Automatic Grammatical Error Correction for Sequence-to-sequence Text Generation: An Empirical Study

Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics, pages 6059–6064 Florence, Italy, July 28 - August 2, 2019. c⃝2019 Association for Computational Linguistics 6059 Automatic Grammatical Error Correction for Sequence-to-sequence Text Generation: An Empirical Study Tao Ge Xingxing Zhang Furu Wei Ming Zhou Microsoft Research Asia {tage, xizhang, fuwei, mingzhou}@microsoft.com Abstract Sequence-to-sequence (seq2seq) models have achieved tremendous success in text genera- tion tasks. However, there is no guarantee that they can always generate sentences without grammatical errors. In this paper, we present a preliminary empirical study on whether and how much automatic grammatical error cor- rection can help improve seq2seq text genera- tion. We conduct experiments across various seq2seq text generation tasks including ma- chine translation, formality style transfer, sen- tence compression and simplification. Exper- iments show the state-of-the-art grammatical error correction system can improve the gram- maticality of generated text and can bring task- oriented improvements in the tasks where tar- get sentences are in a formal style. 1 Introduction Sequence-to-sequence (seq2seq) text generation (Cho et al., 2014; Sutskever et al., 2014) has attracted growing attention in natural language processing (NLP). Despite various advantages of seq2seq models, they tend to have a weakness: there is no guarantee that they can always gener- ate sentences without grammatical errors. Table 1 shows examples generated by seq2seq models in various tasks with grammatical errors. One valid solution to this challenge is conduct- ing grammatical error correction (GEC) for ma- chine generated sentences. Recent GEC systems (Chollampatt and Ng, 2018; Junczys-Dowmunt et al., 2018; Grundkiewicz and Junczys-Dowmunt, 2018; Ge et al., 2018a,b) can achieve human-level performance in GEC benchmarks. We are curi- ous whether they can help improve seq2seq based natural language generation (NLG) models. We therefore propose an empirical study on GEC post editing for various text generation tasks (i.e., ma- chine translation, style transfer, sentence compres- Tasks Examples Machine Translation Das Team-Ereignis ist immer am besten. → The team event is always (the) best. Style Transfer = ) who do u thinks better? → Who do you think (is) better? Sentence Compression Mickey Rooney died yesterday age 93 at his home in Studio City, California... → Mickey Rooney died yesterday (at) age 93. Table 1: Seq2seq model outputs for German-English translation, formality style transfer and sentence com- pression. The texts in round brackets are edits by GEC. sion and simplification) using both automatic and human evaluation methods. Experimental results demonstrate that a state-of-the-art GEC system is helpful for improving the grammaticality of gen- erated text and that it can bring task-oriented im- provements in the tasks where target sentences are in a formal style. The contributions of this paper are twofold: • We present an empirical study on GEC post editing for seq2seq text generation. To the best of our knowledge, it is the first work to study improving seq2seq based NLG models using GEC. • We show some interesting results by thor- oughly comparing and analyzing GEC post editing for various seq2seq text generation tasks, shedding …

Content truncated to 500 words