Context-dependent communication under environmental constraints

TL;DR

This work investigates how context-dependent communication can emerge when environmental constraints restrict referent choice and a vocabulary size cost is imposed. Using a contextual variant of the Lewis signalling game with neural agents, the authors show that granting the sender access to context and penalising vocabulary size yields partly ambiguous signals that still achieve high accuracy, with disambiguation largely offloaded to the environment. When vocabulary is unlimited, signaling becomes unambiguous for both sender types, but with vocabulary penalties, a context-aware sender achieves efficient, context-dependent communication using fewer signals. The findings suggest that context, environment, and production cost jointly shape pragmatic language behavior, offering a framework for understanding real-world situated communication and informing the design of efficient embodied communication systems.

Abstract

There is significant evidence that real-world communication cannot be reduced to sending signals with context-independent meaning. In this work, based on a variant of the classical Lewis (1969) signaling model, we explore the conditions for the emergence of context-dependent communication in a situated scenario. In particular, we demonstrate that pressure to minimise the vocabulary size is sufficient for such emergence. At the same time, we study the environmental conditions and cognitive capabilities that enable contextual disambiguation of symbol meanings. We show that environmental constraints on the receiver's referent choice can be unilaterally exploited by the sender, without disambiguation capabilities on the receiver's end. Consistent with common assumptions, the sender's awareness of the context appears to be required for contextual communication. We suggest that context-dependent communication is a situated multilayered phenomenon, crucially influenced by environment properties such as distribution of contexts. The model developed in this work is a demonstration of how signals may be ambiguous out of context, but still allow for near-perfect communication accuracy.

Figures (8)

• Figure 1: The target-only (T) sender (a) and example vector encodings of object properties (b).
• Figure 2: Target-in-context (TC) sender
• Figure 3: The receiver
• Figure 4: Metric plots for Experiment 1: T and TC sender with no vocabulary penalty. The dark area around curves marks the standard error.
• Figure 5: Metric plots for Experiment 2: T and TC sender combined with 2 different levels of vocabulary penalty. The dark area around curves marks the standard error.