Thursday, September 11, 2008

Broken Chatroom

I found these screenshots on Mobygames. They are quite funny because they illustrate all the problems of my game, Chatroom.
Okay, not that painful yet. This first screenshot illustrates Luke's inability to recognize common acronyms, specifically conversation starters. This is not a major flaw in the dialog system, because there is a chance that acronyms aren't recognizable to everyone in the population, or that they will die out in another 88 years--though it is highly unlikely.

Okay, now there are some clearly evident problems. The first 3 lines illustrate Luke's inability--whenever I say Luke's inability, I really mean my inability to program a proper, functioning AI--to recognize specific subsets of a conversation topic. For instance, 'sculpture', 'drawing,' and 'painting,' for the art topic. Luke would only recognize the art topic, and perhaps the drawing or painting section of it. That's all.

The second problem here is that MobyGamer is asking 3 questions simultaneously, while Luke only responds to the last question. Every other question asked before in the chain is cut off for the last question asked. Somewhat realistic, but not always practical.

The third problem is that the parser doesn't recognize specific questions about a topic, for instance, art. MobyGamer asks "why didn't art school accept you?" but Luke just gives the generic "CHECK FOR 'ART'" response, where he just talks about art in a general sense.

The fourth and final problem contained in this screenshot is Luke's phrase "I'm not going to do your math" which is a bit random. I had it programmed in so if the player typed any of the following ("+, -, =, *, /") math symbols, Luke would give a generic response about not doing math, because he's not a bot or something. I didn't expect people to use hyphens however in a general sense, like MobyGamer did. And as such, you get this awkward, seemingly random response.

Once again, the awkward hyphen detection causes Luke to talk about math in the fourth line of this screenshot.

Second problem here, the player talks about transportation, to which Luke responds "You kidding me? There are none of those left." This is the transportation response, but it was more intended to be for the keywords, "car," "boat," "plane," etc. As such, it doesn't really make grammatical sense right now.

The last problem is that Luke gives a generic response whenever you mention 'war.' This leads to a repeating of the same response over and over, in slightly different phrasings.

The pain is almost over... last screenshot:
I'll just skip all the awkward replies and go down to the bottom of the chatlog where there is one problem with Luke referring to his age again. This was because the previous line, "so they had us wage war?" contained the keyword "AGE" in the word "WAGE." This led to Luke responding about his age, which was again, totally random.

So, yep, Chatroom is far from perfect. I could probably spend more effort on it, and fix some of these problems, but I'm too lazy and I'm not sure the workload would justify the end means. It also would've helped if I used a parser-based engine, like ADRIFT or Inform, or whatever is out there, instead of a graphically based engine such as AGS.

I still had tons of fun making it, and it seemed to garner a lot of publicity once again, just like La Croix Pan did. The End!

Labels: , ,


At 9/11/2008 3:24 PM, Blogger endeavor said...

It'd be cool if you could someone use a thesaurus to catch multiple queries. But for that you'd need internet connection and a way to parse web search results.

At 9/11/2008 3:43 PM, Blogger TheJBurger said...

I'm not sure what you mean.

At 9/11/2008 3:47 PM, Blogger endeavor said...

For example, say someone types in "architecture," but "architecture" is not a key word that you look for. In that case, your game would query a thesaurus website and look for words that are similar to "architecture." Say you find "building" and "building" is one of your key words. With that you can have your character say something about buildings.... That's probably not the best example, but do you get what I mean?

At 9/12/2008 12:43 AM, Blogger TheJBurger said...

Yes... you use a pre-existing, external AI system to catch any parser entries that are not recognizable in game. Then compare those to any recognizable entries using an out of game program. After that, go back into game and relay that answer into the most formidable response, closest to the entry the player typed.

At 9/12/2008 2:27 PM, Blogger ifedajay said...



Post a Comment

<< Home