arXivDaily arXiv每日学术速递 周一至周五更新
arXiv周末暂无论文更新,休息一下吧,周末愉快~~
arXiv 2009.11136cs.CL
收藏

Seq2Edits:使用跨度级编辑操作的序列转换

Seq2Edits: Sequence Transduction Using Span-level Edit Operations

Google Research(谷歌研究院)

Felix Stahlberg, Shankar Kumar

摘要:

我们提出了Seq2Edits,一种开放词汇表的方法,用于自然语言处理(NLP)任务的序列编辑,输入和输出文本之间具有高度的重叠。在这种方法中,每个序列到序列的转换被表示为编辑操作的序列,其中每个操作或者用目标令牌替换整个源跨度或者保持其不变。我们在五个NLP任务(文本规范化,句子融合,句子拆分和重新措辞,文本简化和语法错误纠正)上评估了我们的方法,并全面报告了竞争结果。对于语法错误纠正,与全序列模型相比,我们的方法将推理速度提高了5.2倍,因为推理时间取决于编辑的数量,而不是目标标记的数量。对于文本规范化,句子融合和语法错误纠正,我们的方法通过将每个编辑操作与人类可读的标签相关联来提高可解释性。

英文摘要:

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.

机构、总结、中文摘要均由AI分析而来,可能存在错误,请以论文正文为准。

↑