Been casually trying to increase my ELO on Wordfeud. When you know the dictionary by heart (Wordfeud has rather restricted dictionaries) and have the basics of bonus hunting and hooking down, Scrabble becomes a pure game of imperfect knowledge and strategy. The known information is the prior letter distribution, the played letters on the board and in your rack. The unknown information is your opponent's rack. However, you can compute exact letter probabilities for his rack and strategize accordingly.
Then the game comes down to managing your rack and your opponent's.
Here's are some common techniques/strats:
- Q-pinning: blocking bonus squares or hooks by using difficult letters so your opponent can't lay down a high scoring turn.
- Q-burning: the inverse, taking a hit on current turn score just to get rid of difficult letters.
- E-burning: sometimes having all-low score letters in your rack is worse than having Q-letters as this will hamper your ability to get high scores on good board positions. A well-balanced rank is key to winning.
- Q-swapping: late-game strategy, maintain a difficult letter in your rack to swap in one of the last turns so your opponent draws the impossible letters forcibly. Remember unused rack letters will be deducted from final score. This dirty trick has won me many close games.
> However, you can compute exact letter probabilities for his rack and strategize accordingly.
I think the best programs run Monte Carlo simulations on the contents of their opponent’s rack. That includes modeling their opponent’s moves, and thus improves on just computing these probabilities.
As a simple example, if they played a move, but, if they had an ‘e’, could have made a much better move, chances are they didn’t have an ‘e’ when they made their last move, even if that, a priori, was still highly likely.
This was the idea that I kicked around for years before eventually writing a Scrabble engine in 2018. But by the time I got to the point where Monte Carlo simulation of the opponent's rack leave became feasible, I questioned the utility. Part of the problem is that if your opponent knows you incorporate such a simulation into the analysis, it's possible to subvert the analysis. And the most obvious counter is to not do the simulation. Another issue is that the advantage gained by opponent rack leave simulation is subtle, and it's costly to run enough games to measure the overall utility. I ended up instead putting my effort into controlling search depth termination with uncertainty estimation (spend more time on the portions of the tree that are more volatile).
I don't think either quackle or macondo do opponent leave inference (elise might though), they just run forward simulations using the pool of unseen tiles.
I got sucked into a variant of Scrabble insanity. In 2018 I spent close to six months writing a Scrabble engine, initially as an exercise to learn OCaml. But the algorithmic challenges pulled me deep down a fascinating rabbit hole. Eventually I was recording IRL games with my wife so I could analyze the positions, trying to understand why/how to exchange more often, and trying to make the engine ever stronger. It still lacks an end-game mode (where perfect play is possible due to empty letter bag), but by the time I put it aside, it appeared to be at least as strong as Quackle. (Fair comparison is hard since more CPU time allows deeper analysis.)
It's a project I hope to go back to someday, and maybe follow up on some of the human-like play options and variants I came up with along the way. The tournament word list is proprietary and at the time there were apparently no reasonable licensing options (licensing has since improved), which put a damper on releasing my engine.
I got Scrabble as a Christmas gift about 20 years ago and I immediately fell in love with it.
Fast forward 20 years and now I have a Scrabble Solver app (open source) in my portfolio that I'm proud of. It still needs improvements to run nicely on mobile but I'm happy with how it runs on desktop.
Maybe someone will find it useful, so I'm sharing here :)
That's a really nice front-end! Really good code structure and dev practices too to my JS-nooby eye at least.
A recommendation for the solver part, it seems your using a word trie instead of the more common directed acyclic word graph (DAWG) or the much faster GADDAG. GADDAGs are really cool, +2x faster lookup than DAWGs at 5x space for common Scrabble dictionaries? Even today's CPU cache sizes fit an English GADDAG with ease. [1]
I am working on a project for reinforcement learning for Scrabble, hoping to learn advanced strategies. When I need a front-end, I will definitely reference your project.
I had a brief interest in scrabble, but when I played against online players and computers, I lost the interest all together.
The problem was that Scrabble dictionary is too inclusive, and the Scrabble game rewards making multiple words. So, instead of learning fun and obscure words, it’s more important to spot how to make viable grids that gives you five words with one move, while involving some high-scored letters. Hence, tons of cheesy 2-or-3-letter words, “qi” as mentioned in the article, “mm”, “hm”, currency units, etc.
Once in a blue moon, there comes a moment of absolute brilliancy when you piece together a long word that pokes into the corner, earning a devastating 80+ points. That feels glorious.
But most of the game I played, it’s mainly short scuffles that dominates the mid and late game.
I wish that Scrabble would reward long words more, and get rid of the cheesy words. No more units, or non-sensical onomatopoeia.
The wind off the Faroe Islands, one, a shout of dismay, a three-toed sloth, and cindery lava. Oe, ae, oi, ai, aa. Learning a few cheese words can up the chances of a good score. The problem then becomes convincing family members that they are real words.
Scrabble rewards you using all your letters with a 50 point bingo. Pretty hard to get a bigger bonus, and using all 7 letters as often as possible is a critical strategy to winning.
We keep a hall of fame list of all bingo bonuses with their scores. It’s a meta game to play once you’ve reached the point where games regularly come down to the wire in the mid to high 200s and hunting for those two letter words becomes a boring chore.
I wonder if dominos on a scrabble board, or something like that, could be fun. Anything to replace the spelling. The game described in the article:
> There is something magical, however, about realizing how to play Scrabble even one step above beginner: using the bonus spaces, creating overlapping words, hitting your first bingo. You also learn how to play defense.
> The first time you realize that your opponent is stuck with the Q and you make a crucial play to deny her the legal two- word layup QI, securing yourself the victory, well— that’s magic.
almost sounds like a fun game about territory control or something, but I’m not about to memorize a bunch of random letter combinations to play it.
The two letter words and the common q-without-u words are enough to unlock a much more entertaining version of the game than the one you can play with a normal vocabulary.
Since you already know most of them you're only memorizing a few dozen very short words.
Then a few vowel dumps wouldn't hurt, short V/X/K words...
Also, consider just printing out the legal two letter words and leaving them there for anyone to check ... Saves on the "win by memory" and allows more tactical play by all.
I've been playing like that with my parents (in their 80s) and it makes for a much more fun game!
My wife and I used to play regularly and printed out a list of two and three letter words that we could refer to under our rules. It helped us learn those words.
When I was younger (I've not really played since) we played by a word being valid if you could state what it means and use it in a sentence. An Oxford English Dictionary was used to check. For domain specific words other references were permitted (books on computing for instance, or an encyclopedia). Acronyms were permitted only if in common use in that form. Local slang too, I supposed in modern times adding Urban Dictionary to the references would cover that.
But the important thing was to state the meaning before looking up the word, not just using the reference list to pick out combinations that work with your letters. It being a real word did not count if you didn't understand the word.
This wouldn't work as well for online gaming, I expect. And of course you can game it by memorising the meanings of some obscure words, but that (expanded vocab) was part of the point.
Someone who reads is a reader, viewer, writer, dancer, God loves a tryer, and someone who pares should be a parer.
Although I often try that pattern and have it denied, if you can shelve a project it suggests shelver which is in, if you can table a motion it suggests tabler which isn’t.
It's a logical word (and, as I recall, I eventually went with it) in the sense of someone or something that pares. https://www.thefreedictionary.com/parer But it's fair to say that it's not especially common in the US. Someone would normally say they were going to the manicurist or ask you to hand them the paring knife--though, at least in the latter case, they might intuit the meaning from context.
One of the clever things about Wordle is a very well-calibrated difficulty level, at least for native English speakers. At least for non-hard mode (i.e. can play a word that you already know can't be the final answer) and using a good start word, it requires some luck to get the answer in less than four but you won't lose (i.e. more than six) very often. And it's usually pretty quick.
Thinking about whether something is a likely answer also introduces some strategy itself.
In spite of being something of a professional writer, I don't usually like word games (including Scrabble) much. But I do like Wordle--although I quickly tired of the multiple word variants.
I'm convinced the purest way to play Scrabble is with siblings or close friends who will kick your ass if you get too cute with the "technically permitted" words. Keeps things in check.
I wonder about going the other way, adding more short words that make the game flow better (how about VI? Or just allow any two letter combination?).
My experience having briefly gotten into competitive scrabble 15 years ago is that memorizing the short words is the price of admission for a more fun game. Knowing the short words makes it possible to fit the long words on the board.
Part of what I liked about older scrabble dictionaries (you can use any dictionary you like) was the variance in difficulty in using various letters. Get an 'X'. That is a nice letter to have with high scoring potential with lots of two letter words. A 'J', not so much with only JO as a two letter word. Z, high value with a large word set, but can't go two ways (no ZA). Q, high value but a bit of a pain in the ass and you need a U. Maybe you should turn it in. Now with QI, QAT, etc., the Q is not as cool. But nothing says you have to accept those words. Often we don't play with them.
As suggested elsewhere, we play with a two letter cheat sheet to make the game more fun for those that don't know all those words.
I thought I was good at scrabble but then I realized that playing against my parents does not count. Experts just memorize dictionaries. Just anther example of a game that has been optimized. Would not surprise me if we see a cheating scandal similar to professional chess.
I play Polish Scrabble competitively, and IME there’s much more to the game than just memorizing the dictionary. A program that has full knowledge of the dictionary and always plays the highest-scoring word will easily yield to a human.
There’s virtually no strategy, but tactics is of paramount importance. Things like keeping your rack balanced; which one of several similarly-scoring words to play, based on your knowledge of the game state; when and what to exchange, etc. These factors are all at play very much intuitively, and are learnt with practice over the years.
Polish is especially fun to play because it’s highly inflected, meaning that:
- the number of unique words in the dictionary is about one order of magnitude greater than in, say, English or French – it’d be much harder for Nigel Richards to learn them all;
- keeping an eye on certain combinations of letters (that might form an inflective ending) is much more important.
Plus, there’s a small but vibrant community, with its very own history and folklore to be proud of. :)
My wife is Polish and we tried playing the Polish Scrabble. Allowing for English or Polish words. Took me a few games of getting trounced to realise the letter set is completely different in the Polish version.
Computers are better than humans at Scrabble [0], so the possibility exists for cheating even if their advantage doesn't stem only from mastery of the dictionary.
maven was decidedly not better than the best humans, and even today the best humans are better than the best AI players out there, though the gap is narrowing.
I think the most interesting thing about competitive Scrabble is that it's not really about memorising dictionaries. Once you assume that everyone knows every playable word, the real distinction is in the tactics. There's elements of statistics to determine what your opponent is likely to have, whether to play offensively or defensively, etc.
It's a very different game to "family" Scrabble, but we always had the etiquette rule that you should know the definition of a word you played. It isn't enough to know it's in the Scrabble dictionary.
I come back from time to time to the thought of how exactly to define the threshold where games are fun while you're still improving. I hate playing with people who take the experience professionally, but having some strategy involved is definitely fun.
Nice games are the ones where rules or some important factors are modifiable.
The way we played Quake was by taking hundreds of user made maps and playing them one by one untill they got stale.
It's completely different way of playing than what modern games do. Basically learning just a handful of maps by heart up to specific timings, sound and few pixel spots.
Basically rigidifying naturally variable parts of the game to make it more boring and repetitive for the sake of creating controlled conditions of competition.
Every year my family plays Scrabble on Christmas Day. The score book goes back about a decade. Until 2016 my sister was ruling champion.
Enough was enough. I installed a Scrabble game on my phone. Played every day and took the crown at our next Christmas Scrabble game. The win thoroughly surprised everyone.
That year I had read Carol Dweck’s Growth Mindset. Which the Scrabble win nicely illustrated.
How's your sister? So many years of winning might make your sister incorporate being a scrable champion as a part of her identity. And nobody feels good after part of their identity gets shattered.
Yeah, any kid who owned the only Sega Genesis in a friend group knows practicing Street Fighter too much will just mean no-one wants to play it with you anymore.
Same goes for memorizing useful-in-Scrabble words that occur on Scrabble boards 100x more than they do anywhere else.
> My mood swung on the outcome of each contest, which is heavily subject to chance, what with the letters emerging randomly from the bag
I saw a fun documentary on competitive Scrabble tournaments and I've always wondered if tournaments have considered using computers to randomize the tile draw but constrained to equalize the "luck" between players in a game. I know that competitive Rubik's cube tournaments use a computer to generate a starting cube pattern that's not purely random in that it eliminates easy solves.
I guess the simplest form would be ensuring each player in a game draws an equal distribution of letter-points in a game but I'm sure there are more complex ways to be random but weighted to ensure equal opportunity in a match. What I don't know is if reducing the variability of draws would even be considered an improvement at competitive levels.
There's an infinite number of methods to balance luck and they touch on our conception of fairnes and competition itself.
For instance, which is more fair, letters balanced by pointsbanf partitioned before the game starts, or weighted letter draws based on the current board state and the potential number of scorable points per player? And perhaps more importantly, why?
Try them all and see what they do. I got into simulating games as a kid based on arguments over what the effect of various table-rules were on games. (Spoiler: The effect of every optional rule in Monopoly is to make the game take 5x as long.)
Maybe you could spend points to pick two letter and keep the best...
Why? Because then 1) you'd know what that rule would do, and 2) maybe it'd be neat! Like, maybe you'd find an easy way to provide more balance between newbs and experts, or minimize the variation in score against a given opponent, or whatever.
What's interesting to me is that it's not just anagrams. It's anagrams with letters alphabetically sorted.
So you learn the key, the sorted word, and associate with it the sensibly ordered version (calling it sensible might be a bit of a stretch for some weird words or words in foreign language).
As you play, you sort your letters and try to match then with learned keys by omiting some of the letters and possibly including some letters that you see on the board.
Then out of all the options you have you try to pick the one that's most beneficial for you at this point according to game mechanic because it scores a lot for you, or prevents oponent from scoring, or leaves you enough letter for to score next round.
But you can imagine that memorizing as many pairs of (sorted letter, word) so that they come to you effortlessly is pretty much basic requirement to play the game any good.
If anyone is interested in trying out a novel variant of Scrabble, please check out my game Lexatious, available for Android and iOS. Can be played single-player (against a bot) or two-player.
I’m always amazed at those that can have such singular and deep pursuits and obsessions. I have been watching some video game speed running videos, and they seem to be of a similar makeup. It’s just not the way I’m wired, so it’s always fascinating to read about.
> Once your brain is attuned to anagrams, you start seeing them everywhere.
I had a moment like this when I was obsessed with Scrabble — studying and playing 20+ hours a week.
I saw a guy at work with Demetrius on his nametag. Instantly I saw that “with another U you could make ‘deuteriums.’” It was the weirdest feeling because I had no idea where that realization came from.
I lost probably 4 years in my 20s to Scrabble and then quit cold turkey. I’m about 40 now and I’m pretty good at Wordle and the other NYT word games but that’s about it.
Speaking of Scrabble, is there a decent online version for playing against other people? My wife loved the version that worked on her iPad, despite some clear bugs. I understand that it got killed off, to be replaced essentially with "Words with Friends", with all sorts of infantile blinking animations and unrelated cruft.
She loved being able to play long-distance with family members, and even met some new people to chat and play with from other countries.
I think a 3D version of scrabble would be fun, where you build the words using cubes in 3 dimensions instead of two. You could probably even do it non-digitally if you had magnetic cubes
Then the game comes down to managing your rack and your opponent's. Here's are some common techniques/strats:
- Q-pinning: blocking bonus squares or hooks by using difficult letters so your opponent can't lay down a high scoring turn.
- Q-burning: the inverse, taking a hit on current turn score just to get rid of difficult letters.
- E-burning: sometimes having all-low score letters in your rack is worse than having Q-letters as this will hamper your ability to get high scores on good board positions. A well-balanced rank is key to winning.
- Q-swapping: late-game strategy, maintain a difficult letter in your rack to swap in one of the last turns so your opponent draws the impossible letters forcibly. Remember unused rack letters will be deducted from final score. This dirty trick has won me many close games.