A coding scheme was developed to code a corpus of 500 human-human dialogues recorded at OVR's call centres [Vark et al., 1997a]. This scheme was later adapted for use in an automatic coder [Rats, 1997a]. We intend to represent the meaning of both customer utterances and system utterances in this scheme. The design of the coding scheme is inspired by Conversational Games Theory [Carletta et al., 1995] and the coding scheme of the VERBMOBIL project [Alexandersson and Reithinger, 1995]. However, the Alparon coding scheme has a more hierarchical structure.
The coding scheme also uses the notion of speech acts [Austin, 1962],
[Searle, 1969]. The title of the book where Austin introduced this theory says
quite a lot: "How to do things with language". Speech acts theory views
language use as performing actions, as opposed to earlier theories where
language was seen mainly as referring to objects and states of the world,
being "true" if this corresponded with the real world and "false" otherwise
[Vagle et al., 1993]. There have been many classifications of speech acts - in
the Alparon coding scheme we use "check", "clarification", question",
"reconfirmation" and "statement". As basic building blocks we take speech
acts with propositional contents of the utterance added (i.e. the literal
information contents).
Codes in the Alparon coding scheme have a hierarchical structure consisting of moves, codes, sub codes and arguments [Vark, 1997c]. Moves corresponds to speech acts. Codes, sub codes and arguments will here colloquially be named ``information codes''. An overview of the codes used is given in figure 3.2. As an example, the utterance "The trains leaves at 11:30" will hence be coded as "Statement(Time(Departure(Exact(absolute(11:30)))))"% 3.3. A set of abbreviations has been developed (see table 3.1). Using these, the example will be written as "Sta(Tm(DeT(Exact(absolute(11:30)))))". See [Rats, 1997a] for more details.
