Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Simulating the Emergence of Differential Case Marking with Communicating Neural-Network Agents

Yuchen Lian, Arianna Bisazza, Tessa Verhoef

TL;DR

This study investigates the emergence of Differential Case Marking (DCM) using NeLLCom-X neural agents that first learn a miniature artificial language and then engage in communicative interactions. It shows that DCM-like patterns do not arise from learning alone but emerge when agents communicate, aligning with human studies that emphasize interaction pressures. The work contrasts Dominant and Neutral word-order languages to reveal how typicality and order influence marker usage and demonstrates that neural agents can reproduce human-like DOM effects under communication, while highlighting biases distinct from human learners. Overall, NeLLCom-X provides a scalable, controllable framework to complement experimental research on language evolution and the emergence of case marking.

Abstract

Differential Case Marking (DCM) refers to the phenomenon where grammatical case marking is applied selectively based on semantic, pragmatic, or other factors. The emergence of DCM has been studied in artificial language learning experiments with human participants, which were specifically aimed at disentangling the effects of learning from those of communication (Smith & Culbertson, 2020). Multi-agent reinforcement learning frameworks based on neural networks have gained significant interest to simulate the emergence of human-like linguistic phenomena. In this study, we employ such a framework in which agents first acquire an artificial language before engaging in communicative interactions, enabling direct comparisons to human result. Using a very generic communication optimization algorithm and neural-network learners that have no prior experience with language or semantic preferences, our results demonstrate that learning alone does not lead to DCM, but when agents communicate, differential use of markers arises. This supports Smith and Culbertson (2020)'s findings that highlight the critical role of communication in shaping DCM and showcases the potential of neural-agent models to complement experimental research on language evolution.

Simulating the Emergence of Differential Case Marking with Communicating Neural-Network Agents

TL;DR

This study investigates the emergence of Differential Case Marking (DCM) using NeLLCom-X neural agents that first learn a miniature artificial language and then engage in communicative interactions. It shows that DCM-like patterns do not arise from learning alone but emerge when agents communicate, aligning with human studies that emphasize interaction pressures. The work contrasts Dominant and Neutral word-order languages to reveal how typicality and order influence marker usage and demonstrates that neural agents can reproduce human-like DOM effects under communication, while highlighting biases distinct from human learners. Overall, NeLLCom-X provides a scalable, controllable framework to complement experimental research on language evolution and the emergence of case marking.

Abstract

Differential Case Marking (DCM) refers to the phenomenon where grammatical case marking is applied selectively based on semantic, pragmatic, or other factors. The emergence of DCM has been studied in artificial language learning experiments with human participants, which were specifically aimed at disentangling the effects of learning from those of communication (Smith & Culbertson, 2020). Multi-agent reinforcement learning frameworks based on neural networks have gained significant interest to simulate the emergence of human-like linguistic phenomena. In this study, we employ such a framework in which agents first acquire an artificial language before engaging in communicative interactions, enabling direct comparisons to human result. Using a very generic communication optimization algorithm and neural-network learners that have no prior experience with language or semantic preferences, our results demonstrate that learning alone does not lead to DCM, but when agents communicate, differential use of markers arises. This supports Smith and Culbertson (2020)'s findings that highlight the critical role of communication in shaping DCM and showcases the potential of neural-agent models to complement experimental research on language evolution.

Paper Structure

This paper contains 18 sections, 1 figure, 2 tables.

Table of Contents

  1. Introduction
  2. Differential Case Marking
  3. NeLLCom Framework
  4. The Task
  5. Agent Architecture
  6. Training
  7. Experimental Setup
  8. Meaning Space
  9. Artificial Languages
  10. Evaluation
  11. Model Configuration and Training Details
  12. Results
  13. Dominant Order Language
  14. Effects of communication
  15. Comparison to human results
  16. ...and 3 more sections

Figures (1)

  • Figure 1: Agent communication results for the three initial languages. Column 1: Meaning reconstruction accuracy across communication rounds, computed on the whole test set (orange line), as well as split by non-ambiguous (green) and non-ambiguous (blue) meanings. Col. 2-4: Production preferences (PP) in terms of order proportions and marker use. Specifically, Col. 2 and 3 show PP for non-ambiguous and ambiguous meanings respectively, before and after communication, and Col. 4 shows the difference in PP between $M_{test,\ \neg Amb}$ and $M_{test,\ Amb}$ after communication. Solid diamonds mark the initial language. Each empty circle is an agent and solid circles are the average of all agents, with error bars showing standard deviation. Each experiment is repeated with 50 agent pairs.