Dialogue Act Classification In Human-to-Human Tutorial Dialogues
We present in this paper preliminary results with dialogue act classification in human-to-human tutorial dialogues. Dialogue acts are ways to characterize the actions of tutors and students based on the language-as-action theory. This work serves our larg
- PDF / 912,214 Bytes
- 4 Pages / 439.37 x 666.142 pts Page_size
- 63 Downloads / 248 Views
1
Introduction
A key research question in intelligent tutoring systems (Rus, D’Mello, Hu, & Graesser, 2013) and in the broader instructional research community is understanding what expert tutors do. This goal is motivated by research showing that expert tutors are more effective when it comes to student learning gains (2-sigma effect size which is equivalent to 2 letter grades improvement, e.g. from C to A) than unaccomplished tutors (effect size=0.4; Bloom, 1984). A typical operationalization of this goal of understanding of what good tutors do was to define the behavior of tutors based on their actions. In our case, we model the dialogues as dialogue-act sequences because there are no other types of actions, e.g. task actions as in Boyer and colleagues (2011), that are available in the human-tohuman tutorial dialogues we obtained. Our view of a tutorial dialogue as a sequence of actions is based on the languageas-action theory (Austin, 1962; Searle, 1969) according to which when “we say something, we do something.” Therefore, all utterances in a tutorial dialogue are mapped into corresponding dialogue acts using, in our case, a predefined dialogue or speech act taxonomy. The taxonomy was defined by educational experts and resulted in a two-level hierarchy of 16 top-level dialogue acts and a number of dialogue subacts. The exact number of subacts differs from dialogue act to dialogue act. The overall, two-level taxonomy consists of 140 unique dialogue-act-subact combinations. It should be noted that automatically discovered dialogue act taxonomies are currently being built (Rus, Graesser, Moldovan, & Niraula, 2012) but it is beyond the scope of this paper to automatically discover the dialogue acts in our tutoring sessions.
© Springer Science+Business Media Singapore 2017 E. Popescu et al. (eds.), Innovations in Smart Learning, Lecture Notes in Educational Technology, DOI 10.1007/978-981-10-2419-1_25
185
186
2
V. Rus et al.
The Approach
We adopted a supervised machine learning method to automate the process of dialogue act classification. Machine learning methods imply the design of a feature set which can then be used together with various machine learning algorithms such as Naive Bayes, Decision Trees, and Bayes Nets. In the automated dialogue act classification literature, researchers have considered rich feature sets that include the actual words (possibly lemmatized or stemmed) and n-grams (sequences of consecutive words). In almost every such case, researchers apply feature selection methods because considering all the words might lead to overfitting and, in the case of n-grams, to data sparseness problems because of the exponential increase in the number of feature values. Besides the computational challenges posed by such feature-rich methods, it is not clear whether there is need for so many features to solve the problem of dialogue act classification. We believe that humans infer speakers’ intentions after hearing only few of the leading words of an utterance (Moldovan, Rus, & Graesser, 2011). O
Data Loading...
