# Natural Language Does Not Emerge 'Naturally' in Multi-Agent Dialog

## TL;DR

This paper critically examines the emergence of natural language in multi-agent communication, revealing that most agent-developed languages are effective but not interpretable or compositional, and discusses ways to make them more human-like.

## Contribution

It provides a systematic analysis showing that natural language does not naturally emerge in multi-agent systems and explores methods to induce more human-like communication.

## Key findings

- Most agent languages are effective but not interpretable
- Natural language does not naturally emerge in multi-agent settings
- Restricting communication can lead to more human-like languages

## Abstract

A number of recent works have proposed techniques for end-to-end learning of communication protocols among cooperative multi-agent populations, and have simultaneously found the emergence of grounded human-interpretable language in the protocols developed by the agents, all learned without any human supervision! In this paper, using a Task and Tell reference game between two agents as a testbed, we present a sequence of 'negative' results culminating in a 'positive' one -- showing that while most agent-invented languages are effective (i.e. achieve near-perfect task rewards), they are decidedly not interpretable or compositional. In essence, we find that natural language does not emerge 'naturally', despite the semblance of ease of natural-language-emergence that one may gather from recent literature. We discuss how it is possible to coax the invented languages to become more and more human-like and compositional by increasing restrictions on how two agents may communicate.