Abstract:The evolution of symbolic communication is a longstanding open research question in biology. While some theories suggest that it originated from sub-symbolic communication (i.e., iconic or indexical), little experimental evidence exists on how organisms can actually evolve to define a shared set of symbols with unique interpretable meaning, thus being capable of encoding and decoding discrete information. Here, we use a simple synthetic model composed of sender and receiver agents controlled by Continuous-Time Recurrent Neural Networks, which are optimized by means of neuro-evolution. We characterize signal decoding as either regression or classification, with limited and unlimited signal amplitude. First, we show how this choice affects the complexity of the evolutionary search, and leads to different levels of generalization. We then assess the effect of noise, and test the evolved signaling system in a referential game. In various settings, we observe agents evolving to share a dictionary of symbols, with each symbol spontaneously associated to a 1-D unique signal. Finally, we analyze the constellation of signals associated to the evolved signaling systems and note that in most cases these resemble a Pulse Amplitude Modulation system.

What problem does this paper attempt to address?

The problem that this paper attempts to solve is about the evolutionary mechanism of symbolic communication. Specifically, the authors constructed a synthetic model consisting of senders and receivers. These agents are controlled by continuous - time recurrent neural networks (CTRNNs) and optimized by neuro - evolutionary algorithms such as NEAT. The main objective of the study was to explore how, under different conditions, these agents could evolve a shared symbol system, with each symbol spontaneously associated with a unique one - dimensional signal, thereby enabling the encoding and decoding of discrete information. In addition, the paper also analyzed the constellations of signals and found that in most cases, these signal systems are similar to Pulse Amplitude Modulation (PAM) systems. ### Research Background - **The Origin of Symbolic Communication**: Language is considered the last major transition in the evolutionary process, and language itself is regarded as the apex of the evolutionary history from sub - symbolic communication in the form of icons/indicators to symbolic communication. - **Research Challenges**: Revealing this history and collecting biological evidence are extremely difficult. Therefore, studying the evolution of symbolic communication through computational models (such as artificial - life simulations) has become an important step in understanding the origin of language. ### Main Contributions 1. **Representation of Signal Decoding**: The paper represents signal decoding as a regression or classification task and explores the impact of finite and infinite signal amplitudes on the complexity of the evolutionary search and the level of generalization. 2. **In - depth Analysis of Signal Systems**: The authors conducted a detailed analysis of the evolved signal systems, including signal constellations and cluster analysis, demonstrating how different types of signals emerged from evolution. ### Experimental Setup - **Agent Model**: Senders and receivers are controlled by CTRNNs and optimized by the NEAT algorithm. - **Experimental Conditions**: Four different settings were studied: regression (finite/infinite amplitude) and classification (finite/infinite amplitude). - **Evaluation Metrics**: The performance of the model was evaluated by calculating the correct recognition rate of concepts by the agents. ### Experimental Results - **Evolution of Symbolic Communication**: Under different conditions, agents were able to evolve a shared symbol system, with each symbol associated with a unique one - dimensional signal. - **Impact of Noise**: After introducing Gaussian noise, the impact of different noise levels on communication performance was studied. The results showed that in some cases, a higher noise level actually helps in signal separation and improves the robustness of communication. - **Zero - shot Communication**: In the case of regression tasks with unlimited amplitudes, agents were able to effectively perform zero - shot communication, that is, correctly convey unseen concepts. ### Conclusion By constructing and optimizing a synthetic model, the paper successfully demonstrated how agents can evolve symbolic communication systems under different conditions and conducted an in - depth analysis of the properties of these systems. The research results not only enhance the understanding of the evolutionary mechanism of symbolic communication but also provide a valuable reference for future research.

A Signal-Centric Perspective on the Evolution of Symbolic Communication