【演講公告】2026.11.03「From Cultural Archives to Judicial Reasoning: Structuring Text and Norms across Digital Humanities and Legal Informatics」- 國立政治大學資 訊科學系劉昭麟特聘教授
1. 講者: 國立政治大學資訊科學系劉昭麟特聘教授
2. 題目: From Cultural Archives to Judicial Reasoning: Structuring Text and
Norms across Digital Humanities and Legal Informatics
3. 參加對象: 全校大三以上學生、教職員與長庚醫師
4. 實體地點: 管理大樓 11 樓 AI 講堂
5. 演講時間: 2026 年 11 月 3 日(二)13:00-15:00
6. 線上連結: https://reurl.cc/QNl3rM
演講無需報名, 請於演講時間進入 Teams 線上會議室
講者簡介:
劉昭麟特聘教授現任教於國立政治大學資訊學院資訊科學系。其研究涵蓋自然語言處理、認知歷程、自動推理與機器學習理論、資訊檢索與知識管理系統、數位人文、法學資訊、智慧
型教學系統、智慧型運輸系統、決策支援、規劃及診斷系統等。曾任政大理學院副院長、資
訊學院副院長及資訊科學系主任,並擔任臺灣數位人文學會理事長等學術服務職務。劉教授
亦曾赴密西根大學、哈佛大學及加州大學聖塔芭芭拉校區訪學,致力於以計算技術與人工智
慧推動學術研究及社會發展。
演講摘要:
Generic statistical language models and unstructured text embeddings often struggle when
applied to specialized Chinese corpora, where meaning is strictly governed by historical
conventions or institutional norms. This talk presents an interdisciplinary journey across two
distinct applications of computational semantics: discovering latent patterns in historical
culture and enabling explainable decision-making in modern jurisprudence.
The first half of the presentation focuses on digital humanities, demonstrating how
computational methods overcome unpunctuated and irregular classical formats. We examine
how linguistic heuristics and structural parallelism resolve classical poetry word
segmentation, how large-scale alignment reveals diachronic stylometric patterns and text
reuse across 2,000 years of poetic collections, and how language models and Conditional
Random Fields extract prosopographical knowledge and demarcate biographical entries from
historical local gazetteers (Difangzhi).
The second half turns to legal informatics, exploring how domain priors transform unstructured
prosecutorial and judicial documents into transparent, interpretable outputs. We demonstrate
how statutory article co-citation networks establish objective case similarity in civil disputes;
how deconstructing criminal fraud indictments into statutory constitutive elements yields
explainable case analogies aligned with professional legal judgment; how structured prompt
frameworks guide large language models through multi-cause concurrence and penalty
comparisons; and how procedural trajectories can quantify case disputability and litigation
risk. Together, these case studies demonstrate that computational semantics achieves its
highest reliability not by treating language as generic token distributions, but by formalizing the
structural and normative logic of the target domain.