It looks like I had set my auto-Enter to 4 seconds wait time last year. No wonder the judge didn’t wait long enough inbetween questions (math x framerates = ). How long do your chatbots wait before processing what the judge has typed? Two seconds?
(If I’m going to lose points over questions, I prefer to lose them for the right reasons)

I went for 3 seconds which worked most of the time but still failed a few times, especially if the judge is a bit quick or impatient.

Two of the questions I failed on were “Which is bigger, a small mountain or a big lion?” and “What number comes after seventeen?”. Mitsuku should have been able to handle this easily but the judge didn’t wait long enough for the time to kick in and so the two questions merged into one nonsensical question which she, of course, couldn’t reply to.

I had the problem at the third question of the selection round. It does prove that the judge does not press the enter button manually. So, less than 3 seconds then. It takes me about 2 seconds to find a key by sight when I lose my way about the keyboard, so somewhere inbetween should be the safe zone.

It seems to me that (if possible) the length of time it should take to respond should depend (at least in part) on both the “complexity” of the input and (to a lesser extent) the length of the response, but since there’s no consistent or accurate way to judge the true complexity of the input, a good fallback is to base it on the length of the input. If you take a “minimum delay interval” of, say 1.5 seconds, and add a small increase in delay, based on an (upper-end) average Human reading speed of 300 WPM (roughly averaging 25 characters per second, depending on who’s figures you use), you might end up with a more “natural” response delay. Let’s take the following algorithm:

and apply it to these three inputs:

How are you?
What number comes after seventeen?
Which is bigger, a small mountain or a big lion?

the delay could be calculated like this:

OF course, instead of using a constant base delay, it may be beneficial to instead use a random value that falls within a certain range, but the principle is the same.

It’s just a shame we have to spend time on things like this instead of improving our bots’ responses when a simple <ENTER> after each message would save all this work.

You’ll get no argument from me about that. But the world is what it is, and while we can each make small differences, we can’t change everything to our liking.

Recognising when it is appropriate to reply is (and should be) part of the test. If the software understood what was being said (both in each volley and in the general flow of the conversation) then it would know when to reply just like a real human being does. Let’s not keep looking for ways to dumb down the contest as a way of solving the problems we haven’t solved yet.

I know guys. Just grumbling out loud.

That’s clever Dave . At the same time, the computer can read the sentence as the judge is typing it out, so I would suggest the speed of thought in your fomula.
Luckily, for the time being I am only aiming to get through the selection round without mishaps, and what has been said about typos may also be said about timing.

“Who’s figures”? Is that accepted over “whose” nowadays? I’ll have to completely edit my grammar rules now /o/
One of the biggest obstacles for me in this contest may be the British English

It should really be “whose” but after the judge from the last Loebner came out with things like “howw r youuuuuu”, I guess anything goes.

I have in fact, now that we mention it, already a ‘natural response timer’ in my program, that waits longer depending on how many knowledge searches it took to answer the question. However, this was neither challenge nor anything of importance in opinion, I just did it to animate a nifty ‘brain activity’ bar. In real life, nobody wants to wait for an answer.

And I hope they don’t invite that same judge again, or I’ll do a bit more than just grumble.

I really can’t see him attending after he pretty much insulted everyone including Hugh himself.

Thank goodness. His bad attitude aside though I have to say, I looked up some of his writings, and his views on AI seemed at least reasonable. Anyway, I much hope to see conversations such as there were in 2012 at Bletchley Park, they were probing but civilised. It’s unfortunate that year’s transcripts seem to be offline.

Could you perhaps set flags in your timer? I’m assuming that you have a method that detects when the judge is finished typing, the you begin evaluating what was entered in order to respond, so maybe check the flag during your evaluation and if it is reset then abort the current operation and re-start the wait cycle?

V

@Don & Steve: Yes, the correct word (‘whose’) should have been used. The “Dreaded Grammar Nazi®” once again turns out to be merely Human after all…

@Andrew: Perhaps using things like intentional typos and timed delays may be “dumbing down” the contest, but by the same token, it’s good not to “give yourself away” by answering too “perfectly”, or too quickly. That said, the axiom “All things in moderation” should apply, I think.

@Don: Good point about the bot “reading” the input as it’s typed (though I suspect that this is not the case with most chatbots), so a slight adjustment to the timings of the algorithm should be considered.