--- title: "基于短语及依存的标注规则和短语识别算法研究 — 刘广 2024" description: "Most Chinese dependency parsing runs on word-segmented text and relies on end-to-end supervised models." authors: ["刘广", "涂刚", "李政", "刘译键"] year: 2024 venue: "中文信息学报 (Journal of Chinese Information Processing)" source: "http://jcip.cipsc.org.cn/article/id/zwxxxb_3676" 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/liu-2024-annotation-rules-phrase/" --- [Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/liu-2024-annotation-rules-phrase/) [home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [2024](https://md.research.elamur.ai/free/papers/2024/page/4/) / 基于短语及依存的标注规则和短语识别算法研究 # 基于短语及依存的标注规则和短语识别算法研究 **Authors:** 刘广, 涂刚, 李政, 刘译键 · **Year:** 2024 · **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_3676) · [PDF](http://jcip.cipsc.org.cn/cn/article/pdf/preview/zwxxxb_3676.pdf) ## Summary Most Chinese dependency parsing runs on word-segmented text and relies on end-to-end supervised models. The authors identify two weaknesses in this approach. First, annotation schemes such as LTP and Baidu's use dozens of POS and relation labels, which makes annotation slow and error-prone for non-linguists. Second, sequence-labelling models assign one tag per span and so cannot represent nested constituents. To address this, the authors take the phrase rather than the word as the basic unit and define a simplified scheme with seven nestable phrase types, plus dependency links between phrases. Using this scheme they built the Chinese Phrase Window Dataset (CPWD). They also propose the Syntax Window Model (SWM), which borrows its design from Faster R-CNN object detection. SWM enumerates every candidate span, predicts whether each span is a phrase, adjusts its start and end boundaries with small offsets, and then classifies the phrase type. This lets it recover overlapping phrases and their dependencies jointly. On CPWD, SWM beats a conventional end-to-end baseline by over 1 F1 point. The same method also won first place in the CCL2018 Chinese metaphor sentiment analysis task. The work offers an easier-to-annotate, nesting-aware alternative to fine-grained Chinese parsing schemes. ## Key points - **Seven phrase categories**, each with its own bracket notation: noun, verb, numeral-classifier, prepositional, conjunction, modal-particle, and clause. Pivotal, serial-verb, and subject/object clauses are all marked as clauses. - **CPWD corpus:** - 45,000 mostly modern Chinese sentences drawn from dialogue, news, law, policy texts, and fiction. - Sentences are capped at 50 characters. - The data is split into 35,000 training, 5,000 validation, and 5,000 test sentences. - **SWM architecture:** - A feature encoder (LSTM or BERT; experiments mention a 2-layer BiLSTM) feeds a window-localisation network. - This network scores all N(N+1)/2 spans and regresses start/end offsets of at most one character. - Training combines a cross-entropy loss for phrase detection with an RMSE loss for the offsets. - **Results:** SWM improves F1 by more than 1 point over a traditional end-to-end model on CPWD. In CCL2018 metaphor sentiment analysis, the method added more than 1 F1 point and ranked first. *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{2024, title = {{基于短语及依存的标注规则和短语识别算法研究}}, author = {刘广 and 涂刚 and 李政 and 刘译键}, journal = {中文信息学报 (Journal of Chinese Information Processing)}, year = {2024}, url = {http://jcip.cipsc.org.cn/article/id/zwxxxb_3676} } ``` - [← previous 基于大语言模型的事件常识知识图谱扩展方法](https://md.research.elamur.ai/free/papers/huang-2024-method-expanding-event/) - [next → 基于规则提示的知识图谱通用推理预训练模型](https://md.research.elamur.ai/free/papers/cui-2024-pre-trained-universal/) - [↑ back to papers of 2024](https://md.research.elamur.ai/free/papers/2024/page/4/) ## References 7 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. [ 基于字的分布表征的汉 11] 李国 臣, 党 帅 兵, 王 瑞 波, 等 . 语基本块识别 [ 中 文 信 息 学 报, J]. 2014, 28( 06):18 - 25. [find](https://scholar.google.com/scholar?q=%5B+%E5%9F%BA%E4%BA%8E%E5%AD%97%E7%9A%84%E5%88%86%E5%B8%83%E8%A1%A8%E5%BE%81%E7%9A%84%E6%B1%89+11%5D+%E6%9D%8E%E5%9B%BD+%E8%87%A3%2C+%E5%85%9A+%E5%B8%85+%E5%85%B5%2C+%E7%8E%8B+%E7%91%9E+%E6%B3%A2%2C+%E7%AD%89+%2E+%E8%AF%AD%E5%9F%BA%E6%9C%AC%E5%9D%97%E8%AF%86%E5%88%AB+%5B+%E4%B8%AD+%E6%96%87+%E4%BF%A1+%E6%81%AF+%E5%AD%A6+%E6%8A%A5%2C+J%5D%2E+2014%2C+28%28+06%29%3A18+%2D+25%2E) 2. [ 12] 徐菁 . 面 向 中 文 知 识 图 谱 的 开 放 式 文 本 信 息 抽 取 关 键技术研究[ 国防科技大学硕士学位论文, D]. 2018. [find](https://scholar.google.com/scholar?q=%5B+12%5D+%E5%BE%90%E8%8F%81+%2E+%E9%9D%A2+%E5%90%91+%E4%B8%AD+%E6%96%87+%E7%9F%A5+%E8%AF%86+%E5%9B%BE+%E8%B0%B1+%E7%9A%84+%E5%BC%80+%E6%94%BE+%E5%BC%8F+%E6%96%87+%E6%9C%AC+%E4%BF%A1+%E6%81%AF+%E6%8A%BD+%E5%8F%96+%E5%85%B3+%E9%94%AE%E6%8A%80%E6%9C%AF%E7%A0%94%E7%A9%B6%5B+%E5%9B%BD%E9%98%B2%E7%A7%91%E6%8A%80%E5%A4%A7%E5%AD%A6%E7%A1%95%E5%A3%AB%E5%AD%A6%E4%BD%8D%E8%AE%BA%E6%96%87%2C+D%5D%2E+2018%2E) 3. [ 基于强化学习协同训练的 13] 程钟 慧,陈 珂,陈 刚,等 . 命名实体识别方 法 [ 软 件 工 程,2020,23( J]. 01):7 - 11. [find](https://scholar.google.com/scholar?q=%5B+%E5%9F%BA%E4%BA%8E%E5%BC%BA%E5%8C%96%E5%AD%A6%E4%B9%A0%E5%8D%8F%E5%90%8C%E8%AE%AD%E7%BB%83%E7%9A%84+13%5D+%E7%A8%8B%E9%92%9F+%E6%85%A7%2C%E9%99%88+%E7%8F%82%2C%E9%99%88+%E5%88%9A%2C%E7%AD%89+%2E+%E5%91%BD%E5%90%8D%E5%AE%9E%E4%BD%93%E8%AF%86%E5%88%AB%E6%96%B9+%E6%B3%95+%5B+%E8%BD%AF+%E4%BB%B6+%E5%B7%A5+%E7%A8%8B%2C2020%2C23%28+J%5D%2E+01%29%3A7+%2D+11%2E) 4. [ 题元 理 论 与 汉 语 配 价 问 题 [ 14] 徐烈炯 . J]. 当 代 语 言 学, ( ) : 1998 3 1 - 21. [find](https://scholar.google.com/scholar?q=%5B+%E9%A2%98%E5%85%83+%E7%90%86+%E8%AE%BA+%E4%B8%8E+%E6%B1%89+%E8%AF%AD+%E9%85%8D+%E4%BB%B7+%E9%97%AE+%E9%A2%98+%5B+14%5D+%E5%BE%90%E7%83%88%E7%82%AF+%2E+J%5D%2E+%E5%BD%93+%E4%BB%A3+%E8%AF%AD+%E8%A8%80+%E5%AD%A6%2C+%28+%29+%3A+1998+3+1+%2D+21%2E) 5. [ 生成 语 法 中 词 汇 语 义 与 句 法 的 界 面 研 究 [ 15] 刘宇红 . J]. 外语学刊, 2011( 05):56 - 60. [find](https://scholar.google.com/scholar?q=%5B+%E7%94%9F%E6%88%90+%E8%AF%AD+%E6%B3%95+%E4%B8%AD+%E8%AF%8D+%E6%B1%87+%E8%AF%AD+%E4%B9%89+%E4%B8%8E+%E5%8F%A5+%E6%B3%95+%E7%9A%84+%E7%95%8C+%E9%9D%A2+%E7%A0%94+%E7%A9%B6+%5B+15%5D+%E5%88%98%E5%AE%87%E7%BA%A2+%2E+J%5D%2E+%E5%A4%96%E8%AF%AD%E5%AD%A6%E5%88%8A%2C+2011%28+05%29%3A56+%2D+60%2E) 6. [ 基于大规模 语 义 知 识 库 的“ 词 汇—句 法 语 义” 16] 孙道功 . 接口研究[ 语言文字应用, J]. 2016( 02):125 - 134. [find](https://scholar.google.com/scholar?q=%5B+%E5%9F%BA%E4%BA%8E%E5%A4%A7%E8%A7%84%E6%A8%A1+%E8%AF%AD+%E4%B9%89+%E7%9F%A5+%E8%AF%86+%E5%BA%93+%E7%9A%84%E2%80%9C+%E8%AF%8D+%E6%B1%87%E2%80%94%E5%8F%A5+%E6%B3%95+%E8%AF%AD+%E4%B9%89%E2%80%9D+16%5D+%E5%AD%99%E9%81%93%E5%8A%9F+%2E+%E6%8E%A5%E5%8F%A3%E7%A0%94%E7%A9%B6%5B+%E8%AF%AD%E8%A8%80%E6%96%87%E5%AD%97%E5%BA%94%E7%94%A8%2C+J%5D%2E+2016%28+02%29%3A125+%2D+134%2E) 7. [ 施 事、 受 事 句 法 实 现 的 义 类 17] 亢世 勇, 许 小 星, 马 永 腾 . 制约[ 语文研究, J]. 2011( 04):36 - 40. [find](https://scholar.google.com/scholar?q=%5B+%E6%96%BD+%E4%BA%8B%E3%80%81+%E5%8F%97+%E4%BA%8B+%E5%8F%A5+%E6%B3%95+%E5%AE%9E+%E7%8E%B0+%E7%9A%84+%E4%B9%89+%E7%B1%BB+17%5D+%E4%BA%A2%E4%B8%96+%E5%8B%87%2C+%E8%AE%B8+%E5%B0%8F+%E6%98%9F%2C+%E9%A9%AC+%E6%B0%B8+%E8%85%BE+%2E+%E5%88%B6%E7%BA%A6%5B+%E8%AF%AD%E6%96%87%E7%A0%94%E7%A9%B6%2C+J%5D%2E+2011%28+04%29%3A36+%2D+40%2E)