Chinese Sentence Paraphrasing

Anonymous

Abstract: Sentence paraphrasing involves understanding the semantics and generating alternative expressions that are equivalent to the original sentence but not identical. However, there lack of an evaluation metric for paraphrasing that aligns well with human annotation and a lack of high-quality Chinese paraphrase datasets which makes it difficult to train a Chinese paraphrase model. To address these challenges, we present the first large-scale automatically constructed Chinese sentence paraphrase corpus, consisting of 9.45 million annotated sentence pairs for paraphrasing. We also introduce a core dataset with 2.5 thousand Chinese sentence pairs that are completely paraphrased by the crowd and annotated by experts. With this high-quality data, we establish an automatic evaluation metric for Chinese paraphrasing evaluation, achieving a Spearman coefficient of 0.726 in human-annotated data and significantly outperforming existing metrics. Additionally, we build a strong baseline for Chinese paraphrasing generation with few entity and logical errors while preserving the meaning of the sentence and generating diverse and innovative sentences.

Paper Type: long

Research Area: NLP Applications

Contribution Types: NLP engineering experiment

Languages Studied: Chinese

Preprint Status: There is no non-anonymous preprint and we do not intend to release one.

A1: yes

A1 Elaboration For Yes Or No: 8

A2: yes

A2 Elaboration For Yes Or No: 8

A3: yes

A3 Elaboration For Yes Or No: 1

B: yes

B1: yes

B1 Elaboration For Yes Or No: 2,3,5

B2: yes

B2 Elaboration For Yes Or No: 1

B3: yes

B3 Elaboration For Yes Or No: 3,5

B4: yes

B4 Elaboration For Yes Or No: 3

B5: yes

B5 Elaboration For Yes Or No: 4,5

B6: yes

B6 Elaboration For Yes Or No: 5

C: yes

C1: no

C1 Elaboration For Yes Or No: The model is open source and the training process only requires the most basic computing resources and a few hours.

C2: yes

C2 Elaboration For Yes Or No: 5

C3: yes

C3 Elaboration For Yes Or No: 5

C4: yes

C4 Elaboration For Yes Or No: 4,5

D: yes

D1: yes

D1 Elaboration For Yes Or No: 2

D2: yes

D2 Elaboration For Yes Or No: 2

D3 Elaboration For Yes Or No: 2

D4: no

D4 Elaboration For Yes Or No: Not involving ethical review.

D5: yes

D5 Elaboration For Yes Or No: 2

E: yes

E1: yes

E1 Elaboration For Yes Or No: 2