Hangman2020/01/20

Rules of the Game

The game of Hangman involves one player thinking of a word or phrase, and another player guessing that word by guessing letters one at a time to reveal information. If the guesser guesses too many letters incorrectly, that player loses.

What is the most difficult Hangman word to guess?

Strategy 1 - Most Common Letter

In this article, I assume that the only valid words are those listed in a certain version of SOWPODS. This dictionary does not include proper nouns and does not include words longer than 15 letters.

A typical strategy involves

  1. determining all words that match a given template, then
  2. determining which letter appears in most of the valid words.

If multiple letters have the same rank in Step 2, then any is chosen with equal probability. The difficulty of a word under this strategy is the average number of misses made before the word is guessed. The most difficult words by length are listed in the following table:

BI
8.50
DI
8.50
GI
8.50
HI
8.50
KI
8.50
LI
8.50
MI
8.50
PI
8.50
QI
8.50
SI
8.50
JAI
17.49
KOI
17.17
COZ
16.88
EAU
16.36
FOU
16.33
VEG
16.09
JEU
16.01
JOE
15.88
HEH
15.84
ZEK
15.84
DIFF
14.50
JIFF
14.50
KOIS
14.42
VAGI
14.30
BABU
14.20
BUZZ
14.00
EAUS
14.00
FOUS
14.00
LACS
14.00
MIME
14.00
HUZZY
14.85
JIGGY
13.50
WIGGY
13.50
JUTTY
13.48
VUTTY
13.48
CUTTY
13.28
HAJES
13.18
FUZZY
13.09
DUDDY
13.04
BACHS
13.00
FOXING
13.06
JIBING
13.00
JOYING
12.96
YUCKED
12.44
YUKKED
12.11
CACHED
12.10
JAZZED
12.10
HUGGED
12.06
JUGGED
12.06
MOMZER
12.00
WUDDING
12.33
YUKKING
12.15
CACHING
12.00
JAZZING
12.00
WAXWING
12.00
COZYING
11.00
HUTTING
11.00
JUTTING
11.00
HUZZIES
10.88
HUGGERS
10.75
CUPPIEST
9.05
BUZZIEST
9.02
MUMMINGS
8.97
CUTTLING
8.75
MUFFLING
8.75
CUFFLING
8.50
FUZZLING
8.50
JAZZIEST
8.50
MUZZLING
8.50
PUZZLING
8.50
JUDDERING
7.20
FULLERING
6.71
JAZZINESS
6.66
CAMPINESS
6.38
HAMMINESS
6.21
HAPPINESS
6.21
RUBBISHLY
6.17
RUTTISHLY
6.17
YUCKINESS
6.13
UNPUZZLED
6.10
BUCKJUMPER
5.00
JOKINESSES
5.00
PICKPOCKET
5.00
BACKSTALLS
4.83
FLUFFINESS
4.75
TITTUPPING
4.58
RUBBISHING
4.57
CURFUFFLED
4.50
KURFUFFLED
4.50
PACKSTAFFS
4.50
JAZZINESSES
6.66
CAMPINESSES
6.38
HAMMINESSES
6.21
HAPPINESSES
6.21
YUCKINESSES
6.13
HUFFINESSES
6.07
FUZZINESSES
5.85
TUBBINESSES
5.61
POTTINESSES
5.45
PUFFINESSES
5.37
FLUFFINESSES
4.75
KLUTZINESSES
4.25
SULPHHYDRYLS
4.00
UNTRUTHFULLY
4.00
GRUBBINESSES
3.89
CLUBBINESSES
3.79
FRUMPINESSES
3.67
CHATTINESSES
3.50
QUAGGINESSES
3.50
SMUDGINESSES
3.50
CATTISHNESSES
4.17
BULLISHNESSES
4.00
CURRISHNESSES
3.79
RUTTISHNESSES
3.63
CULTISHNESSES
3.33
RAFFISHNESSES
3.17
RAMMISHNESSES
3.17
WAGGISHNESSES
3.17
BLACKCURRANTS
3.00
BRACHYDACTYLY
3.00
QUADRANGULARLY
3.00
UNTRANSLATABLY
3.00
WAGELESSNESSES
3.00
WOODLESSNESSES
2.88
HARMLESSNESSES
2.75
HOPELESSNESSES
2.75
CHROMATOGRAPHS
2.50
CHROMATOGRAPHY
2.50
CONGRATULATORS
2.50
CONGRATULATORY
2.50
ULTRASTRUCTURAL
3.00
ANTHROPOPHAGOUS
2.00
ASTROPHOTOGRAPH
2.00
AUTOCHTHONOUSLY
2.00
BRACHYDACTYLISM
2.00
BRACHYDACTYLOUS
2.00
CHROMATOPHOROUS
2.00
CHROMOXYLOGRAPH
2.00
CRYSTALLOGRAPHY
2.00
GNATHOSTOMATOUS
2.00

Typically, shorter words are more difficult than longer words, with the most difficult being JAI (or KOI for a common word). However, a human player may find longer words to be harder, because Step 1 is extremely difficult to calculate. I asked several players to guess two words:

  1. VISUALIZED—a 10-letter word with difficulty 0.
  2. BAKING—a 6-letter word with difficulty 11.58.

One player reached the state

_IS___I__D
NT

of which only 2 valid words remained.1 The player was unable to find either easily and ended with 12 misses. The difficulty of this word also came from the presence of a V and Z which players tend to overlook. The best score I saw was 7.

On the other hand, many players reached the state

_A_ING
???

and they were aware that many, many words fit this template.2 Guessing the word became a game of chance, which a computer cannot overcome either. The best score I saw was 8.

Strategy 2 - All Valid Letters

An alternate strategy generalizes Step 2 of Strategy 1. Instead of guessing only the letter that is most common, the computer will guess any valid letter with probability proportional to the letter’s frequency in the remaining words.

Some “easy” words became more difficult, because the computer now had a small chance to guess uncommon letters that missed. But “hard” words also became easier, because if the computer happened to guess an uncommon letter that was correct, the number of valid words would decrease dramatically. Compare the distribution of difficulties between Strategy 1 and Strategy 2 by word length.3

Strategy 1
Strategy 2

Strategy 2 plays more consistently so it is harder to outplay, but it is worse overall. These are the best words for Strategy 2:4

ZO
10.03
QI
9.79
KI
9.63
LO
9.42
PO
9.40
GO
9.39
LI
9.28
AL
9.03
DO
8.92
OF
8.85
ZUZ
14.60
QAT
13.50
VAV
13.41
ZAX
13.40
ZZZ
12.90
SAX
12.82
KAK
12.70
HOX
12.46
FAG
12.24
FAW
12.20
ZIZZ
14.23
JAZZ
13.90
JIZZ
13.49
JILL
13.33
ZILL
13.16
MIZZ
12.93
MOZZ
12.89
QATS
12.76
ZZZS
12.68
MUZZ
12.42
ZILLS
12.50
JILLS
12.36
JAZZY
12.31
JUJUS
12.19
BUZZY
12.18
VILLS
11.95
WILLS
11.81
ZIFFS
11.57
FIZZY
11.34
FAZED
10.96
JAZZED
12.22
ZIZZED
11.96
JAZZER
11.92
JAGGED
11.49
ZAGGED
11.36
FAFFED
11.13
FOXING
11.08
JOGGED
10.94
BUZZER
10.84
FUZZED
10.76
JAZZING
11.68
ZIZZING
11.27
JAZZERS
10.66
FAFFING
10.49
BUZZING
10.48
JAPPING
10.40
ZAPPING
10.30
JAZZIER
10.04
BIZZIES
9.72
JAGGING
9.61
ZIZZLING
9.43
JAZZIEST
8.91
CHIZZING
8.80
WUZZLING
8.69
QUIZZING
8.61
WHIZZING
8.59
WHIFFING
8.07
BUZZINGS
8.05
WHIPPING
7.79
WAZZOCKS
7.66
JAZZINESS
7.71
BEZZAZZES
7.09
PAZZAZZES
7.01
FUZZINESS
6.92
KOOKINESS
6.83
BOWWOWING
6.69
MUZZINESS
6.69
KAVAKAVAS
6.66
JIBBERING
6.37
QUIZZINGS
6.31
JOKINESSES
6.78
FOXINESSES
6.43
BOXINESSES
6.37
FOZINESSES
6.29
WAVINESSES
6.18
HOKINESSES
6.13
COXINESSES
5.84
HAZINESSES
5.66
PIPINESSES
5.43
WIGWAGGING
5.07
JAZZINESSES
7.58
FUZZINESSES
6.88
KOOKINESSES
6.73
MUZZINESSES
6.61
HUFFINESSES
6.08
WOOZINESSES
5.98
PUFFINESSES
5.91
BOOZINESSES
5.91
BUGGINESSES
5.88
BATTINESSES
5.32
FLUFFINESSES
5.17
STUBBINESSES
4.94
QUIZZINESSES
4.93
STUFFINESSES
4.80
SHABBINESSES
4.77
SNAZZINESSES
4.56
SCABBINESSES
4.41
CHUMMINESSES
4.33
JAGGEDNESSES
4.29
FLAGGINESSES
4.09
FADDISHNESSES
4.50
WAGGISHNESSES
4.33
CADDISHNESSES
4.25
FOPPISHNESSES
4.25
RAFFISHNESSES
4.20
DOGGISHNESSES
4.17
JOBLESSNESSES
4.15
JOYLESSNESSES
4.04
BOOKISHNESSES
4.03
HUFFISHNESSES
3.95
WAGELESSNESSES
4.03
WOODLESSNESSES
3.71
FACELESSNESSES
3.58
BOOTLESSNESSES
3.54
FOODLESSNESSES
3.52
HOPELESSNESSES
3.50
WORKLESSNESSES
3.49
FECKLESSNESSES
3.39
WORDLESSNESSES
3.20
SHEEPISHNESSES
2.94
SENSELESSNESSES
2.95
SHAMELESSNESSES
2.88
SMOKELESSNESSES
2.84
SHAPELESSNESSES
2.81
STATELESSNESSES
2.64
PULSELESSNESSES
2.56
INEFFABLENESSES
2.52
POSSESSEDNESSES
2.50
INDIVIDUALIZING
2.48
ENJOYABLENESSES
2.41

Strategy 2 does the worst when the word contains very uncommon letters such as J and Z, and like Strategy 1, it finds repeated letters difficult.

Strategy 3 - Most Knowledge

For another strategy, the focus is on the number of valid words remaining after every guess. To judge the value of guessing the letter A, first all currently valid words are assumed to be in play with equal probability. The guess of A is applied to each valid word, and all the valid words after each application are counted up. The guess that results in the fewest number of valid words is chosen. Again, if multiple letters tie, then either is chosen with equal probability.

This strategy was too computationally expensive for me to calculate anything meaningful. It seems to perform the worst too, since it disregards whether a guess is likely to be right or wrong.5

Data

All the data and code used can be found on the GitHub repo.


  1. The other word is MISCLAIMED.

  2. 102 words match _A_ING, and the frequencies of remaining letters are somewhat evenly distributed.
    B-6, C-11, D-11, E-7, F-10, H-11, J-3, K-10, L-13, M-12, O-1, P-10, Q-0, R-23, S-12, T-15, U-0, V-8, W-19, X-5, Y-10, Z-6.

  3. While the difficulty values for Strategy 1 were computed exactly, the values for Strategy 2 were approximated. Computing the probabilities of every possible path became too time-consuming, so instead the computer played every word 100 times and averaged the results. In some instances, the difficulty could fluctuate by a few points.

  4. After computing difficulties using 100 trials, the top 50 words in each category had their difficulties recalculated using 1000 trials for more accurate rankings. As a result, these values may differ from the data file where only the results using 100 trials were recorded.

  5. Using a dictionary of ABC, BAC, DDD, DDE, Strategy 3 is on average 3.94× worse than Strategy 1 and 3.06× worse than Strategy 2.