Seq2Edits: Sequence Transduction Using Span-level Edit Operations

TL;DR

Seq2Edits introduces a sequence editing method using span-level operations for NLP tasks, enhancing speed and interpretability, especially in tasks with high input-output overlap.

Contribution

The paper presents a novel span-level edit operation framework for sequence transduction, improving efficiency and explainability over traditional sequence-to-sequence models.

Findings

Achieves competitive results across five NLP tasks.

Speeds up grammatical error correction inference by up to 5.2x.

Enhances interpretability with human-readable edit tags.

Abstract

We propose Seq2Edits, an open-vocabulary approach to sequence editing for natural language processing (NLP) tasks with a high degree of overlap between input and output texts. In this approach, each sequence-to-sequence transduction is represented as a sequence of edit operations, where each operation either replaces an entire source span with target tokens or keeps it unchanged. We evaluate our method on five NLP tasks (text normalization, sentence fusion, sentence splitting & rephrasing, text simplification, and grammatical error correction) and report competitive results across the board. For grammatical error correction, our method speeds up inference by up to 5.2x compared to full sequence models because inference time depends on the number of edits rather than the number of target tokens. For text normalization, sentence fusion, and grammatical error correction, our approach improves explainability by associating each edit operation with a human-readable tag.