Revisiting the Architectures like Pointer Networks to Efficiently Improve the Next Word Distribution, Summarization Factuality, and Beyond

TL;DR

This paper challenges the dominance of softmax in language models, proposing efficient pointer network alternatives that improve next word prediction and summarization factuality without significant speed loss.

Contribution

It introduces simplified pointer network-based methods as effective softmax alternatives, enhancing language model performance and factual accuracy in summarization tasks.

Findings

Outperforms mixture of softmax in GPT-2

Improves factCC scores by 2 points in CNN/DM and XSUM

Increases MAUVE scores by 30% in BookSum

Abstract

Is the output softmax layer, which is adopted by most language models (LMs), always the best way to compute the next word probability? Given so many attention layers in a modern transformer-based LM, are the pointer networks redundant nowadays? In this study, we discover that the answers to both questions are no. This is because the softmax bottleneck sometimes prevents the LMs from predicting the desired distribution and the pointer networks can be used to break the bottleneck efficiently. Based on the finding, we propose several softmax alternatives by simplifying the pointer networks and accelerating the word-by-word rerankers. In GPT-2, our proposals are significantly better and more efficient than mixture of softmax, a state-of-the-art softmax alternative. In summarization experiments, without significantly decreasing its training/testing speed, our best method based on T5-Small improves factCC score by 2 points in CNN/DM and XSUM dataset, and improves MAUVE scores by 30% in BookSum paragraph-level dataset.