I was hoping to catch your interest. I am a fan of your blog and have read your bit on the glossaries for corrective feedback. It was Joseph Rezeau who first described using the Glossary for corrective feedback here on Moodle.org in 2005. I developed quite an extensive glossary module for that purpose but have since given up and have developed a site dedicated to corrective feedback on L2 writing errors here: http://virtualwritingtutor.com . It is a work in progress, but it was your summary of Rod Ellis's work that inspired me to go beyond the glossary module and try something more adventurous. Frankie Kam has done an interesting hack with it here: http://moodurian.blogspot.ca/2012/08/extend-functionality-of-ouwiki-by.html Ah, Frankie! What a wonderful person to know!
Okay, as for artificial intelligence chatbots, I have tried them and I find them annoying. Shulze is right. As for intelligent people, I have tried them and I find them fascinating. So for me, a worthy development goal is to capture the charm and beauty of people by inviting them to create a chatbot of themselves.
So to explain, the system I am presenting here has two components: 1) a speech recognition enabled video or audio chatbot to practice a conversation with, and 2) a chatbot maker component for creating a chatbot for others to have a conversation with. For language learners, both could be interesting forms of speaking practice. Interviewing a chatbot for the person's phone number and other details is an interactive listening activity. You have to ask for the number and address to hear them. There should also be an learning advantage to having a tireless conversation partner to interview online. Medical interviews are very formulaic. Many other work-related and counter-service work conversations are very formulaic also.
Creating a chatbot is a different language learning experience entirely. It is more like creating an eportfolio of answers to standard questions. The fact that students can then interact with their peers' chatbots after they have been created promises to be an interesting opportunity for peer-talk and peer-evaluation.
You raise the question of open dialogue systems where a user must formulate a question that the chatbot knows. To create such a system, you must anticipate all the questions that a user might ask and have videos at the ready for each one. A closed dialogue system simply feeds the user the lines to say. Other than having access to the questions beforehand, the two system types are identical. My design provides the questions using a slider with clickable sentences linked to audio models. It works well-enough for langauge learning. Combining an open dialogue system with a grammar checker to improve the system's "understanding" is a very interesting idea. Something to think about.
With Google's implementation of x-webkit-speech in HTML5, enabling speech recognition on a webpage is as easy as putting in a textbox. If you know how to play a different video or audio file depending on what is submitted to the texbox, then you have a chatbot of a real person with real intelligence and charm. Only the conversation part is simulated.
The problem with recognizing speech using Google's speech recognition service is that it is designed to recognize Standard L1 American speech. Longer sentences cause the system to hang and accents can be a problem. Adding an additional training phase with a speech recognizer that can score pronunciation before the chatbot conversation would be great. I may have to switch from Google to a different system altogether. I am looking into finding a speech recognition service I can afford. For now, it is Google's free service.
Thanks for the information about Flash and its WAV (SPEEX) format. Right now, I use a Java video recorder and Java MP3 recorder. When students use them from home, they sometimes dismiss the request to run the applet instead of accepting, and then of course the Java aplets don't work. Students send me mystified emails. Students seem to be more familiar with Flash, so they are more likely to accept. Is there no third option?
The other reason why chatbotmaker is Chrome only is that my Java video recorder records MP4s, a problem for some browsers. So for now, its Chrome for the free version, and I'll have to come up with a pro version with cross-browser compatibilty.