WRT: Writer Response Theory

Jabberwacky, a chatbot system which learns continuously from its conversations with web users, has won a Loebner prize for the second year in a row. A competitor in the Loebner contest for simulated human conversation since 2002, the chatbot system has sported both incremental improvements and some recent dramatic facelifts.

Last year, the Jabberwacky chatbot winner was George, a chatbot speech and voice-recognition enabled chatbot whose avatar goes beyond the Oddcast SitePal technology now common in online Pandorabots by displaying emotional body language to suit the topic and general mood. George has been getting a lots of press coverage of late, although this year’s winning Jabberwacky chatbot is actually Joan.

Particularly interesting in the Joan interface is the option of providing emotional feedback and coaching via drop-down boxes - indicating the mood or tenor of statements to help Joan learn about facial expression and inflection. As with the basic Jabberwacky approach to conversation, this appears to be an attempt to collect a kind of collective wisdom about the tenor of conversational memes - but mood is so stateful that I wonder if the data will provide for a mood significantly more convincing than just cycling mood states at random intervals. Perhaps with the Jabberwacky entry in 2007, we’ll see even more.

More resources:

• Jabberwacky press releases
  Wikipedia: Jabberwacky
  ‘Interview’ with George