Imagination is All You Need! Curved Contrastive Learning for Abstract Sequence Modeling Utilized on Long Short-Term Dialogue Planning

TL;DR

This paper introduces Curved Contrastive Learning (CCL), a novel space-time inspired method for representing multi-turn dialogues, enabling zero-shot response ranking and future sequence likelihood assessment with improved efficiency.

Contribution

We propose CCL, a curvature-inspired contrastive learning technique that models turn distances in dialogues, facilitating zero-shot response ranking and sequence likelihood estimation.

Findings

Achieves higher efficiency by encoding only the last utterance during inference.

Demonstrates planning capability over multiple turns in dialogues.

Shows improved response ranking performance in zero-shot settings.

Abstract

Inspired by the curvature of space-time (Einstein, 1921), we introduce Curved Contrastive Learning (CCL), a novel representation learning technique for learning the relative turn distance between utterance pairs in multi-turn dialogues. The resulting bi-encoder models can guide transformers as a response ranking model towards a goal in a zero-shot fashion by projecting the goal utterance and the corresponding reply candidates into a latent space. Here the cosine similarity indicates the distance/reachability of a candidate utterance toward the corresponding goal. Furthermore, we explore how these forward-entailing language representations can be utilized for assessing the likelihood of sequences by the entailment strength i.e. through the cosine similarity of its individual members (encoded separately) as an emergent property in the curved space. These non-local properties allow us to imagine the likelihood of future patterns in dialogues, specifically by ordering/identifying future goal utterances that are multiple turns away, given a dialogue context. As part of our analysis, we investigate characteristics that make conversations (un)plannable and find strong evidence of planning capability over multiple turns (in 61.56% over 3 turns) in conversations from the DailyDialog (Li et al., 2017) dataset. Finally, we show how we achieve higher efficiency in sequence modeling tasks compared to previous work thanks to our relativistic approach, where only the last utterance needs to be encoded and computed during inference.