Conclusion

Abstract

In Chap. 3, we proposed a method to automatically evaluate argumentation skills from the records of argumentation exercises. Using two scoring sheets Round A (competitive argumentations) and Round B (cooperative argumentations) of an intercollegiate negotiation competition as a reference, several evaluation items were selected for each type of argumentation (cooperative argumentation, competitive argumentation). We then asked the experts (A, B, and C) to score each of the sample argumentation documents subjectively, represented the individual argumentation records as a series of “speech acts + factors” (training data), and performed multiple regression analysis between the pattern of appearance of “speech acts + factors” in the training data and the scoring results. The scoring prediction model was constructed by performing multiple regression analysis between the pattern of occurrence of “speech act + factor” and the scoring results in the training data. This predictive model reveals what each scorer pays attention to when making evaluations. Using this model, we scored the argumentation records of the garbage house problem and found that the accuracy of judgments of high-scoring argumentations with a negotiation strategy score of 3.5 or higher was 80%, 70%, and 90% for scorers A, B, and C. The estimation accuracy of the record of high-scoring utterances with a good working relationship score of 3.5 or higher was as high as 90% for scorer A, 80% for scorer B, and 100% for scorer C. This indicates that it is possible to identify excellent answers with high accuracy even for evaluation items such as negotiation strategies and good working relationships, which are conventionally difficult to evaluate objectively.