signed speech is extremely brief, lasting only about 75 milliseconds on average. Under these pressures, language should be impossible for our brains to process. Yet miraculously, we manage to speak and understand one another despite these staggering odds.
Given the challenges posed by the nature of the speech signal, our brains face a strong pressure to immediately process incoming material before the signal is lost or overwritten by new material. One way the brain achieves this is through a memory process called chunking. Chunking allows the brain to recode incoming information from its smallest units into larger bits of information that is easier for us to remember. To illustrate, imagine you were tasked with remembering the following set of letters: U A C F I S B I A. Remembering this sequence requires keeping nine pieces of information in memory, which exceeds the average working memory limit of approximately four pieces of information. However, if we contrast this with trying to recall the following sequence (C I A F B I U S A), this proves to be much easier to remember for most English speakers. This is because in the second example, we can chunk these 9 letters into 3 acronyms (CIA, FBI, USA), which significantly reduces the amount of information that we need to commit to memory. This effect is also observed for non-linguistic information such as sequences of numbers or musical sequences, meaning that chunking is a domain-general process rather than domain-specific: it is involved in the processing of a wide variety of information rather than just language. In fact, some of the earliest demonstrations of chunking showed how expert chess players remembered the positions of pieces on the board by chunking them into larger configurations, such as the French Defense or the Queen’s Gambit. There is even evidence of chunking in non-human animals such as monkeys and pigeons, which suggests that this memory process may have deep evolutionary roots.
Chunking has been shown to play a key role in human language learning. It can enable us to discover the structures present in speech by helping us group individual syllables into words and words into phrases, which is one of the first challenges of learning a new language. Individual differences in chunking, or how well participants perform on memory tasks like the example provided above, strongly predicts important linguistic abilities such as reading and other language skills. Deficits in chunking abilities have even been correlated with language-related difficulties in clinical populations, suggesting that such memory abilities play a critical role in our ultimate linguistic outcomes.
Beyond the timescale of the individual, chunking on a moment-to-moment basis has major implications for how human language evolves over time. The way we learn and the conversations we have on the timescale of milliseconds shapes language across the millennia. Words and phrases that are easy for us to chunk and retain in memory will be favored in use and passed onto the next generation. Those that are difficult for us to chunk will be used infrequently and eventually die out. This repeated cultural transmission of information between speakers gives rise to the linguistic structures that we are familiar with today. Differences in the kinds of information that is transmitted can help give rise to new languages and dialects over many generations.
Through continued practice, we learn to grapple with—or how to chunk—the immense complexity of our linguistic input, breaking it down into manageable bits of information that we can easily remember and pass on. Rather than the brain evolving to learn language, we as learners shape our own destinies by promoting the endurance of information that best serves our cognitive capacities and communicative needs.
For further reading on this subject, see Isbilen, E. S. & Christiansen, M. H. (2020). Chunk-based memory constraints on the cultural evolution of language. Topics in Cognitive Science.
Erin İşbilen (PhD), after studying psychology (major) and literature (minor) at UCSC, finished his PhD at Cornell University. Her main interest is how domain-general cognitive mechanisms can explain many complex linguistic phenomena. Her dissertation how individual differences dealt with how individual differences in statistical learning and memory predict linguistic proficiency in adults, and on developing improved methods of testing these phenomena.
The Cognizer is a publishing platform initiated by CogIST, a cognitive science community from Turkey. On this platform, articles and essays on different topics from different fields of cognitive science are published in a way that would bridge the gap between public audience and experts.