Talk:20Q

From Wikipedia, the free encyclopedia
Jump to: navigation, search

Before exploring more in the hand held version of 20Q it might be noted that there is a larger version with a base of about 8 inches that supports a purple dome on which the questions and guesses are projected by means of some motorized mechanism and light source. The games is controlled by an on and off switch under the unit and a hole for reset to be activated by a paperclip or similar item. Reset is for use when static electricity causes the game to fail. It has buttons for Unknown, Rarely, No, Yes (which also functions as "on" and "new game"), sometimes, undo (if the player incorrectly answers the question the player can press undo and go back to the question to change the answer) and sound on/off. The domed version always starts with the selections of Animal? Vegetable? Mineral? Other? or Unknown? The toy units ask 20 questions then displays its guess. If the item is not identified in 20 questions, the games try five more questions and guesses again. There is also a "big" size model which is rectangular.

That having been said, it may be sufficient to say that: The Twenty Questions game is also the basis for plastic toy versions by toy maker Radica (r) under licence from Radica Games Ltd. which produces versions of a game called 20Q, a trademark of 20Q NET, INC. The company also has an online version of the game which as of this writing is found online at WWW.20Q.COM. Also see WWW.RADICAGAMES.COM. The company also has an online message board. The plastic toy versions do not learn from play but the online game keeps track of answers and "learns" from being played.74.67.49.203 07:06, 3 October 2007 (UTC)

Existing (past/present) languages for the handheld version[edit]

I am trying to record for which languages a handheld version is/was available. I've got the Italian version (found in a toystore on the airport in Venice). It is the real thing, because of the brand Radica: and references to the official websites. According to the official site, only English, French and German versions are sold. Bcurfs 10 May 2006

20Q: Italian was removed form the website at the request of the Italian distributor. Dutch was among the first languages translated and the product should be in stores in the Netherlands.

And indeed, by now I found the Dutch version. Besides the previously mentioned languages (including Italian) no other languages seem to exist. It be mentioned that the online version of 20Q features the following languages: American, British, Canadian, German, French, Spanish, Italian, Chinese (2 languages), Dutch, Hungarian, Czechian, Greek, Danish, Swedish, Portugese. Bcurfs 21:51, 30 May 2006 (UTC)

The other languages exist on the website. The 20Q AI plays online at www.20Q.net in: Canadian English, British English, American English, German, French, Italian, Spanish, Dutch, Swedish, Polish, Portuguese, Danish, Finnish, Hungarian, Greek, Czech, Cantonese, Mandarin, Japanese, Korean (in test mode). The pocket games are sold (generally in the market for their specific language) in: English, French, Italian, German, Japanese, Chinese, Dutch. There are new specialty versions of the handheld game available in the United States and Canada. The toy is a bonus, the online game continues to learn and the original game played its 44,843,491st game today. Someone other than me launched this page about 20Q. I try to check in from time to time to make sure the information is accurate.

Functioning of handheld version.[edit]

It seems a bit unknown how the handheld version works.

For one thing, what is the reset button for?

The reset button restarts the game.

I read somewhere that it is stated that the handheld version does not learn from the responses it receives. Can anyone confirm / deny this based on fact or documentation?

Correct, the pocket games do not learn; the cost to produce the chip that would contain a learning knowledgebase would be very expensive.

I can't think of a proper function for the reset button if its internal information structure does NOT change with time. Bcurfs 10 May 2006

I contacted 20Q via the form on their website www.20q.net and got an answer to these two questions. 20Q: The reset button on the handheld "reboots" the chip, starting the program over in the event the game locks up. The buttons on the left and right of the product adjust the speed of the text scrolling. 20Q: In order to keep the cost of the product low, the handheld does not learn. We are considering this feature for future products. Bcurfs 21:42, 30 May 2006 (UTC)


NPOV note[edit]

The advertorial tone of the handheld section has been revised.20Qer 21:17, 3 November 2006 (UTC)

Over time, I'm learning more about editorial policies, however, I'd like to know how long it takes to have the messages that appear at the top of the article about cleanup, NPOV, etc., removed by the unseen, unnamed Benevolent Wikipedia Editor20Qer 21:17, 3 November 2006 (UTC)

How does it work?[edit]

This article should point a curious reader in the right direction to really learn how 20Q works. It links to artificial neural network, which is all well and good, but not enough. Even neglecting the learning aspect of things, how can a neural network take a series of yes/no answers and (a) decide on the next question to ask, and (b) guess what the answer is? How much of the algorithm was hand-coded and how much did the neural network figure out? —Ben FrantzDale 20:51, 26 April 2007 (UTC)

Very good questions. My pure conjecture is that it is not a standard neural network in configuration. Rather it consists of a huge matrix of questions and objects. It starts out with the special hard-coded question of "Animal, Vegetable, or Mineral". Given that answer it ranks the possible objects according to the number of times a person has identified the object in question in that category. At no point does the system ever rule any object out (unless it asks that object, and you say no), but it keeps tracks of a probability for every object. Then it randomly chooses a question from the list of questions not yet asked, weighting the chosen question according to how it is to narrow down the list of most probable objects by approximately 50%. Updating the probabilities of every object according to the question result. This continues until it is reasonably certain about the object in question. At this point it will start asking some other questions about the object, attempting to gain more information about it. (Only in versions with learning enabled, such as the online version) These new questions are often quite noticeably different from the previous questions, and are normally questions that it has very few answers for this particular object. Then it will guess the object. At the end if it has successfully determined the object, it will update the matrix to include the responses given for each question asked (so it might add one to Matrix["Hot dog"]["Is it edible"][YES], and Matrix["Hot dog"]["Is it commonly found in nature?"][NO], etc.) 66.254.242.239 (talk) 21:59, 18 April 2009 (UTC)
Also keep in mind that they are not just yes or no questions, but "Yes/No/Irrelevant/Probably/Doubtful/Sometimes" questions. That helps a bit. 66.254.242.239 (talk) 22:03, 18 April 2009 (UTC)
Just see the patent, it is covered in some detail there. pgr94 (talk) 22:31, 18 April 2009 (UTC)


I understand that the objects are learned by users who play the game. However, where are the questions learned from? Full Decent (talk) 20:20, 29 July 2009 (UTC)

How smart can it be?[edit]

You were thinking MAMMOTH! does it have a trunk? you said yes. 20q thinks: doubtful. >.> —Preceding unsigned comment added by 64.122.63.142 (talk) 19:34, 10 December 2007 (UTC)

The problem with this is, that most people do NOT think of a mammoth and when asked the question 'Does it have a trunk?', the player may interpret this word as the trunk of a car... and the program may even ask this question in the series of questions when the player is thinking of a CAR (which is a much more popular word). In other words, there is no guarantee that any question has only one meaning. In fact this is almost certainly not the case for any question. The program only tries to narrow down its search and if the answer should be 'yes', this does not completely contradict doubtful. It is pretty smart of the program to then conclude that a mammoth 'doubtful' has a trunk. To the player this indicates that there is some evidence of a mixup of words. The program will reflect this mixup in 'strange' conclusions. But the players are the ones that give the question different meanings. Bcurfs (talk) 07:07, 12 February 2011 (UTC)

Misspelling of "20Q"[edit]

I would like to take this time to remind editors of the proper spelling of "20Q". Many misspellings of the name are "2oQ". Please do not make this mistake. the name consists of the number "20" and the letter "Q". I have seen the title misspelled many times on many pages on Wikipedia. VmKid (talk) 00:05, 6 July 2009 (UTC)