Human language can seem inefficient compared with computer code, but researchers argue that its structure is optimized for the brain rather than for maximum compression. A new modeling study suggests people rely on familiar patterns to reduce mental effort during real-time conversation.
The work, by linguist Michael Hahn and cognitive scientist Richard Futrell, was published in Nature Human Behaviour. Using information-theory-based modeling, they examined why languages worldwide tend to favor predictable word patterns instead of highly compact encodings.
Efficiency for brains, not bits
In principle, the same message could be transmitted with fewer symbols, similar to how computers use binary strings. The researchers contend that such a system would be harder for humans to learn and process because it would not align with how people store knowledge and anticipate meaning.
Natural language, they argue, is tightly linked to shared experience, letting listeners map words onto familiar concepts quickly. That connection helps speakers avoid creating arbitrary, maximally compressed labels that would be information-dense but difficult to interpret.
Predictability lowers cognitive load
The model emphasizes that comprehension is incremental: listeners use each word to narrow down likely meanings before a sentence ends. This predictive processing makes everyday communication feel almost automatic, even if it is not mathematically optimal in terms of compression.
As an illustration, the authors point to how grammatical order guides expectations in languages such as German. When familiar cues arrive in the expected sequence, the brain can prune unlikely interpretations early, whereas scrambled word orders force more effortful processing.
What the findings suggest for AI
The researchers say the results help explain why languages converge on structures that are learnable and robust under noisy, fast conditions like speech. Rather than chasing minimal code length, languages appear to balance expressiveness with the constraints of memory, attention, and prediction.
The same logic could inform how developers evaluate and design large language models, which already rely heavily on predicting likely next words. The study suggests that systems built to communicate smoothly with people may benefit from prioritizing human-friendly predictability over pure information compression.
Leave a Reply