Many heuristics can be used to distinguish moves that would make the task of winning the game harder for a human opponent.

The idea would be to find moves that put the opponent in one or many of the following situations:

- There are few winning moves
- Winning sequences are long (6+ plies, say)
- Winning moves score much less than the highest scoring moves
- Winning moves do not block a huge play for the opponent
- Winning moves keep a bad leave
- Winning moves are hard to find, either because:
- They make use of no coloured square
- They involve long words (9+ letters)
- They involve words with atypical sequences of letters
- They make use of a blank as the first letter of the word played
- They make use of a blank as a heavy letter
- They involve hooking on multiple (possibly non-adjacent) letters already on the board
- They involve playing on both sides of a (3+)-letter word already on the board

- Et cætera

It is player A's turn and he is trailing by 84 points.

Player A's rack: DDKOSUY

Player B's rack: AEEOOOQ

In this example, player A has no winning move. However, one move stands out as giving him the best chances of stealing the game against a human: [DE 4J 3 DKOSUY], with the idea of playing [DO I3 9 KSUY] and [KYUS H1 78] next. This move stands out because:

- Player B must play [AÈDE 4H 6 EOOOQ] to win the game
- If player B plays [AÈDE 4H 6 EOOOQ] then he keeps a bad leave
- If player B plays [AÈDE 4H 6 EOOOQ] then he lets player A play the lucrative [KYUS B11 48 DDOU]

Indeed, the setup that player A wants to execute after [DE 4J 3 DKOSUY] is subtle and player B might be tempted to block [KYUS B11 48 DDOU] anyway by playing [OU B12 2 AEEOOQ], getting rid of an O and thinking that player A would be stuck with one of his two heavy letters - The winning sequences are relatively long and precise calculations are required

Pre-endgames are situations where there are a few tiles left in the bag, say between one and four. Our endgame solver could be used to solve such problems.

We could have a mode that finds the moves that win all the time and more precise modes that would give statistics about each move, such as:

- Winning percentage, with details about each eventuality
- Worst possible scenario, e.g. it might be crucial to know that, given optimal play by both sides, a move will let you gain at least X points over your opponent
- Expected value and variance of the spread differentials
- Et cætera

Without too much additional efforts, we are already able to work out the winning moves in pre-endgames where one tile is left in the bag.

This example between Nigel Richards and John O’Laughlin is taken from here. Nigel had a 22 points deficit.

His rack: ?AENNTU

Unseen tiles from his point of view: AAEEIILQ

Nigel played [ENAUNTEr B4 61] which would have lost him the game had he drawn the Q.

Our analysis shows that at least two moves were winning no matter what letter Nigel would have picked: [AUNT C6 16 ?EN] and [UR 7C 3 ?AENNT].

This example between Pakorn Nemitrmansuk and Nigel Richards is taken from here. Pakorn had a 22 points deficit.

His rack: ?DLRSTU

Unseen tiles from his point of view: ABDEEOOO

Pakorn played [SOD I1 6 ?LRSTU] and won the game.

Our analysis shows that exactly three moves were winning no matter what letter Pakorn would have picked: [SOD I1 6 ?LRSTU], [OD 12L 2 ?LRSTU] and [OR 12L 2 ?DLSTU].

This example between Pakorn Nemitrmansuk and Nigel Richards is taken from here. Pakorn had a 44 points deficit.

His rack: ?GINNWX

Unseen tiles from his point of view: AEIILOTT

Pakorn played [DRENCHING 15G 48 ?NWX] which would have lost him the game had he drawn the O.

Our analysis shows that one move, and only one, was winning no matter what letter Pakorn would have picked: [WING L3 46 ?NX].

This example between Nigel Richards and Jesse Day is taken from here. Nigel had a 13 points deficit.

His rack: AAHIRTY

Unseen tiles from his point of view: AEELLNSU

Nigel played [ARIGHT L2 20 AY] and won the game.

Our analysis shows that the only three moves that won all the time (no matter what letter Nigel would have picked) were [ARIGHT L2 20 AY], [AIGHT L3 18 ARY] and [AY A14 32 AHIRT].

When too many tiles are left in the bag, precise solutions are unreachable. However, heuristics and simulations can be used to get a good idea of what the best moves are. Here are some examples of information we could try to provide:

- Which moves block the most bingos for the opponent?
- Which moves open the board the most, in the sense that the number of playable moves with the letters not already on the board is maximal?
- Illustrating with a heat map, if it was the opponent’s turn, on which square would he likely score the most?
- Which moves offer the “best” leaves based on the remaining letters and the current board position?
- What kind of rack might the opponent have based on his last play(s), assuming that he plays sensible moves?
- Based on the unseen letters and the current board position, what is the probability that the opponent has a play of at least X points?
- Based on a given rack and board position, and assuming you can play as many moves as you want in a row, what are the most lucrative setups?
- Et cætera

As far as we know, the game of scrabble is well behind the game of chess in terms of in-game exercises. Indeed, most chess sites offer puzzles, and some even specialize in that, e.g. chesstempo that offers thousands of them and categorizes them. What our community does have, though, are tools to train with anagrams and practice in the duplicate format.

We have the tools to automatically generate endgames, solve them and categorize them. Therefore, we could offer exercises in a website for players to dynamically practice endgames.

Everything described above could be part of a software that would allow you to train, play against the computer and analyze your games. We would make sure that players of all levels would gain something playing against the top engine.

Return to home page

This site uses web cookies, click to learn more. Our privacy policy.
© Ortograf Inc. Updated on February 23th 2021. Informations & Contacts. |