"... In PAGE 2: ... Due to the labor intensive nature of the experiment, ten rather short texts were used, three of which were considered simplistic, ranging from paragraph to page in length. In Table1 , the combination of information types and representation forms outlined and used by Sparck Jones is illustrated. The conclusion of the paper by Sparck Jones (1993) is that there is no single representation of discourse structure that capture the information in a text needed to produce a good summary, and that a more comprehensive discourse model is required to do so.... ..."

"... In PAGE 9: ...nseen documents i.e. the test set. In this study we split the training and test sets into different sizes to evaluate the impact of larger training sets on the SVM. Table1 outlines the different size training and test sets used. Each of the experiments conducted on the different repre- sentations is evaluated in terms of precision and recall.... ..."

"... In PAGE 7: ... Optionally, Collagen can also use speech recognition software to allow the user to speak these utterances rather than creating them through the GUI, and it can use speech synthesis software to allow the agent to speak its utterances. 4 Tutorial Behaviors as Collaborative Discourse Acts Table1 is a summary of our progress in integrating ITS and CDS: it lays out in detail how eachofPaco apos;s tutorial behaviors is generated from Collagen apos;s discourse state representation and Paco apos;s student model. The #0Crst column of the table is a ranked list of the tutorial act types.... In PAGE 9: ...hich remains there until the agent provides the help #28e.g., line 41#29. Using the generic capabilities of Collagen to record information about a user, Paco maintains a simple overlay model #5B4#5D that records, for each step in a recipe, whether the student has been exposed to it. In Table1 , the condition #5Cthe student knows step ! quot; means that the student has been taught this step before. The condition #5Cstudent knows step ! needs to be done quot; means the student has been taught all the steps that connect ! to the root of the current plan.... In PAGE 9: ... However, this approach leaves open the question of howtochoose which act to perform. Paco chooses which act to perform based on the rankings of the discourse acts, given in the #0Crst column of Table1 . For example, Paco prefers to give initiative when the student knows what to do next rather than teach or remind her what to do next.... ..."

"... In PAGE 8: ... Optionally, Collagen can also use speech recognition software to allow the user to speak these utterances rather than creating them through the GUI, and it can use speech synthesis software to allow the agent to speak its utterances. 4 Tutorial Behaviors as Collaborative Discourse Acts Table1 is a summary of our progress in integrating ITS and CDS: it lays out in detail how eachofPaco apos;s tutorial behaviors is generated from Collagen apos;s discourse state representation and Paco apos;s student model. The #0Crst column of the table is a ranked list of the tutorial act types.... In PAGE 10: ...hich remains there until the agent provides the help #28e.g., line 41#29. Using the generic capabilities of Collagen to record information about a user, Paco maintains a simple overlay model #5B4#5D that records, for each step in a recipe, whether the student has been exposed to it. In Table1 , the condition #5Cthe student knows step ! quot; means that the student has been taught this step before. The condition #5Cstudent knows step ! needs to be done quot; means the student has been taught all the steps that connect ! to the root of the current plan.... In PAGE 10: ... However, this approach leaves open the question of howtochoose which act to perform. Paco chooses which act to perform based on the rankings of the discourse acts, given in the #0Crst column of Table1 . For example, Paco prefers to give initiative when the student knows what to do next rather than teach or remind her what to do next.... ..."

"... In PAGE 6: ... Optionally, Collagen can also use speech recognition software to allow the user to speak these utterances rather than creating them through the GUI, and it can use speech synthesis software to allow the agent to speak its utterances. 4 Tutorial Behaviors as Collaborative Discourse Acts Table1 is a summary of our progress in integrating ITS and CDS: it lays out in detail how eachofPaco apos;s tutorial behaviors is generated from Collagen apos;s discourse state representation and Paco apos;s student model. The #0Crst column of the table is a ranked list of the tutorial act types.... In PAGE 8: ...hich remains there until the agent provides the help #28e.g., line 41#29. Using the generic capabilities of Collagen to record information about a user, Paco maintains a simple overlay model #5B4#5D that records, for each step in a recipe, whether the student has been exposed to it. In Table1 , the condition #5Cthe student knows step ! quot; means that the student has been taught this step before. The condition #5Cstudent knows step ! needs to be done quot; means the student has been taught all the steps that connect ! to the root of the current plan.... In PAGE 8: ... However, this approach leaves open the question of howtochoose which act to perform. Paco chooses which act to perform based on the rankings of the discourse acts, given in the #0Crst column of Table1 . For example, Paco prefers to give initiative when the student knows what to do next rather than teach or remind her what to do next.... ..."

"... In PAGE 6: ...2 Performance Evaluation First, we tested the speech act analysis model and the dis- course analysis model by using the same training and test sets. Table6 shows the results for each analysis model. In Table 6, the results of speech act analysis are obtained by using the correct structural information of discourse, i.... In PAGE 6: ... Table 6 shows the results for each analysis model. In Table6 , the results of speech act analysis are obtained by using the correct structural information of discourse, i.e.... In PAGE 6: ... Simi- larly, the results of discourse structure analysis are obtained by using the correct speech act information from the anno- tated dialogue corpus. As shown in Table6 , the proposed models show better results than previous works such as Lee (1997) [13], Kim (2003) [26], and Kim (1998) [27]. Model- II shows better results than Model-I in all cases.... ..."

"... In PAGE 6: ... Note that we make the assumption that a new sentence can be connected to the sentences on the right most edge in the discourse tree (in Figure 5, S5 is not allowed to connect to S1 and S2). (1) Cue phrases Examples of rules for matching cue phrases are shown in Table1 . Each rule speci es a condition for a pair of a new sentence and a possible connected sentence: the range of possible connected sentences (how far from the new sentence) and patterns for the two sentences.... ..."