WRT: Writer Response Theory

(See video of chatbot Ben Franklin’s Ghost in action)

You walk up to the oversized book at the front of a curious glassed in stage, something of a cross between a puppet theater and a mausoleum. It’s the tricentennial of Ben Franklin’s birth and you haven’t seen any sign of him yet. On closer inspection, the book turns out to be the frame for two touch screens. On each, you see that familiar Revolutionary scrawl you’ve seen all over Philadelphia, remediated, of course. Choosing from Ye Old Verbs, Nouns, and Adjectives, you build a question. But who will answer it? Pressing the red wax seal marked “Ask Ben,” you summon the transparent video image of Ben Franklin himself, at least in that semi-cartoonish, kid-friendly version of a historical figure that we reserve for our coins, our foundation myths, and our advertising campaigns. Ben is reborn as a bot for all seasons.
Ben Franklin’s Ghost is a Pennsylvania original:

Ben Franklin’s Ghost was funded by the Pennsylvania Department of Community and Economic Development managed through the Pittsburgh Film Office, Carnegie Mellon University’s Entertainment Technology Center and Lights of Liberty, Inc.

Of course, They’ve chosen their subject well. Franklin no doubt invented his own chatbot, using water-power and old book bindings. (In this fascinating post, and excellent note on the proto-chatbot Turk, Rob McDougal notes: Franklin was, it turns out, a friend of the Turk’s “inventor” Wolfgang von Kempelen, and a fan of Kempelen’s chess-playing Turk.)

According to the Pittsburgh Post Gazette, Franklin’s Ghost is built on The Synthetic Interview, a chatbot system designed by Scott Stevens (writing below with Donald Marinelli) at CMU’s Human-Computer Interaction Institute and Michael Christel of the School of Computer Science. Franklin is the “first public exhibit to use the technology.” The Synthetic Interview device was previously used to generate Albert Einstein.
The PG article aslo quotes Ann Meredith, who says, “It’s the closest people will come to being able to speak to Benjamin Franklin.” Meredith certainly shows how far we’ve come from concerns about the Turing test, completely skipping over the question of whether or not machines can enter the astral plane in the firsplace. The article also sites, Jessica Trybus, director of edutainment at ETC, a position that is itself another sign of the times.
Parts of Speech & Parsing Speech

I watched as children asked a variety of questions. Tell me about electricity? Did you invent electricity, that sort of thing. They stayed and stayed and stayed. In fact, the children I graciously let go before me turned out to have been back 3 times to talk to Ben. Meanwhile, I prepared my bad behavior question (what these botmasters call “pool” questions). Did you marry John Adams to invent the President? To my chagrin, pithy Ben replied with a rather riddling axiom. Apparently he saw me coming.

The interface is rather remarkable for the way in which the user constructs sentences. Rather than allowing free input, the system takes the unusual constraint of requiring the user to select from a menu of choices, reducing the number of possible statements infinitely, while still offering a practically infinite set of combinatoric possibilities. [Admittedly, choosing from “Common Words” in this milieu of nostalgic American foundation-myths has a particularly democratic feel–though the “Proper Nouns” get the lion’s share of the matching attention.] One might expect that the parsing would take advantage of this system. However after much testing and bad behavior from two WRT members, the system proved itself to only due a bit of keyword matching. Entering these parts of speech in any order seemed to produce the same response and verbs did not seem to act differently than, say, nouns, as keywords.

Stevens and Marinelli describe their system:

American readers will more easily grasp what we are doing if we describe the system as a really big “Jeopardy” game, i.e., a database of answers in search of the proper or appropriate questions. (Marinelli and Stevens)

The system takes advantages of 5,000 most common English words and at least 1,000 domain specific words.

The domain specific language model includes idiosyncratic terms, proper names, foreign words and phrases, and geographic references that are part of the world of the persona being interviewed. (Marinelli and Stevens)

Curiously, they use the word “persona” rather than “person” to describe the human that has been re-presented. Perhaps the fact that Synthetic Interview is geared towards reproducing popular images of historical and legendary figures also helps sidestep the Turing dilemma.

All in all the combination of the physical exhibit, the performance, and the interface seemed to “deliver” in ways that the Liberty Bell couldn’t hope to. This was certainly a step beyond the Hall of Presidents, at least as far as information content was concerned. Another victory for hyperreal history? Let chatbots ring!

Marinelli, Donald and Scott Stevens. “Synthetic interviews: the art of creating a “dyad” between humans and machine-based characters.” Proceedings of the sixth ACM international conference on Multimedia: Technologies for interactive movies. 1998: 11-16.

Spice, Byron. “CMU puts words in Ben Franklin’s mouth.” Post Gazette.com. Local news. Education. 30 June, 2005.