What traits should the ideal chatbot authoring language have?
Gary, as you probably know, there’s been a lot of talk on Robitron and the Chatbots.org AIZone about Bruce Wilcox’s ChatScript (versus AIML). There is also RiveScript by Noah Petherbridge, CML (Conversation Markup Language) by Chongguan Yang, and AIScript (PersonalityForge) by Benji Adams. The list goes on.
Current Loebner champion Wilcox has waxed eloquent about this in March 2011, “Beyond Façade: Pattern Matching for Natural Language Applications”:
He revealed more details about his work in June 2011, “Suzette, the Most Human Computer”:
As I’ve mentioned previously on Chatbots.org AIZone, its not so much the language as the interpreter that makes the most difference to the learning curve. (Though I do agree theoretically with the logic for the improvements of ChatScript over AIML.) Its in fact the (cloud) infrastructure of new services like Chatbot4u that make the most difference. It just feels like to me that the old days of hosting an interpreter on your own hardware are quickly passing away.
That said, I’m a great believer in MODULARITY, open plug and play frameworks. I believe, both Apple Siri and IBM Watson were made in this way. Currently, the biggest bottleneck is the voice-in/voice-out lipsync avatar. I am a big promoter of XMPP as the lingua franca for open, modular conversational agents. But, there is no good IM-Voice bridge available at this time (with or without avatar).