next up previous
Next: Conclusions Up: Discussion Previous: Character Identification

Speaker identification

The identity of the speaker might be subtle, with embedded in self-reference:

"Tuk! little Tuk!"</QUOTE> said a voice.
It was a very little person who spoke. 
He was dressed as a sailor, and looked 
small enough to be a middy, but he was
 not one. <QUOTE TYPE="NEW"> "I bring
you many greetings from Corsor ... ,"
</QUOTE> said <CHARACTER ID="CORSOR">
CLASS="properName"> Corsor;</CHARACTER>
It is not obvious how to make the connection between a voice and Corsor. We believe that there will be limits on what can be done automatically, and that the more adult the stories the more complex this may be, but we still feel we can do better than our current version.

Nested spoken speech can also be hard (e.g., stories within stories, or quotes within quotes). In addition to the syntactic issues in identifying such quotations, there are the more complex issues of identifying the (sub)speaker. There is also the issue of how such speech should be rendered; should it be in the quoted speaker's voice, or in the main speaker's voice but rendered in such a way as to sound like the quoted speaker, or at least different from the main speaker?



Alan W Black 2003-10-20