Out of curiosity, why do most chatbots not use a user feedback to improve their conversations? I just tried web-hal and it asked which one of it’s responses would be the most suitable one.
I understand most botmasters manually go through the scripts, but wouldn’t a more direct user feedback make the improvement quicker?

For me, it’s easier to go through the logs. That way, I can see and correct instantly any errors. I offer a contact address where people are welcome to send in suggestions (and often do). On the Loebner version of my bot, I encourage people to send in comments after speaking to it and post a selection on the bot’s page.

The bot allows the users to correct replies as they are speaking to it but I get these new responses emailed to me before I decide whether to include them permanently in it’s responses. I would rather manually correct any errors than let the users do so for obvious reasons.

Do you use any specific mechanism to allow users to feedback if a particular answer e.g. Didn’t make sense?

Yes, if the user says something like “that’s not right”, “you were wrong”, “your answer was incorrect” or something similar, it will prompt them for a new response which gets emailed to me.

Human: Do you like eggs?
Bot: I don’t know
Human: bad answer
Bot: OK, what should I have said?
Human: I liked fried eggs
Bot: OK, so I should have said, ” I like fried eggs ” in response to you saying, ” do you like eggs “. Is that correct?
Human: yes
Bot: I’ll try to remember that. If you say ” do you like eggs ” to me again, I will remember what you told me.
Human: Do you like eggs?
Bot: I like fried eggs.

They can also do something along these lines:

Human: When I say, “Jump”, you say, “How high”.
Bot: I’ll try to remember that. If you say ” Jump” to me again, I will remember what you told me.
Human: Jump
Bot: How high

If this was a genuine example, I would include this in its responses but most of the time people just teach it rubbish like “Tim is gay. What is Tim” and so I daren’t make this process automatic.