Supervised Seeded Iterated Learning for Interactive Language Learning

TL;DR

This paper introduces Supervised Seeded Iterated Learning, a novel approach combining existing methods to effectively prevent language drift in interactive language models, demonstrated through a language-drift translation game.

Contribution

It proposes a new training method that merges Supervised Selfplay and Seeded Iterated Learning to address weaknesses in existing approaches for reducing language drift.

Findings

Supervised Seeded Iterated Learning reduces language drift effectively.

The method outperforms existing approaches in the language-drift translation game.

It minimizes late-stage training collapses and negative likelihood issues.

Abstract

Language drift has been one of the major obstacles to train language models through interaction. When word-based conversational agents are trained towards completing a task, they tend to invent their language rather than leveraging natural language. In recent literature, two general methods partially counter this phenomenon: Supervised Selfplay (S2P) and Seeded Iterated Learning (SIL). While S2P jointly trains interactive and supervised losses to counter the drift, SIL changes the training dynamics to prevent language drift from occurring. In this paper, we first highlight their respective weaknesses, i.e., late-stage training collapses and higher negative likelihood when evaluated on human corpus. Given these observations, we introduce Supervised Seeded Iterated Learning to combine both methods to minimize their respective weaknesses. We then show the effectiveness of \algo in the language-drift translation game.