COMPUTATIONAL MODELS OF LANGUAGE EVOLUTION

Аннотация

The evolution of language represents a complex and multifaceted phenomenon central to human cognitive and social development. Understanding the mechanisms that underpin language emergence, change, and stabilization over time has been a longstanding challenge in linguistics, anthropology, and cognitive science. Recently, computational modeling has emerged as a powerful approach to investigate language evolution by simulating the interaction of agents engaged in learning, communication, and cultural transmission of language. This article explores the major computational paradigms employed to model language evolution, including agent-based models, iterated learning frameworks, and evolutionary game theory models. These computational approaches illuminate how linguistic structure can arise, be transmitted, and evolve within populations. Furthermore, advances in neural network-based learning models and large-scale empirical data integration are discussed as promising avenues for future research. Despite limitations, computational models continue to provide valuable insights into the dynamic processes shaping language and hold the potential to bridge gaps between theoretical predictions and observed linguistic phenomena. Keywords Language evolution, computational modeling, agent-based simulation, iterated learning, evolutionary game theory, neural networks, cultural transmission, language acquisition, linguistic change, artificial intelligence The evolution of language has fascinated researchers for decades due to its fundamental role in human society and cognition. As a uniquely human trait, language enables the expression of abstract thought, social coordination, and cultural knowledge transmission. Yet, understanding how language emerged and evolved remains a significant challenge, primarily because direct empirical evidence of early human language is inaccessible. Traditional linguistic and anthropological approaches have relied on indirect methods such as the study of language families, fossil records, and comparative analysis of extant languages, but these methods are limited in their capacity to capture the dynamic processes of language change and emergence. The advent of computational modeling has provided new methodological tools that allow researchers to simulate and 90 INNOVATIVE RESEARCH IN SCIENCE International scientific-online conference explore the processes driving language evolution within artificial populations of interacting agents. Computational models of language evolution attempt to reproduce the mechanisms by which language is learned, used, and transmitted in populations over time. By simulating agents equipped with cognitive and communicative capacities, these models test hypotheses about the origins of linguistic structure, the role of learning biases, and the influence of social interactions on language change. Central to these efforts are frameworks that include agent-based models, iterated learning models, and evolutionary game-theoretic approaches. Each paradigm emphasizes different aspects of language dynamics but collectively contributes to a richer understanding of the evolutionary process. Agent-based models simulate populations of autonomous agents that communicate with one another to coordinate shared linguistic conventions. Each agent holds an internal representation of linguistic knowledge—such as a lexicon or grammar—and learns from interaction outcomes. Through repeated communication and feedback, agents adapt their internal representations, leading to the emergence of shared vocabularies or grammatical structures. A seminal example is the Naming Game, wherein agents negotiate names for objects until a consensus is reached. Such models highlight how simple interaction rules and local learning can give rise to global linguistic coherence without centralized control. Extensions of agent-based models incorporate more complex structures such as syntax and compositional semantics, enabling exploration of how hierarchical and recursive language features might arise. Iterated learning models focus on the cultural transmission of language across generations. Unlike agent-based models that primarily consider synchronous interactions, iterated learning models simulate a sequential process in which each generation learns the language from the data produced by the previous one. This framework captures how learning biases and constraints shape language structure over time, filtering linguistic variants through the bottleneck of acquisition. Bayesian learning algorithms are often employed to model how learners infer underlying language rules from limited and noisy data. Notably, experiments in iterated learning have demonstrated the spontaneous emergence of compositionality—an essential property of human language— over repeated transmission cycles. These models provide insights into the evolutionary pressures that promote the regularization and complexity of language. 91 INNOVATIVE RESE