[Bouchacourt and Baroni, 2019]

"Miss Tools and Mr Fruit: Emergent communication in agents learning about object affordances."

⁠

arXiv preprint arXiv:1905.11871 (2019)⁠

// EXTRACTS WITH LIGHT EDITS

Contribution

Introduce a game meeting many desiderata for a natural communication environment

We propose here a new task capturing crucial aspects of the human environment, such as natural object affordances, and of human conversation, such as full symmetry among the participants

Two-pronged analysis of emerging communication, at the pragmatic and semantic levels

full symmetry is not a sufficient condition for a common language to emerge

⁠

Findings

Enabling the agents to communicate greatly increases performance compared to the no-communication ablation, both with and without memory, despite the high baseline set by agents that learn about tool usefulness without communicating

Is memory ablation necessary for communication to matter?

Did they however developed shared conventions to refer to them, as in human language?

Genuine bi-lateral communication helps them to reach higher reward

Even perfectly symmetric agents converge to distinct idiolects instead of developing a single, shared code

Setup

Picked 16 tool and 31 fruit categories

The agents are fully symmetric and cannot specialize to a fixed role or turn-taking scheme. The number of turns is open and determined by the agents.

Reward depends instead on tool and fruit affordances.

For example, if the majority of fruits requires to be cut, a knife is intrinsically more useful than a spoon.

Message Effect is computed on single turns and uses causal theory (Pearl et al., 2016) to quantify how much what an agent utters impacts the other, compared to the counterfactual scenario in which the speaking agent said something else.

Conclusions

Our experiments show that these agents do develop genuine communication even when

Successful communication per se is not directly rewarded;

The observable environment already contains stable, reliable information helping to solve the task (object affordances); and

The agents are not artificially forced to rely on communication by erasing their memory.

Something like ritual might be needed after all

Graesser et al. (2019), who study a simple signaling game, similarly conclude that training single pairs of agents does not lead to the emergence of a common language, which requires diffusion in larger communities

equipping the agents with a feedback loop in which they also receive their own messages as input might encourage shared codes across speaker and listener roles

Want to print your doc?
This is not the way.

Try clicking the ⋯ next to your doc name or using a keyboard shortcut (

CtrlP

) instead.