--- title: "基于预训练语言模型的汉语字形谜解谜模型 — 王涵玉 2026" description: "The paper addresses automatic solving of Chinese character-shape riddles (字形谜), in which the clue describes a character's components and their spatial…" authors: ["王涵玉", "梁循", "关真宇", "张森森", "牛思敏", "王孟威", "宋世超", "杨丽"] year: 2026 venue: "中文信息学报 (Journal of Chinese Information Processing)" source: "http://jcip.cipsc.org.cn/article/id/zwxxxb_4045" 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/wang-2026-chinese-character-riddle/" --- [Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/wang-2026-chinese-character-riddle/) [home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2026](https://md.research.elamur.ai/free/papers/2026/page/5/) / 基于预训练语言模型的汉语字形谜解谜模型 # 基于预训练语言模型的汉语字形谜解谜模型 **Authors:** 王涵玉, 梁循, 关真宇, 张森森, 牛思敏, 王孟威, 宋世超, 杨丽 · **Year:** 2026 · **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_4045) · [PDF](http://jcip.cipsc.org.cn/cn/article/pdf/preview/zwxxxb_4045.pdf) ## Summary The paper addresses automatic solving of Chinese character-shape riddles (字形谜), in which the clue describes a character's components and their spatial arrangement. The authors note that general LLMs handle these poorly; GPT-4, for example, gave 玥 instead of 金 for a "人 above 玉" clue. To fill the data gap, they built a character resource from two sources. Basic attributes such as strokes, pinyin and radical for about 3,500 common characters came from the mapull/chinese-dictionary project, and decompositions came from kfcd/chaizi. Riddles were then generated automatically by filling structure-specific templates with each character's components. The solver runs as a pipeline with three steps: - a BGE encoder with a small feed-forward classifier predicts the character's structure type; - a fine-tuned BART model extracts the key components; - a lookup dictionary maps the structure and components to the answer character. On a test set built from templates never seen in training, the system reaches 96.90% overall accuracy. This shows that fine-tuned pre-trained models can capture sub-character structure, with possible uses in language teaching and cultural applications. ## Key points - **Structure types:** The dataset covers five layouts (top–bottom, left–right, top–middle–bottom, left–middle–right, full enclosure). - **Split:** Data is divided 7:3 into 1,052 training and 452 test riddles. The heavy skew toward left–right and top–bottom characters is offset with template-generated pseudo-data and oversampling. - **Accuracy:** Structure prediction reaches 98.92%, keyword extraction 97.30%, and the combined answer 96.90%. Character accuracy equals the combined figure. - **Error sources:** Errors stem from rare components such as 丩, from function words being skipped (而 misread), and from confusing top–middle–bottom with top–bottom clues. - **Code:** The code is released openly on GitHub (repository "zimi"). *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{2026, title = {{基于预训练语言模型的汉语字形谜解谜模型}}, author = {王涵玉 and 梁循 and 关真宇 and 张森森 and 牛思敏 and 王孟威 and 宋世超 and 杨丽}, journal = {中文信息学报 (Journal of Chinese Information Processing)}, year = {2026}, url = {http://jcip.cipsc.org.cn/article/id/zwxxxb_4045} } ``` - [← previous 基于预训练模型的文本常识合理性估计研究综述](https://md.research.elamur.ai/free/papers/cui-2026-commonsense-plausibility-estimation/) - [next → 多元知识图谱链接预测研究综述](https://md.research.elamur.ai/free/papers/wei-2026-link-prediction-over/) - [↑ back to papers of 2026](https://md.research.elamur.ai/free/papers/2026/page/5/) ## References 11 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. [ 12] 陈潇, 王晶晶, 李寿 山, 等 . 基 于 多 通 道 BERT 的 跨 语 言 属性级情感分类方法[ J]. 中文信息学报,2022,36 [find](https://scholar.google.com/scholar?q=%5B+12%5D+%E9%99%88%E6%BD%87%2C+%E7%8E%8B%E6%99%B6%E6%99%B6%2C+%E6%9D%8E%E5%AF%BF+%E5%B1%B1%2C+%E7%AD%89+%2E+%E5%9F%BA+%E4%BA%8E+%E5%A4%9A+%E9%80%9A+%E9%81%93+BERT+%E7%9A%84+%E8%B7%A8+%E8%AF%AD+%E8%A8%80+%E5%B1%9E%E6%80%A7%E7%BA%A7%E6%83%85%E6%84%9F%E5%88%86%E7%B1%BB%E6%96%B9%E6%B3%95%5B+J%5D%2E+%E4%B8%AD%E6%96%87%E4%BF%A1%E6%81%AF%E5%AD%A6%E6%8A%A5%2C2022%2C36) 2. ( 2):121 - 128. [ 13] 胡星武, 桂韬, 张 奇, 等 . 基 于 对 比 学 习 的 表 达 式 偏 序 关系建模[ J]. 中文信息学报,2023,37( 4):166 - 174. [find](https://scholar.google.com/scholar?q=%28+2%29%3A121+%2D+128%2E+%5B+13%5D+%E8%83%A1%E6%98%9F%E6%AD%A6%2C+%E6%A1%82%E9%9F%AC%2C+%E5%BC%A0+%E5%A5%87%2C+%E7%AD%89+%2E+%E5%9F%BA+%E4%BA%8E+%E5%AF%B9+%E6%AF%94+%E5%AD%A6+%E4%B9%A0+%E7%9A%84+%E8%A1%A8+%E8%BE%BE+%E5%BC%8F+%E5%81%8F+%E5%BA%8F+%E5%85%B3%E7%B3%BB%E5%BB%BA%E6%A8%A1%5B+J%5D%2E+%E4%B8%AD%E6%96%87%E4%BF%A1%E6%81%AF%E5%AD%A6%E6%8A%A5%2C2023%2C37%28+4%29%3A166+%2D+174%2E) 3. [ 14] 桑杰端珠, 才让加 . 基于词典注入的藏汉机器翻译模型 预训练方法[ J]. 中文信息学报,2023,37( 8):43 - 51. [find](https://scholar.google.com/scholar?q=%5B+14%5D+%E6%A1%91%E6%9D%B0%E7%AB%AF%E7%8F%A0%2C+%E6%89%8D%E8%AE%A9%E5%8A%A0+%2E+%E5%9F%BA%E4%BA%8E%E8%AF%8D%E5%85%B8%E6%B3%A8%E5%85%A5%E7%9A%84%E8%97%8F%E6%B1%89%E6%9C%BA%E5%99%A8%E7%BF%BB%E8%AF%91%E6%A8%A1%E5%9E%8B+%E9%A2%84%E8%AE%AD%E7%BB%83%E6%96%B9%E6%B3%95%5B+J%5D%2E+%E4%B8%AD%E6%96%87%E4%BF%A1%E6%81%AF%E5%AD%A6%E6%8A%A5%2C2023%2C37%28+8%29%3A43+%2D+51%2E) 4. [ 15] 赵云肖, 李茹, 李 欣 杰, 等 . 基 于 汉 字 形 音 义 多 元 知 识 和标签嵌入的文本语义匹配模型[ J]. 中文信息 学 报, 2024,38( 3):42 - 55. [find](https://scholar.google.com/scholar?q=%5B+15%5D+%E8%B5%B5%E4%BA%91%E8%82%96%2C+%E6%9D%8E%E8%8C%B9%2C+%E6%9D%8E+%E6%AC%A3+%E6%9D%B0%2C+%E7%AD%89+%2E+%E5%9F%BA+%E4%BA%8E+%E6%B1%89+%E5%AD%97+%E5%BD%A2+%E9%9F%B3+%E4%B9%89+%E5%A4%9A+%E5%85%83+%E7%9F%A5+%E8%AF%86+%E5%92%8C%E6%A0%87%E7%AD%BE%E5%B5%8C%E5%85%A5%E7%9A%84%E6%96%87%E6%9C%AC%E8%AF%AD%E4%B9%89%E5%8C%B9%E9%85%8D%E6%A8%A1%E5%9E%8B%5B+J%5D%2E+%E4%B8%AD%E6%96%87%E4%BF%A1%E6%81%AF+%E5%AD%A6+%E6%8A%A5%2C+2024%2C38%28+3%29%3A42+%2D+55%2E) 5. [ 16] 刘梦迪, 梁循 . 基于 偏 旁 部 首 知 识 表 示 学 习 的 汉 字 字 形相似 度 计 算 方 法 [ J]. 中 文 信 息 学 报,2021,35 [find](https://scholar.google.com/scholar?q=%5B+16%5D+%E5%88%98%E6%A2%A6%E8%BF%AA%2C+%E6%A2%81%E5%BE%AA+%2E+%E5%9F%BA%E4%BA%8E+%E5%81%8F+%E6%97%81+%E9%83%A8+%E9%A6%96+%E7%9F%A5+%E8%AF%86+%E8%A1%A8+%E7%A4%BA+%E5%AD%A6+%E4%B9%A0+%E7%9A%84+%E6%B1%89+%E5%AD%97+%E5%AD%97+%E5%BD%A2%E7%9B%B8%E4%BC%BC+%E5%BA%A6+%E8%AE%A1+%E7%AE%97+%E6%96%B9+%E6%B3%95+%5B+J%5D%2E+%E4%B8%AD+%E6%96%87+%E4%BF%A1+%E6%81%AF+%E5%AD%A6+%E6%8A%A5%2C2021%2C35) 6. ( 12):47 - 59. 9. [ 17] 吴少扬, 余正涛, 黄 于 欣, 等 . 面 向 汉 越 跨 语 言 事 件 检 索的事件 预 训 练 方 法 [ J]. 中 文 信 息 学 报,2024,38 [find](https://scholar.google.com/scholar?q=%28+12%29%3A47+%2D+59%2E+9%2E+%5B+17%5D+%E5%90%B4%E5%B0%91%E6%89%AC%2C+%E4%BD%99%E6%AD%A3%E6%B6%9B%2C+%E9%BB%84+%E4%BA%8E+%E6%AC%A3%2C+%E7%AD%89+%2E+%E9%9D%A2+%E5%90%91+%E6%B1%89+%E8%B6%8A+%E8%B7%A8+%E8%AF%AD+%E8%A8%80+%E4%BA%8B+%E4%BB%B6+%E6%A3%80+%E7%B4%A2%E7%9A%84%E4%BA%8B%E4%BB%B6+%E9%A2%84+%E8%AE%AD+%E7%BB%83+%E6%96%B9+%E6%B3%95+%5B+J%5D%2E+%E4%B8%AD+%E6%96%87+%E4%BF%A1+%E6%81%AF+%E5%AD%A6+%E6%8A%A5%2C2024%2C38) 7. ( 4):78 - 85. [ 18] 王艺臻 . 汉 语 字 谜 理 解 的 认 知 神 经 机 制 研 究 [ D]. 杭 州:浙江大学博士学位论文,2021. [find](https://scholar.google.com/scholar?q=%28+4%29%3A78+%2D+85%2E+%5B+18%5D+%E7%8E%8B%E8%89%BA%E8%87%BB+%2E+%E6%B1%89+%E8%AF%AD+%E5%AD%97+%E8%B0%9C+%E7%90%86+%E8%A7%A3+%E7%9A%84+%E8%AE%A4+%E7%9F%A5+%E7%A5%9E+%E7%BB%8F+%E6%9C%BA+%E5%88%B6+%E7%A0%94+%E7%A9%B6+%5B+D%5D%2E+%E6%9D%AD+%E5%B7%9E%3A%E6%B5%99%E6%B1%9F%E5%A4%A7%E5%AD%A6%E5%8D%9A%E5%A3%AB%E5%AD%A6%E4%BD%8D%E8%AE%BA%E6%96%87%2C2021%2E) 8. [ 19] 卜小倩 . 体 认 语 言 学 视 域 下 汉 语 谜 语 的 认 知 机 制 研 究[ D]. 成都:四川外国语大学硕士学位论文,2021. [find](https://scholar.google.com/scholar?q=%5B+19%5D+%E5%8D%9C%E5%B0%8F%E5%80%A9+%2E+%E4%BD%93+%E8%AE%A4+%E8%AF%AD+%E8%A8%80+%E5%AD%A6+%E8%A7%86+%E5%9F%9F+%E4%B8%8B+%E6%B1%89+%E8%AF%AD+%E8%B0%9C+%E8%AF%AD+%E7%9A%84+%E8%AE%A4+%E7%9F%A5+%E6%9C%BA+%E5%88%B6+%E7%A0%94+%E7%A9%B6%5B+D%5D%2E+%E6%88%90%E9%83%BD%3A%E5%9B%9B%E5%B7%9D%E5%A4%96%E5%9B%BD%E8%AF%AD%E5%A4%A7%E5%AD%A6%E7%A1%95%E5%A3%AB%E5%AD%A6%E4%BD%8D%E8%AE%BA%E6%96%87%2C2021%2E) 9. [ 20] 朱新秤, 李瑞菊, 周治金 . 谜语问题解决中线索的作用 [ J]. 心理学报,2009,41( 5):397 - 405. [find](https://scholar.google.com/scholar?q=%5B+20%5D+%E6%9C%B1%E6%96%B0%E7%A7%A4%2C+%E6%9D%8E%E7%91%9E%E8%8F%8A%2C+%E5%91%A8%E6%B2%BB%E9%87%91+%2E+%E8%B0%9C%E8%AF%AD%E9%97%AE%E9%A2%98%E8%A7%A3%E5%86%B3%E4%B8%AD%E7%BA%BF%E7%B4%A2%E7%9A%84%E4%BD%9C%E7%94%A8+%5B+J%5D%2E+%E5%BF%83%E7%90%86%E5%AD%A6%E6%8A%A5%2C2009%2C41%28+5%29%3A397+%2D+405%2E) 10. [ 21] 李建华, 陈涛, 贾 旭 东, 等 . 中 文 大 规 模 谜 语 问 答 数 据 集构建与评测方法研究[ J]. 五邑大学学报( 自然 科 学 版),2023,37( 4):38 - 46. [find](https://scholar.google.com/scholar?q=%5B+21%5D+%E6%9D%8E%E5%BB%BA%E5%8D%8E%2C+%E9%99%88%E6%B6%9B%2C+%E8%B4%BE+%E6%97%AD+%E4%B8%9C%2C+%E7%AD%89+%2E+%E4%B8%AD+%E6%96%87+%E5%A4%A7+%E8%A7%84+%E6%A8%A1+%E8%B0%9C+%E8%AF%AD+%E9%97%AE+%E7%AD%94+%E6%95%B0+%E6%8D%AE+%E9%9B%86%E6%9E%84%E5%BB%BA%E4%B8%8E%E8%AF%84%E6%B5%8B%E6%96%B9%E6%B3%95%E7%A0%94%E7%A9%B6%5B+J%5D%2E+%E4%BA%94%E9%82%91%E5%A4%A7%E5%AD%A6%E5%AD%A6%E6%8A%A5%28+%E8%87%AA%E7%84%B6+%E7%A7%91+%E5%AD%A6+%E7%89%88%29%2C2023%2C37%28+4%29%3A38+%2D+46%2E) 11. 梁循( 1965—), 通 信 作 者, 博 士, 教 授, 博 士 生 导 师, 主要研究领域为自然语言 处 理、 商 务 智 能、 社 [find](https://scholar.google.com/scholar?q=%E6%A2%81%E5%BE%AA%28+1965%E2%80%94%29%2C+%E9%80%9A+%E4%BF%A1+%E4%BD%9C+%E8%80%85%2C+%E5%8D%9A+%E5%A3%AB%2C+%E6%95%99+%E6%8E%88%2C+%E5%8D%9A+%E5%A3%AB+%E7%94%9F+%E5%AF%BC+%E5%B8%88%2C+%E4%B8%BB%E8%A6%81%E7%A0%94%E7%A9%B6%E9%A2%86%E5%9F%9F%E4%B8%BA%E8%87%AA%E7%84%B6%E8%AF%AD%E8%A8%80+%E5%A4%84+%E7%90%86%E3%80%81+%E5%95%86+%E5%8A%A1+%E6%99%BA+%E8%83%BD%E3%80%81+%E7%A4%BE)