Emergence of Continuous Signals As Shared Symbols Through Emergent Communication

Abstract

Humans create shared symbols through interactions with others and the environment, enabling effective communication within a specific community. This process, known as emergent communication, facilitates the exchange of intentions and meanings. Furthermore, humans combine phonemes, the smallest units of sounds, to create a myriad of continuous symbols, a property called compositionality. The generation and recognition capabilities of these speech signals are acquired through interactions with others. Although existing studies have promoted the emergence of discrete symbols representing observed objects through two agents, the emergent creation of continuous signals with compositionality is yet to be studied. This study proposes a novel probabilistic generative model that integrates the Metropolis-Hastings naming game (MHNG) and Gaussian process hidden semi-Markov model (GP-HSMM), allowing the emergence of continuous symbols. MHNG allows symbol emergence through communication between independent individuals without directly observing the internal states of others. Moreover, GP-HSMM supports the unsupervised segmentation of continuous signals and acquires continuous signals with compositionality as shared symbols. We conducted experiments using the upper and lower portions of MNIST as the observed images to evaluate the efficacy of our proposed approach. The results demonstrate that the agents successfully generate and recognize shared continuous signals representing MNIST digits despite possessing different internal representations.

Cite this work

No public copy is stored for this record, so only the metadata is cited.

Stored only when both papers are in the corpus — 5 citations exist upstream.

Papers that cite the same references — the relational fallback while embeddings are still backfilling.