Experiments
By taking part in these experiments, you'll be making a valuable contribution to the Convo project.
Our character Ditto the donkey will undergo training with examples taken from the conversation logs. Different people converse in different ways, so we want as many people as possible to interact with him.
We currently have five experiments running, plus a demonstration (links to items on this page):
- Experiment 1: Name Recognition
- Experiment 2: Simple Emotion Modelling
- Experiment 3: English or Gibberish
- Experiment 4: Question Recognition
- Experiment 5: Babbling - NEW!
- Demonstration: The Kuleshov Effect
More experiments coming soon (October 2006).
Experiment 1: Name Recognition
The aim of this experiment is to teach Ditto the donkey to recognise his own name. As pet owners know, many animals can recognise their name, so it seemed an appropriate place to start teaching Ditto. Try saying things to him and see if he can pick out his name among a jumble of other words. We'll use your messages to train him further, so he should improve over time.
Take part in Experiment 1: Name Recognition
Experiment 2: Simple Emotion Modelling
We're continuing Ditto's schooling by teaching him to react emotionally to English sentences, despite the fact that he doesn't understand them yet. In a sense he responds to your "tone of voice". We want you to insult Ditto and to compliment him, and hopefully he'll react appropriately. We'll use your messages to train him further, and he should gradually improve in his ability to understand the emotions behind English sentences, even if he doesn't understand the words yet.
Take part in Experiment 2: Simple Emotion Modelling
Experiment 3: English or Gibberish
In this experiment Ditto is being taught to distinguish English from other languages or garbage typing, which we've told him are called "gibberish". As far as Ditto's concerned, humans speak either English or Gibberish, and he has to decide whether your messages are in one language or the other.
Take part in Experiment 3: English or Gibberish
Experiment 4: Question Recognition
Questions are easy to recognise... as long as they end with a "?". But sometimes they don't, do they. In this experiment Ditto is being taught to identify questions.
Take part in Experiment 4: Question Recognition
Experiment 5: Babbling
In this experiment Ditto babbles like a baby in response to you. A donkey says his first words of English!
Take part in Experiment 5: Babbling
Demonstration: The Kuleshov Effect
It's part of human nature to try and make sense of randomness, and sometimes we see patterns where there aren't any. The early Russian filmmaker Kuleshov noticed that people interpreted an actor's facial expression depending on the context. But the actor wasn't acting at all: the same neutral shot was spliced into several different scenes, and people imagined they saw a range of emotions. A similar effect can happen with a conversational system programmed to give random responses. People try to make sense of the replies, even though they're completely arbitrary.
Experience The Kuleshov Effect for yourself