如何分析課堂發言：人工智能與統計方法的結合

統計人工智能專家系統（SAIES）可以自動分析課堂發言並幫助學生學習。自動轉錄的課堂發言文稿可以形成一個數據庫，計算語言學使得統計語篇分析（SDA）進行自動變量分類成為可能。這一方面從根本上改變了SDA模型的後續分析流程，另一方面可以解釋多層次變量對目標動作產生了怎樣的影響。SAIES將理論模型轉換為統計模型，對數據進行測試，並對結果進行解釋。筆者用SAIES分析了17組學生與教師組合在13周內進行的課程設計討論。分析結果表明在以下四種情況下，更容易出現微小創新︰（1）學生們過去的學習成績更好；（2）在較近的發言次序中出現了微小創新；（3）在錯誤答案出現之後有學生提出了不同意見；（4）小組的問題解決方案得分越高。
Statistics Artificial Intelligence Expert System (SAIES) automatically analyzes classroom participation and aid students’ learning. After its commercialization, automated transcription can create a database and computational linguistics can automatic categorization of variables for Statistical Discourse Analysis (SDA). SDA models pivotal actions that radically change subsequent processes as well as effects of explanatory variables at multiple levels (sequences of turns/messages, time period, person, group, organization, and so on) on target actions. SAIES translates a theoretical model into a statistical model, tests the model on the data, and eventually interprets the results (SAIES rewrites itself to run revised analyses if needed). As this paper discusses, applying SAIES to 30,569 words in 3,296 turns of talk by 80 students in 20 groups shows more correct, new ideas (micro-creativity) when: 1) students have higher past achievement, 2) recent turns of talk show micro-creativity, 3) a wrong answer is followed by a disagreement, or 4) groups have higher solution scores. Copyright © 2019 北京大學.