Abstract: Cross-lingual summarization (CLS) is the task of generating a summary in a target language from a document in a source language. Recently, end-to-end CLS models have achieved impressive results using large-scale, high-quality datasets typically constructed by translating monolingual summary corpora into CLS corpora. However, due to the limited performance of low-resource language translation models, translation noise can seriously degrade the performance of these models. In this paper, we propose a fine-grained reinforcement learning approach to address low-resource CLS based on noisy data. We introduce the source language summary as a gold signal to alleviate the impact of the translated noisy target summary. Specifically, we design a reinforcement reward by calculating the word correlation and word missing degree between the source language summary and the generated target language summary, and combine it with cross-entropy loss to optimize the CLS model. To validate the performance of our proposed model, we construct Chinese-Vietnamese and Vietnamese-Chinese CLS datasets. Experimental results show that our proposed model outperforms the baselines in terms of both the ROUGE score and BERTScore.

基于细粒度强化学习增强噪声数据的低资源跨语言摘要

[1]Ayana, Shen SQ, Chen Y, et al., 2018. Zero-shot cross-lingual neural headline generation. IEEE/ACM Trans Audio Speech Lang Process, 26(12):2319-2327.

[2]Bai Y, Gao Y, Huang HY, 2021. Cross-lingual abstractive summarization with limited parallel resources. Proc 59^th Annual Meeting of the Association for Computational Linguistics and the 11^th Int Joint Conf on Natural Language Processing, p.6910-6924.

[3]Böhm F, Gao Y, Meyer CM, et al., 2019. Better rewards yield better summaries: learning to summarise without references. Proc Conf on Empirical Methods in Natural Language Processing and the 9^th Int Joint Conf on Natural Language Processing, p.3110-3120.

[4]Cao Y, Liu H, Wan XJ, 2020. Jointly learning to align and summarize for neural cross-lingual summarization. Proc 58^th Annual Meeting of the Association for Computational Linguistics, p.6220-6231.

[5]Dou ZY, Kumar S, Tsvetkov Y, 2020. A deep reinforced model for zero-shot cross-lingual summarization with bilingual semantic similarity rewards. Proc 4^th Workshop on Neural Generation and Translation, p.60-68.

[6]Dyer C, Chahuneau V, Smith NA, 2013. A simple, fast, and effective reparameterization of IBM Model 2. Proc Conf of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, p.644-648.

[7]Hermann KM, Kočiský T, Grefenstette E, et al., 2015. Teaching machines to read and comprehend. Proc 28^th Int Conf on Neural Information Processing Systems, p.1693-1701.

[8]Hu BT, Chen QC, Zhu FZ, 2015. LCSTS: a large scale Chinese short text summarization dataset. Proc Conf on Empirical Methods in Natural Language Processing, p.1967-1972.

[9]Javed A, Ali Khan A, 2022. Shot classification and replay detection for sports video summarization. Front Inform Technol Electron Eng, 23(5):790-800.

[10]Jiang SY, Tu DB, Chen XS, et al., 2022. ClueGraphSum: let key clues guide the cross-lingual abstractive summarization. https://arxiv.org/abs/2203.02797

[11]Kang XM, Zhao Y, Zhang JJ, et al., 2020. Dynamic context selection for document-level neural machine translation via reinforcement learning. Proc Conf on Empirical Methods in Natural Language Processing, p.2242-2254.

[12]Kim S, Jang JY, Jung M, et al., 2021. A model of cross-lingual knowledge-grounded response generation for open-domain dialogue systems. Findings of the Association for Computational Linguistics, p.352-365.

[13]Kumar G, Foster G, Cherry C, et al., 2019. Reinforcement learning based curriculum optimization for neural machine translation. Proc Conf of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, p.2054-2061.

[14]Lai H, Gao YM, Huang YX, et al., 2022. Evaluation method of text generation based on multi-granularity feature. J Chin Inform Process, 36(3):45-53,63(in Chinese).

[15]Leuski A, Lin CY, Zhou L, et al., 2003. Cross-lingual C*ST*RD: English access to Hindi information. ACM Trans Asian Lang Inform Process, 2(3):245-269.

[16]Li HQ, Huang J, Cao Z, et al., 2023. Stochastic pedestrian avoidance for autonomous vehicles using hybrid reinforcement learning. Front Inform Technol Electron Eng, 24(1):131-140.

[17]Li P, Tang C, Xu XH, 2021. Video summarization with a graph convolutional attention network. Front Inform Technol Electron Eng, 22(6):902-913.

[18]Liang YL, Meng FD, Zhou CL, et al., 2022. A variational hierarchical model for neural cross-lingual summarization. Proc 60^th Annual Meeting of the Association for Computational Linguistics, p.2088-2099.

[19]Lim JM, Kang IS, Lee JH, 2004. Multi-document summarization using cross-language texts. Proc NTCIR-4.

[20]Lin CY, 2004. ROUGE: a package for automatic evaluation of summaries. Proc 4^th Workshop on Annual Meeting of the Association for Computational Linguistics, p.74-81.

[21]Nguyen TT, Luu AT, 2022. Improving neural cross-lingual abstractive summarization via employing optimal transport distance for knowledge distillation. Proc 36^th AAAI Conf on Artificial Intelligence, 36(10):11103-11111.

[22]Orǎsan C, Chiorean OA, 2008. Evaluation of a cross-lingual Romanian-English multi-document summariser. Proc Int Conf on Language Resources and Evaluation.

[23]Ouyang J, Song BY, McKeown K, 2019. A robust abstractive system for cross-lingual summarization. Proc Conf of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, p.2025-2031.

[24]Paulus R, Xiong CM, Socher R, 2017. A deep reinforced model for abstractive summarization. https://arxiv.org/abs/1705.04304

[25]Rennie SJ, Marcheret E, Mroueh Y, et al., 2017. Self-critical sequence training for image captioning. Proc IEEE Conf on Computer Vision and Pattern Recognition, p.1179-1195.

[26]Rippeth E, Post M, 2022. Additive interventions yield robust multi-domain machine translation models. Proc 7^th Conf on Machine Translation, p.220-232.

[27]Takase S, Okazaki N, 2020. Multi-task learning for cross-lingual abstractive summarization. https://arxiv.org/abs/2010.07503

[28]Unanue IJ, Parnell J, Piccardi M, 2021. BERTTune: fine-tuning neural machine translation with BERTScore. https://arxiv.org/abs/2106.02208

[29]Vaswani A, Shazeer N, Parmar N, et al., 2017. Attention is all you need. Proc 31^st Int Conf on Neural Information Processing Systems, p.6000-6010.

[30]Wang JA, Meng FD, Lu ZY, et al., 2022. ClidSum: a benchmark dataset for cross-lingual dialogue summarization. Proc Conf on Empirical Methods in Natural Language Processing, p.7716-7729.

[31]Wu LJ, Zhu JH, He D, et al., 2019. Machine translation with weakly paired documents. Proc Conf on Empirical Methods in Natural Language Processing and the 9^th Int Joint Conf on Natural Language Processing, p.4375-4384.

[32]Xiong LL, Tang Y, Liu CS, et al., 2023. A home energy management approach using decoupling value and policy in reinforcement learning. Front Inform Technol Electron Eng, 24(9):1261-1272.

[33]Yoon W, Yeo YS, Jeong M, et al., 2021. Learning by semantic similarity makes abstractive summarization better. https://arxiv.org/abs/2002.07767

[34]You YJ, Jia WJ, Liu TY, et al., 2019. Improving abstractive document summarization with salient information modeling. Proc 57^th Annual Meeting of the Association for Computational Linguistics, p.2132-2141.

[35]Zhang TY, Kishore V, Wu F, et al., 2020. BERTScore: evaluating text generation with BERT. https://arxiv.org/abs/1904.09675

[36]Zhao H, Xie J, Lv Y, et al., 2013. Common error analysis of machine translation output. The 9^th China Workshop on Machine Translation.

[37]Zhao J, Zhao YP, Wang WX, et al., 2022. Coach-assisted multi-agent reinforcement learning framework for unexpected crashed agents. Front Inform Technol Electron Eng, 23(7):1032-1042.

[38]Zhou J, Ke P, Qiu XP, et al., 2023. ChatGPT: potential, prospects, and limitations. Front Inform Technol Electron Eng, early access.

[39]Zhu JN, Wang Q, Wang YN, et al., 2019. NCLS: neural cross-lingual summarization. Proc Conf on Empirical Methods in Natural Language Processing and the 9^th Int Joint Conf on Natural Language Processing, p.3054-3064.