--- title: "基于强化学习的古今汉语句子对齐研究 — 喻快 2025" description: "Neural machine translation between Classical (ancient) and Modern Chinese lacks large, high-quality sentence-aligned parallel data." authors: ["喻快", "邵艳秋", "李炜"] year: 2025 venue: "中文信息学报 (Journal of Chinese Information Processing)" source: "http://jcip.cipsc.org.cn/article/id/zwxxxb_3831" licence: "Free access (jcip.cipsc.org.cn), no open licence stated — readonly; © 中文信息学报" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/papers/yu-2025-sentence-alignment-ancient/" --- [Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/yu-2025-sentence-alignment-ancient/) [home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2025](https://md.research.elamur.ai/free/papers/2025/page/3/) / 基于强化学习的古今汉语句子对齐研究 # 基于强化学习的古今汉语句子对齐研究 **Authors:** 喻快, 邵艳秋, 李炜 · **Year:** 2025 · **Venue:** 中文信息学报 (Journal of Chinese Information Processing) · **Licence:** Free access (jcip.cipsc.org.cn), no open licence stated — readonly; © 中文信息学报 **Links:** [Publisher's page](http://jcip.cipsc.org.cn/article/id/zwxxxb_3831) · [PDF](http://jcip.cipsc.org.cn/cn/article/pdf/preview/zwxxxb_3831.pdf) ## Summary Neural machine translation between Classical (ancient) and Modern Chinese lacks large, high-quality sentence-aligned parallel data. Older alignment methods score sentence pairs with length, lexical, and shared-character features. They capture meaning poorly and struggle with many-to-many alignments. The authors frame alignment within already-matched paragraph pairs as an unsupervised sequential decision problem. A BERT model works as a binary matcher, using a next-sentence-prediction-style input of the form [CLS] ancient [SEP] modern. It is first warmed up on rough sentence pairs that the LCS algorithm extracts, with randomly sampled negatives added. It is then trained as a policy network inside a dynamic-programming search over 1-1, 1-2, 2-1 and 2-2 candidate merges. The training signal is a reward for covering as many sentence pairs as possible, optimized with policy gradient plus a baseline. Matches involving more than one sentence earn larger rewards, which pushes the model to notice non-1-1 patterns. On a manually aligned test set from the Twenty-Four Histories, the approach beats the strongest earlier baseline. The biggest gains are on many-to-many alignments. This provides a way to mine parallel data for low-resource historical translation. ## Key points - **Data:** About 300 articles from 《二十四史》 (Twenty-Four Histories) were crawled with their modern translations, giving roughly 9,000 hand-aligned paragraph pairs. - 200 of these paragraphs were sentence-aligned by hand, yielding 1,550 test sentence pairs labeled by alignment pattern. - The remaining paragraphs serve as training data. - **Alignment patterns:** Earlier statistics show 1-1 alignments make up about 90.12% of the data, and patterns up to 2-2 cover about 98.98%. The model therefore handles only patterns up to 2-2, and folds 1-0/0-1 cases into 1-2/2-1. - **Rewards and complexity:** A rejected match earns 0 reward, an accepted 1-1 match earns 2, and accepted 1-2, 2-1 and 2-2 matches earn 3 each. The DP runs in O(MN) time, so the method suits paragraph-aligned corpora rather than whole-document ones. - **Claimed contributions:** - first use of a pre-trained language model for ancient–modern Chinese sentence alignment; - a global RL objective that maximizes alignment coverage; - a newly built alignment corpus. - **Publication:** The PDF header shows the paper appeared in 中文信息学报 vol. 38, no. 12 (Dec 2024), pp. 30–38. It comes from Beijing Language and Culture University. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/) · [Natural language processing](https://md.research.elamur.ai/free/shelves/nlp/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @article{2025, title = {{基于强化学习的古今汉语句子对齐研究}}, author = {喻快 and 邵艳秋 and 李炜}, journal = {中文信息学报 (Journal of Chinese Information Processing)}, year = {2025}, url = {http://jcip.cipsc.org.cn/article/id/zwxxxb_3831} } ``` - [← previous 基于大语言模型的自动代码修复综述](https://md.research.elamur.ai/free/papers/xu-2025-large-language-model/) - [next → 基于正负例思维链的表格-文本混合金融数据自动问答方法](https://md.research.elamur.ai/free/papers/li-2025-positive-negative-chain/) - [↑ back to papers of 2025](https://md.research.elamur.ai/free/papers/2025/page/3/) ## References 5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Pr o c e ed i ngso ft he55 t hAnnua lMe e t i ngo ft heAs s o - c i a t i onf o rCompu t a t i ona lL i ngu i s t i c s, 2017:562 - 570. [find](https://scholar.google.com/scholar?q=Pr+o+c+e+ed+i+ngso+ft+he55+t+hAnnua+lMe+e+t+i+ngo+ft+heAs+s+o+%2D+c+i+a+t+i+onf+o+rCompu+t+a+t+i+ona+lL+i+ngu+i+s+t+i+c+s%2C+2017%3A562+%2D+570%2E) 2. [ 21] DEVLINJ, CHANG M W, LEE K, e ta l.Be r t:Pr e - t r a i n i ng o fde ep b i d i r e c t i ona lt r ans f o rme r sf o rl a n - //Pr a eund e r s t a nd i ng[ C] o c e e d i ng so ft h e NAACL, gu g 2014:4171 - 4186. [find](https://scholar.google.com/scholar?q=%5B+21%5D+DEVLINJ%2C+CHANG+M+W%2C+LEE+K%2C+e+ta+l%2EBe+r+t%3APr+e+%2D+t+r+a+i+n+i+ng+o+fde+ep+b+i+d+i+r+e+c+t+i+ona+lt+r+ans+f+o+rme+r+sf+o+rl+a+n+%2D+%2F%2FPr+a+eund+e+r+s+t+a+nd+i+ng%5B+C%5D+o+c+e+e+d+i+ng+so+ft+h+e+NAACL%2C+g) 3. [ ] , 22 SUN Y WANGS, LIY, e ta l.Er n i e:Enhanc edr ep - r e s en t a t i ont hr ough knowl edgei n t eg r a t i on[ J]. a rXi v r ep r i n ta rXi v: 1904. 09223, 2019. p [ 23] LIS, ZHAOZ,HU R, e ta l.Ana l og i c a lr e a s on i ngon // Ch i ne s e mo r l og i c a lands eman t i cr e l a t i ons[… [find](https://scholar.google.com/scholar?q=%5B+%5D+%2C+22+SUN+Y+WANGS%2C+LIY%2C+e+ta+l%2EEr+n+i+e%3AEnhanc+edr+ep+%2D+r+e+s+en+t+a+t+i+ont+hr+ough+knowl+edgei+n+t+eg+r+a+t+i+on%5B+J%5D%2E+a+rXi+v+r+ep+r+i+n+ta+rXi+v%3A+1904%2E+09223%2C+2019%2E+p+%5B+23%5D+LIS%2C+ZHAOZ%2CHU+R%2C+e+ta) 4. [ 24] KINGMA DP, BAJ.Adam:A me t hodf o rs t o cha s t i c [ ] / / op t imi z a t i on C Pr o c e ed i ngso ft he3r dI n t e r na t i on - a lCon f e r enc ef o rLe a r n i ngRep r e s en t a t i ons,2015. [find](https://scholar.google.com/scholar?q=%5B+24%5D+KINGMA+DP%2C+BAJ%2EAdam%3AA+me+t+hodf+o+rs+t+o+cha+s+t+i+c+%5B+%5D+%2F+%2F+op+t+imi+z+a+t+i+on+C+Pr+o+c+e+ed+i+ngso+ft+he3r+dI+n+t+e+r+na+t+i+on+%2D+a+lCon+f+e+r+enc+ef+o+rLe+a+r+n+i+ngRep+r+e+s+en+t+a+t+i+ons%2C) 5. 谢文秀( 1993—), 博 士 研 究 生, 主 要 研 究 领 域 为 问答、 机 器 学 习、 自 然 语 言 处 理、 健 康 信 息 学、 公 [find](https://scholar.google.com/scholar?q=%E8%B0%A2%E6%96%87%E7%A7%80%28+1993%E2%80%94%29%2C+%E5%8D%9A+%E5%A3%AB+%E7%A0%94+%E7%A9%B6+%E7%94%9F%2C+%E4%B8%BB+%E8%A6%81+%E7%A0%94+%E7%A9%B6+%E9%A2%86+%E5%9F%9F+%E4%B8%BA+%E9%97%AE%E7%AD%94%E3%80%81+%E6%9C%BA+%E5%99%A8+%E5%AD%A6+%E4%B9%A0%E3%80%81+%E8%87%AA+%E7%84%B6+%E8%AF%AD+%E8%A8%80+%E5%A4%84+%E7%90%86%E3%80%81+%E5%81%A5+%E5%BA%B7+%E4%BF%A1+%E6%81%AF+%E5%AD%A6%E3%80%81+%E5%85%AC)