Why human language is not compressed like computer code

Human language looks messy and redundant compared to the sleek efficiency of computer code. Yet new research suggests that this apparent inefficiency is exactly what makes speaking and understanding language so effortless for the brain.

A team led by linguist Michael Hahn from Saarland University in Saarbrücken, working with Richard Futrell at the University of California, Irvine, has developed a mathematical model to explain why human languages are structured the way they are. Their study, published in Nature Human Behaviour, argues that languages are optimized not for maximum compression, but for minimizing mental effort during everyday communication.

Human language versus binary code

Across the world, around 7,000 languages are spoken, ranging from those with only a few remaining speakers to global languages such as Chinese, English, Spanish and Hindi that are used by billions. Despite their enormous diversity, they all work on the same basic principle: words are combined into phrases, phrases into sentences, and each element contributes its own piece of meaning to the overall message.

From the viewpoint of information theory, a computer-like digital code should be able to convey the same information more compactly, for instance as sequences of ones and zeros. That raises a puzzle: if nature tends to favor efficient systems, why did human communication evolve in such a seemingly complex, redundant form rather than in a tightly compressed digital code?

The new study’s conclusion is that compression is not the main priority for the human brain. Instead, language is organized to make it easier for listeners and readers to predict and interpret meaning in real time, using what they already know about the world.

Language grounded in real-world experience

According to Hahn, human language is anchored in shared experience. Inventing an abstract term for an unfamiliar hybrid such as “half a cat and half a dog” would leave listeners at a loss, because it does not relate to anything they have actually encountered. Likewise, scrambling familiar words into an unintelligible string of letters may preserve all the characters, but it destroys meaning.

By contrast, an ordinary phrase like “cat and dog” is immediately understandable because both animals are common in everyday life. This anchoring in lived experience gives natural language its power: words are not arbitrary symbols in a vacuum, but cues that connect to a rich web of knowledge and expectations.

Why the brain favors familiar patterns

The researchers argue that, counterintuitively, the more “complicated” structure of natural language is actually easier for the brain to handle than a highly compressed code. When people speak or listen, they continually draw on long-term experience with words, grammar and real-world situations. This allows the brain to anticipate what is likely to come next and to process incoming information with relatively little effort.

Hahn likens this to a daily commute. A shorter but unfamiliar route can feel more exhausting than a slightly longer one that is well known, because the unfamiliar path demands greater attention and monitoring. Similarly, a perfectly compressed digital code might be shorter, but without stable, familiar patterns, the brain would need to work much harder to decode and interpret it.

From an information-theoretic perspective, the researchers argue that the brain effectively has fewer “bits” to process when language follows predictable, natural patterns that match our expectations. A purely binary communication system would sacrifice this predictability and context, increasing cognitive load for both speaker and listener.

Predictive processing in everyday speech

One of the clearest illustrations of this idea comes from word order. In German, the phrase “Die fünf grünen Autos” (“the five green cars”) feels natural and instantly interpretable to native speakers, whereas “Grünen fünf die Autos” (“green five the cars”) does not.

In the well-formed version, each word narrows down what can plausibly follow. “Die” signals a certain grammatical category and rules out many others. “Fünf” indicates that something countable is coming next, not an abstract concept. “Grünen” adds information about number and color, limiting the range of likely nouns to a set of plural, green objects. By the time the listener hears “Autos” (cars), there is only one coherent interpretation left.

Because the words arrive in an order that fits the listener’s expectations, the brain can begin to assemble meaning long before the sentence is complete. When the order is scrambled, this predictive process breaks down. The same words become much harder to integrate because the usual cues come in the wrong sequence, and the brain cannot easily build up a single, coherent prediction.

A model of language that prioritizes mental effort

Hahn and Futrell formalized these ideas in a mathematical model that treats language as a trade-off between compressing information and keeping cognitive demands low. Their analysis suggests that the structure of human languages reflects a systematic bias toward reducing mental workload, even when that means sacrificing some degree of raw compression.

In practice, this means that languages typically favor word orders, grammatical patterns and vocabulary that help listeners make good predictions at each moment in a sentence. This design allows humans to communicate complex ideas quickly and reliably without overwhelming their limited attention and memory resources.

Implications for artificial intelligence

The study’s findings may also have consequences for how artificial intelligence systems handle language. Large language models, which power many generative AI tools, already rely on prediction: they estimate the most likely next word based on previous context. A deeper understanding of how the human brain manages this predictive process could help researchers refine these systems to work more naturally and efficiently.

By modeling not just how much information a message contains, but also how easy it is for a human brain to process, future AI tools might better align with the rhythms and constraints of everyday communication. That, in turn, could make interactions with digital assistants and conversational systems feel more intuitive and less mentally demanding for users.