On Decoding Strategies for Neural Text Generators

TL;DR

This paper provides a comprehensive analysis of how different decoding strategies affect the quality and diversity of generated text across various natural language generation tasks, revealing task-specific behaviors and trade-offs.

Contribution

It systematically evaluates the interaction between decoding strategies and tasks, highlighting the variability and task-specific nature of decoding effects in neural text generation.

Findings

Decoding strategy effects vary significantly across tasks.

The diversity-quality trade-off is task-dependent.

Beam search's length bias is not consistent across tasks.

Abstract

When generating text from probabilistic models, the chosen decoding strategy has a profound effect on the resulting text. Yet the properties elicited by various decoding strategies do not always transfer across natural language generation tasks. For example, while mode-seeking methods like beam search perform remarkably well for machine translation, they have been observed to lead to incoherent and repetitive text in story generation. Despite such observations, the effectiveness of decoding strategies is often assessed with respect to only a single task. This work -- in contrast -- provides a comprehensive analysis of the interaction between language generation tasks and decoding strategies. Specifically, we measure changes in attributes of generated text as a function of both decoding strategy and task using human and automatic evaluation. Our results reveal both previously-observed and surprising findings. For example, the nature of the diversity-quality trade-off in language generation is very task-specific; the length bias often attributed to beam search is not constant across tasks.