Abstract: One of the driving forces of language evolution is the selection of variants that suit the communicative needs of its users. Crucially, fitness of linguistic variants may largely depend on the structure of the environment in which language is learned, transmitted, and used. This hypothesis has gained support in various domains. We apply it in the context of scalar terms with a major focus on quantifiers, such as most. Based on a model that combines logic and evolutionary game theory, we argue that such signals might have evolved as stable semantic units through adaptation to general communicative principles and distributional properties of the environment such as normality.
Introduction
In the study of language evolution, a central question is why certain linguistic structures are universal. One such universal is the proportional quantifier most. In formal semantics, most is typically defined by a threshold of 1/2 (e.g., "Most As are B" is true if more than 50% of As are B). While logically there is no reason the threshold couldn't be 60% or 80%, cross-linguistic evidence suggests a consistent association with the 1/2 threshold. This paper investigates how environmental constraints shape these semantic thresholds, proposing that the 1/2 threshold for most is an evolutionary adaptation to normally distributed environments.
Modeling Communication as a Game
To measure the "fitness" of linguistic variants, the study employs a Discrimination Game involving a sender and a receiver. The agents are presented with two different contexts (e.g., two groups of people with different heights). Success occurs only if both agents, using their internal thresholds, can distinguish between the contexts using a shared signal.
This communicative success is then analyzed through the lens of Evolutionary Game Theory (EGT). In this framework, linguistic strategies (thresholds) are evaluated for their evolutionary stability. A strategy $p$ is stable if a small group of "mutants" using a different threshold $q$ cannot successfully invade and replace the prevalent strategy in the population.
The Role of Environmental Distribution
A critical component of this model is the environment, represented as a random variable $X$ following a normal distribution $\mathcal{N}(\mu, \sigma^2)$. This reflects the natural clustering of properties like height, weight, or proportions around an average. The paper argues that the structure of this distribution directly dictates which thresholds are most robust against invasion.
Evolutionary Stability of Thresholds
The model demonstrates that while multiple thresholds might technically be immune to invasion, they are not equally fit. Strategies closer to the environmental mean ($\mu$) exhibit higher communicative fitness and greater stability.
As shown in the figure above, when a prevalent strategy is far from the mean (e.g., at the extremes of -2 or 2), the population threshold required for a better strategy (one closer to the mean) to take over is very low. This indicates that populations using sub-optimal thresholds are highly vulnerable to invasion by more efficient strategies. Conversely, strategies near the mean (0 in a standard normal distribution) require a much larger "mutant" population to be displaced, representing a state of high evolutionary stability.
The Road to "Most"
The paper bridges the gap between simple scalar terms (like "tall") and abstract quantifiers. If first-order concepts evolve thresholds at their means due to environmental pressure, then second-order properties—such as the proportion of objects satisfying a property—will also follow a normal distribution. For proportions, the "mean of means" naturally converges to 1/2.
Consequently, the threshold for most evolves toward 1/2 because that point provides the maximum discriminative power in a world of normal distributions. This formal proof aligns mathematical modeling with empirical linguistic data, suggesting that the semantics of most are a direct result of optimizing communication for our environment.
Conclusion
While cognitive constraints like memory and processing speed undoubtedly influence language, this research highlights the profound impact of external environmental structures. The 1/2 equilibrium for most represents a stable outcome of agents adapting to the distributional properties of the world. Future work may extend this framework to other fuzzy quantifiers like many or few, further mapping the relationship between nature and the evolution of meaning.
Summary updated