"... In PAGE 5: ... We group conversational actions into one operator if they achieve the same effect, so that two superficially different utterances constitute examples of the same operator if they take the dialogue in the same direction. Table1 summarizes the operators supported by the Adaptive Place Advisor. Let us first consider the operators available to the dialogue manager for advancing the conversation.... ..."

"... In PAGE 3: ...3 The next step was to analyze the conversations linguistically and in terms of library related features. We rst categorized dialogues into several groups, listed in Table1 , according to the type of task of the patron, assuming that di erent tasks would exploit di erent strategies not only in solving the task but also in conversational structure. This categorization would con rm or nullify this assumption, but also simplify further analysis.... In PAGE 3: ... These included length of conversations: number of words, utterances, and turns; control of conversations; the amount of inherent structure in the conversations; and task-solving strategy. Table1 gives some counts for typical conversations in each category. Categories with lots of structure allowed pre-stored responses, so were good candidates for early development, because we only had to consider conversational features without worrying about how to access databases to provide correct answers for queries.... ..."

"... In PAGE 3: ... 3 The next step was to analyze the conversations linguistically and in terms of library related features. We #0Crst categorized dialogues into several groups, listed in Table1 , according to the type of task of the patron, assuming that di#0Berent tasks would exploit di#0Berent strategies not only in solving the task but also in conversational structure. This categorization would con#0Crm or nullify this assumption, but also simplify further analysis.... In PAGE 3: ... These included length of conversations: number of words, utterances, and turns; control of conversations; the amount of inherent structure in the conversations; and task-solving strategy. Table1 gives some counts for typical conversations in each category. Categories with lots of structure allowed pre-stored responses, so were good candidates for early development, because we only had to consider conversational features without worrying about how to access databases to provide correct answers for queries.... ..."

"... In PAGE 5: ... We group conversational actions into one operator if they achieve the same effect, so that two superficially different utterances constitute examples of the same operator if they take the dialogue in the same direction. Table1 summarizes the operators supported by the Adaptive Place Advisor. Let us first consider the operators available to the dialogue manager for advancing the conversation.... ..."

"... In PAGE 5: ... We group conversational actions into one operator if they achieve the same effect, so that two superficially different utterances constitute examples of the same operator if they take the dialogue in the same direction. Table1 summarizes the operators supported by the Adaptive Place Advisor. Let us first consider the operators available to the dialogue manager for advancing the conversation.... ..."

"... In PAGE 5: ... We group conversational actions into one operator if they achieve the same effect, so that two superficially different utterances constitute examples of the same operator if they take the dialogue in the same direction. Table1 summarizes the operators supported by the Adaptive Place Advisor. Let us first consider the operators available to the dialogue manager for advancing the conversation.... ..."

"... In PAGE 5: ... We group conversational actions into one operator if they achieve the same effect, so that two superficially different utterances constitute examples of the same operator if they take the dialogue in the same direction. Table1 summarizes the operators supported by the Adaptive Place Advisor. Let us first consider the operators available to the dialogue manager for advancing the conversation.... ..."

"... In PAGE 5: ... Our current system design contains approximately thirty dialogue messages, which are intended to promote a lim- ited, but interesting human-machine conversation. A selec- tion of these messages is given in Table2 . Note that Table 2 only describes the content of messages (what is said), and not the expression (how it is conveyed).... ..."

"... In PAGE 6: ... Based on the above dialogue structure, we propose a linear system for annotating dialogue acts for the MTCC. As shown in Table2 , we use thirty-seven annotation tags to mark up the discourse functions of the utterances. Unlike the sequential dialogue structure, the main discussion of a topic here is rather dynamic.... In PAGE 6: ... 3.2 Dialogue Acts Based on Table2 , this section presents a brief introduction to the annotation tags without giving explicit examples due to the lack of space4. To start a conversation contains only one annotation tag.... ..."