Congratulations Bruce! I won my bet.

Robby.

Well done, Bruce!

Congrats to all the competitors. Just getting to the finals is a noteworthy feat.

Many congratulations Bruce on your 3rd win.

It was a disaster for Mitsuku this year, as the network at the Loebner kept sending packets of data too early and splitting many of the judge’s inputs into separate messages. One of the judges asked Mitsuku “What is your favourite book?” and it turned out as below :

Judge: What is your fa
Mitsuku: Are you asking about my fa?
Judge: vourite book?
Mitsuku: If the universe is expanding, what is it expanding into?

This affected all the bots (and the webcast too it seems) but Rose was able to deal with it best and so not only is congratulations due for winning but congratulations also for being ranked in 1st place by each of the 4 judges.

I’ve posted the transcripts for Mitsuku here: http://www.square-bear.co.uk/mitsuku/loebner2014.htm Not sure if there are any official transcripts yet.

I’ll get you next year Bruce!

Congratulations Bruce! And I trust the rest had a good time too.

Hh el l o h o w a r e y o u t o d a y ?

Round 1, question 1, cheap shot right off the bat *sigh*. This is more a task for visual text scanners or phonetic voice recognition. Or one could make a nonsense detector to ignore it and always start with a proper greeting.

the network at the Loebner kept sending packets of data too early

Well that is concerning. How does that work exactly? It sends over a bunch of letters in a spurt and holds the rest back for seconds long so that chatbots thought they’d stopped typing? Bruce’s entries keep an eye on punctuation as well as presumably still have a more lenient timer.

Don Patrick - Nov 17, 2014:

Well that is concerning. How does that work exactly? It sends over a bunch of letters in a spurt and holds the rest back for seconds long so that chatbots thought they’d stopped typing? Bruce’s entries keep an eye on punctuation as well as presumably still have a more lenient timer.

Spot on. I could see on the console for Mitsuku that 5 or 6 letters were being sent at once, then a break of a few seconds then another 5 or 6 letters until the message was complete. This was ok if the break was just a second or so but anything longer than 3 or 4 seconds and Mitsuku assumed the judge had finished typing and so started her response.

I looked at the other bots and they seemed to be doing the same. It looked like Rose was also doing the same, as I saw lots of inputs like “even during the” which indicated to me, a mid sentence but Bruce must have coded a way round it.

I use a break of 3 seconds to detect end of input and so this killed me.

Don’t get me wrong though, it was a great day and was good to meet up with “fellow geeks” as well as James May from the telly. My wife was insanely jealous when I phoned her to say that he was there. Hopefully, the programme will be shown on the BBC.

I am very positively impressed by the quality of Mitsuku conversation.
You did a great job these Steve, I see that when looking at the round 1 transcript.
I see what you mean when looking at the transcripts starting at Round 2…

Can you describe exactly what happened here?
Is that from the way the things have been parsed?
Please, let’s get into the details here.

One of the problems early on with the LPP judges’ perl components is that they would max out the CPU usage. Has that been fixed? Windows 8 or whatever they used this year may not dedicate the same amount of resources to the task at hand.

I think it was a network issue rather than CPU, as when the judge program and bot program is on the same machine, this problem doesn’t occur. I was speaking to one of the organisers who said that the qualifiers were held on the same machine and so no network was involved.

If only we were allowed to use <return> to indicate end of message, this problem wouldn’t occur but I guess it’s something for entrants to try and cater for next year’s contest.

If your program can do LPP it can handle speech recognition easier too!

Back in 1990, when the LPP was first written, hardware specs were far, far less than what they are now (CPU’s are actually 85 times faster now, on average, and RAM capacity is over two thousand times greater), so I seriously doubt that hardware performance is an issue. Networks, however (even the gigabit networks we have today), are still favorite homes for gremlins, ghosts and other invisible nuisances, and a single buffer overflow on a network switch can foul up communications quicker than you can turn around. It’s disappointing that these things happen at what seems like the worst possible times, but <yoda> happen they do, and prepare for them we must; yyyeeeesss… </yoda>

LPP?  I just read the spec and my head exploded.  Creating directories in order to transmit text??!?!!?  Why not use XMPP or some other standard chat protocol?  What the heck is the reasoning behind this weirdo protocol?

One thing you must bear in mind is that the protocol was devised back in the 1990’s, and that back then, there wasn’t such a thing as a “chat protocol”. In fact, there wasn’t much of anything on the web beyond a few thousand simple HTML pages, at that point in time. The LPP is, without question, an “odd duck”, and is certainly a shining beacon of “old school”, but it is what it is, and using it is the price of admission to one of the most universally famous chatbot competitions on the planet.

Perhaps we, as a community, should try to open a discussion with Dr. Loebner, with the goal of devising a new protocol that takes advantage of improvements in security and efficiency (not to mention improved simplicity)?

Steve Worswick - Nov 17, 2014:

one of the organisers who said that the qualifiers were held on the same machine and so no network was involved.

See? As I had assumed . Still, Will Rayer was the wiser to test his entry with a network connection. I have no idea how to even establish one.

Steve Worswick - Nov 17, 2014:

but I guess it’s something for entrants to try and cater for next year’s contest.

Nah, I have to disagree that technical difficulties with the provided hardware is the entrants’ responsibility. Next we’d have to bring our own power supply in case of a blackout. I think Dr. Keedwell is professional enough to agree that the organisation will have to solve this issue, as there have been plenty of contests where this was no problem.

Windows 8 in particular has shown a noteworthy delay of about one second with the LPP running unnecessarily full steam on the same computer as my program, but I do believe a network could add to that. I’m pretty sure the organisation would welcome us programming a better, faster, prettier LPP, but personally I’d need a good reason to sacrifice a week of my time for a show I’m not in.

Though thinking about it, if we could have the programs run on the judge computers and just overlay a full-screen interface to hide them from view, then patching the keyboard keys through to a Windows program would be a sinch.

That’s the ideal solution Don. The judge program can be maximised to cover the bot workings behind it and would remove any network issue.

Congrats Bruce!  I haven’t seen your transcripts, but the vote was unanimous this year singling out Rose for the #1 spot.  Nice job!