Annotator differences

As discussed in chapter 4, annotators were intentionally given only the instruction to ``mark wherever the topic changes'' within documents. As a result, different annotators marked documents in slightly different ways: At one extreme, annotators marked only the start or end of large, clear, coherent topics. At the other extreme, some annotators made a mark every time a topic drifted in or out of immediate focus. For instance, in figure 5.13, all `disjoint' discussion is partitioned individually, such as the non-sequitur utterance ``I can go home now'', seen in figure 5.2.5. We can see here that the topic boundaries are used to constrain a topic amid a sea of noise--an assertion that the change from topic to no-topic is a topic change.

**Figure 5.13:** NIST dialogue,
$\begin{figure}\centering \epsfig{file=graphs/NIST_20020305-1007-catherine,width=1\textwidth}\end{figure}$

**Figure 5.14:** NIST dialogue 2,
$\begin{figure}\centering \epsfig{file=graphs/NIST_20020214-1148-catherine.ps,width=1\textwidth}\end{figure}$

**Figure 5.15:** A snippet from figure 5.13's source dialogue, showing unexpected topic-boundary marking style
$\begin{figure}{\tt 019: oh, that's a good idea \par 005: that's \par 003: you le... ...} \par 029: \{breath\} \par 006: \{vocalization\} \par <break /> }\end{figure}$

James Ballantine 2005-02-19