The Computational Grammar of Human Behaviour in a Biological Environment
This paper introduces DynaClade, an unsupervised algorithm that reveals human spontaneous behavior in virtual environments is organized by a second-order, context-dependent computational grammar rather than pre-planned sequences, characterized by distinct latent states and a newly identified non-locomotor sway pattern.
Original paper licensed under CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/). This is an AI-generated explanation of a preprint that has not been peer-reviewed. It is not medical advice. Do not make health decisions based on this content. Read full disclaimer
Human beings are constantly in motion, a flowing stream of steps, gestures, and shifts in posture that allow us to navigate the world and achieve our goals. While we can easily describe a single movement like picking up a cup, understanding how the brain strings together thousands of these spontaneous moments into a coherent life is a profound puzzle. Scientists have long wondered whether we move by following a rigid, pre-written script, or if we simply choose the next step based on what we just did and where we are right now. This question touches on the very nature of free will and the limits of our mental planning. If our actions were entirely pre-planned, our minds would need to calculate every future step before taking a single one, a task that would quickly overwhelm our cognitive capacity. Instead, many theories suggest that our behavior is more fluid, shaped by immediate needs and the history of our last few moves, rather than a distant, fixed blueprint.
To investigate this, researchers created a digital world where people could move freely without being told exactly what to do. They built a virtual reality environment filled with fruit trees and hidden dangers, such as snakes, bears, and panthers. Twenty-three participants wore motion-capture suits and headsets, allowing the scientists to record their every movement in high detail as they wandered through these risky landscapes. The goal was simple: gather as much fruit as possible while avoiding contact with the threats. Crucially, the participants were not given a script. They were free to run, walk, crouch, or stand whenever they felt it was necessary. The researchers then used a new computer program, which they named DynaClade, to watch the hours of recorded movement and automatically sort it into distinct actions. This program did not know what a "run" or a "stand" looked like beforehand; it had to discover these patterns on its own by looking at the speed and shape of the body's movements.
The computer successfully identified a rich vocabulary of twenty-two broad types of movement, ranging from walking and running to more subtle actions like kneeling to pick fruit or standing still. One of the most interesting discoveries was a specific behavior the researchers labeled "hesitant." This was not a freeze of fear, but a slow, swaying movement on the spot, often occurring right before a person decided to move in a new direction. It appeared to be a moment of gathering information, a physical pause where the person seemed to be weighing their options. The program also found that these actions were not random. When a person was near a fruit bush, they were far more likely to reach or kneel. When a threat appeared, they were far more likely to run or stand still, depending on the type of animal. The movements were clearly tied to the environment, proving that the computer's classification matched real-world logic.
The deeper mystery, however, lay in the order of these actions. Do people move in fixed chains, like a domino effect where one move forces the next? Or do they choose each step based only on the one immediately before it? The researchers tested this by looking at the sequences of actions. They found that the order was indeed non-random, but it was not a rigid script either. The data showed that the best way to predict what a person would do next was to look at what they were doing right now and what they had just done. This is known as a second-order structure, meaning the immediate past matters more than a long history of previous moves. The study ruled out the idea that humans are constantly running complex, long-term plans for every single step they take. Instead, the evidence suggests that our brains use a much simpler, more efficient strategy. We seem to select our next move based on our current situation and the action we just finished, adapting moment by moment to the world around us.
This approach allows us to handle the complexity of real life without getting stuck in endless calculation. The researchers found that once they accounted for the context—such as whether a person was foraging or escaping a threat—the sequence of actions became highly predictable. It was as if the person switched between different sets of rules depending on the situation, but within each situation, the next move was largely determined by the one before it. This finding challenges the notion that we are constantly executing pre-assembled sequences of behavior. Instead, it paints a picture of human movement as a flexible, responsive flow, guided by the immediate present and the very recent past. By observing people in a realistic, risky setting, the study reveals that our daily lives are not a series of rigid routines, but a dynamic dance of adaptation, where we constantly adjust our next step based on where we are and what we just did.
Drowning in papers in your field?
Get daily digests of the most novel papers matching your research keywords — with technical summaries, in your language.