The Tile Bag

by James Leong, Editor

Dear reader: Thank you for turning your eye to the first edition of the Tile Bag. I’m James Leong, your columnist, whom you can refer to as Mr. Tile. I’m here to answer your questions about anything SCRABBLE: Learning words, strategies, computer programs, rules—the works. If there are questions I can’t answer, I’ll get in touch with those who might and, hopefully, we’ll learn something together.

Anyway, without any further ado, let’s get to some questions. It seems today’s theme is technology — specifically, computer programs. Let’s start at the beginning: A.

What is Aerolith? – PM

Aerolith is an online tool developed by Cesar del Solar to help review words. It is a program you can install on your computer that allows you to access an online server where you can quiz yourself on jumbled words, whether by length or other interesting characteristics. These jumbles appear in sets of 50 words, and you solve them one by one within a specific time limit. If you do not finish before time is up, you can then review which words you have missed. One specific benefit is that you can unscramble these jumbles with friends or in daily competitions against others, to see how well your abilities compare to the field. Aerolith also provides definitions of words, plus front and back hooks. A similar program to Aerolith is JumbleTime, developed by David and Oliver Johnson, which differs mainly in display and somewhat in content.

Basically, these programs are designed to help improve our anagramming skills. Aerolith can be found at www.aerolith.org and JumbleTime at www.jumbletime.com.

Skipping all the way down to Q, here are some questions about our favorite duck.

Walking through an endgame on Quackle tonight, Five-Minute Championship Player offered no choices for blocking my opponent's only option for a Q play. By sticking the opponent with the Q, I lost the game by less than any conclusions CP initially gave. Why is it that sticking the opponent with the Q didn't seem to factor at all into CP's strategy? – TF

When checking to see what move Quackle deems "best" early on in a game, what is a sufficient number of iterations? For each number of plies (2-6, Many), how many iterations are sufficient for Quackle to decide which move is best? - KJ

Well, two questions about Quackle! Let’s have a small primer for everyone. Quackle is a program developed by Jason Katz-Brown and John O’Laughlin that can help players review their moves in games they have played. For each move, Quackle can spit out a set of possible moves and evaluate which it calculates are best, according to how likely it determines each move will result in a win. If the game is decided, Quackle will determine how best to augment spread in a given direction. One can also play a Scrabble game against one of four inbuilt opponents (in a variety of languages). Quackle can be found at www.quackle.org.

Older versions of Quackle sometimes did not play the endgame perfectly, particularly when there were only a few tiles left in the bag. With respect to TF’s question, these mistakes were made more often than not with an empty bag. Quackle calculates every potential ending move, and picks optimally — assuming everyone plays perfectly. A Q-stick, for instance, might initially seem to be the best option, but there are many ways your opponent could respond to score well in return. Maybe that didn’t happen in this case, and as a result you lost by less than what Quackle had predicted, as it had factored that your opponent would have made moves that did not occur.

With respect to KJ’s question, let’s describe how Quackle decides a move is best. Basically, Quackle will create a series of trials comparing the outcomes of a move for a certain number of turns thereafter, called “ply.” Each trial comparing the outcome is called a “simulation” (or sim), as it randomly generates what your opponent has and what you next receive, and plays out the game for those turns, using its computerized evaluation to pick what it sees as the best moves. At the end, it will tell you the move it will pick for you in that moment (mind you, only for the ply you restrict it to).

Quackle uses its own standards and abilities — that is, it can find every play for each side, regardless of how impossible the play might be for a human being to find. As the ply increases, this deviation between computer and human play grows. This means that unless you play just like the computer (and your opponent does, too), the predictive capabilities of Quackle in a person-to-person setting worsen the more you want the program to “see into the future.”

Keeping this in mind, if Quackle tells you that one move will win games, say, 80% of the time, and the next best move will win 30%, you really won’t need a large number of simulations to back this up because the drift of those statistics is not going to vary by a large degree. When Quackle tells you two moves are basically tied in how often they will win, and the margin stays within a small band for your definition over simulations, this is a time your intuition is likely to help, because as a person you have a leg up on knowing how people act. Sometimes Quackle will not pick the best move immediately, and so you might need to type in the move you prefer and then run the simulator to see how it sorts out — usually until the numbers stop jumping around by huge overlapping margins — which often is a short time.

Where a large number of simulations might be handiest is where the win percentages of all moves are very small to begin with. For instance, if the leading move wins 4% of the time and the second best wins 2% of the time, it basically means you are doubling your chances of possibly winning if you pick the leading move, and thus making sure the numbers are accurate matters a whole lot more than, say, with moves that have a 98% vs. 96% winning probability.

Basically, my suggestion is to give Quackle a try. If you like what it tells you, use that. If you don’t and have good reason to dislike its choices, then it might be best to disregard its advice. Beyond a point, the number of simulations isn’t going to tell you that much since Quackle very rarely changes its mind once it sets on a path. With respect to ply, remember that computers and human beings aren’t one and the same and that whatever accuracy you seek with simulations might be lost by the logical differences between the two entities.

Got a question? Don’t hesitate to write, no matter how strange or silly it may sound! Just fire away to tilebag@gmail.com.

James Leong is a top player from Vancouver, BC, Canada. He was the winner of the 2007 Players’ Championship in Dayton, OH.