Lexitron

A command-line regex search engine for the English language.

Requirements

The only requirement is Python 3.9 or later.

If you try to install Lexitron and something goes wrong, let me know what your system details are and I'll try to get it fixed.

Installation

Lexitron is available on the Python Package Index (pip). To install, simply type

$ pip install lexitron

at the command line.

Once the install is complete, you can access Lexitron with the lx command at the terminal, and its rhyming dictionary with ry (see Rhymes).

Quick tour

• lx '.*icide': words ending in "icide"
• lx -s rdb: words containing "rdb" anywhere
• lx 'b.a.k': five-letter words matching B?A?K
• lx -a listen: anagrams of "listen"
• lx -f aeilnrst: words you can spell from these letters
• lx -f abcdef.: the same, with a blank tile
• lx -l 4- -rf eihnglw -s h: today's Spelling Bee (center letter "h")
• lx -i q -e u: words with a "q" but no "u"
• lx -t a.a.a: words shaped like "rarer"
• lx -w crane:..y.g: Wordle candidates
• lx --hooks cat: Scrabble hooks for "cat"
• ry orange: rhymes (well, near rhymes) for "orange"
• lx -n '.*tion': how many words end in "tion"

More in Examples.

Usage

Usage syntax is

$ lx [options] [expression]
$ lx -f|-a [options] letters [expression]

where expression is a regular expression and [options] are as follows. The expression can be left out when another option says what to look for (lx -t abba, lx -w crane:..y.g).

The expression must match the whole word, so lx cat finds only "cat", and lx ".*cat" finds words ending in "cat". To match the expression anywhere within a word, use -s (so lx -s cat finds "cat", "catalog", "bobcat", etc).

Inflected forms (plurals, conjugations, etc, like "went" or "oxen") are searched along with base forms. Use -b to search base forms only. Phrases with spaces are listed in their base form only ("sperm whale", not "sperm whales").

Searches are case-insensitive unless you use -c. Options can go before or after the expression. Quote the expression ('...') so your shell doesn't interpret characters like *, ? and $.

zsh tip: add alias lx='noglob lx' to your ~/.zshrc and you can skip the quotes for * and ?: lx .*icide, lx colou?r.

A little regex

Expressions, letters and patterns

The same characters mean different things in different places:

• In the expression, they're regex: lx 'dot?' finds "do" and "dot" (? makes the "t" optional).
• In letters (-f, -a), each letter is a tile and . is a blank: lx -a dot. finds "dote", "doth", "toad" and so on.
• In a pattern (-t), each letter is a placeholder: lx -t abba finds "deed", "noon" and so on.

How the expression matches

Spelling from letters

With -f and -a, the letters come first, optionally followed by an expression: lx -a aerst finds all anagrams of "aerst", lx -a aerst 's.*' only the ones starting with "s". Lengths and letters count letters only, ignoring apostrophes, hyphens and spaces.

In the letters, . is a blank tile that can stand for any one letter: lx -f abcdef.. With -a, blanks must be used, like the rest of the letters.

More filters

In a pattern, each letter stands for some letter: the same letter for the same letter, different letters for different ones. . is any letter, and as in regex, .* is any run of letters (.+ one or more, .? zero or one). So abba matches "deed" but not "dddd", a.a.a matches "rarer", and ab.*ab matches words that start and end with the same two letters, like "enlighten". The pattern only describes shape; to fix actual letters, add an expression: lx -t a.a.a 'r.*'.

Word games

-w sets the length from the guess, and uses level 50 unless you give -L, since Wordle answers are everyday words. A repeated letter is handled the way Wordle does it: in geese:.y..., the grey second and third "e" mean the word has exactly one "e".

With -w, the best guesses come first (-o best), each with its score: how evenly its colors would split the remaining candidates, as expected information in bits. A higher score means fewer candidates left, on average. With more than 500 candidates, the scores are estimated from an evenly spaced sample of them. -o best sorts up to 5000 words.

Sometimes a word that can't be the answer would narrow things down more, for example by trying five new letters rather than repeating a green one. In a terminal, up to 10 of these come first when they beat every candidate, dimmed (or marked * when NO_COLOR is set). On a tie the candidate wins, since it might be right. Piped output lists only the candidates. Use --probes to see only these words: the 20 best and, after a ... row, the 20 worst.

--plus, --minus and --change take an optional count, cat:2 for two letters. These and --hooks skip words with apostrophes. In a terminal, --hooks shows the front hooks, the word and the back hooks; with -x or when piped, it lists the hooked words.

Which words to search

Output

Type $ lx -h for full help text.

If you aren't familiar with regular expressions, it isn't too hard to learn the basics. There are many resources online. A good starting point is the Wikipedia article.

Output

By default, Lexitron will output a well-formatted (potentially multi-column) list of words. Use -I to add a header describing the results.

Words are sorted alphabetically, except that inflected forms (plurals, conjugations and so on) come right after their base form when it's also in the results: "run ran running runs runabout". In a terminal, the inflected forms are dimmed, so each family stands out, and the other words are colored by how common they are: bold green for core vocabulary, then green, yellow and orange, to red for the most obscure (see Word list). -o alpha sorts strictly alphabetically, with the same styling. Set NO_COLOR in your environment to turn styling off.

The results are separated into "proper" words (containing capitals, like "France" or "USA") and "common" words (lowercase, like "boat").

Using the -x flag will return a more machine-readable output with one word per line. This is also what you get automatically when the output isn't a terminal (e.g. when piping into another command or redirecting to a file). In that case the -I header goes to stderr, so it doesn't get mixed in with the words.

Word list

Lexitron's dictionary comes from the English Speller Database (SCOWL), which grades every word by how well known it is:

All spellings are included: American, British, Canadian and Australian ("color" and "colour", "organize" and "organise"). Accents are removed ("café" becomes "cafe"). Possessives, abbreviations and Roman numerals are left out.

Rhymes

Lexitron comes with a rhyming dictionary too, the ry command:

$ ry -L 50 time
1 syllable
chime    crime    grime    mime     rhyme    thyme    rime
climb    dime     lime     prime    slime    clime

2 syllables
enzyme        daytime       primetime     flextime      smalltime
sublime       lifetime      ragtime       halftime      mistime
centime       lunchtime     sometime      longtime      noontime
quicklime     meantime      springtime    mealtime      teatime
subprime      nighttime     wartime       onetime
bedtime       pastime       downtime      peacetime

3 syllables
pantomime     flexitime     overtime      anytime       summertime
paradigm      maritime      wintertime    realtime

Usage syntax is

$ ry [options] word [expression]

A rhyme matches the sounds from the last stressed vowel on, so "cat" rhymes with "acrobat" and "happy" with "unhappy". Rhymes are grouped by syllables, most common first, and colored by how common they are, as in lx. Weak rhymes, which just put another word in front ("daytime" for "time"), come last and dimmed. Words that are pronounced exactly the same ("right" for "write") are left out.

Long rhymes come first: words that also match the vowels from the stress before, so "understand" gets "wonderland" and "education" gets "dedication" and "celebration". Those with the most alike consonants come first. They come first for near rhymes and assonance too: ry -A weblog starts with "headlong" and "headstrong". -g shows only the long rhymes.

$ ry -L 50 -y 3 understand
long rhymes
underhand     wonderland    motherland

3 syllables
ampersand      hinterland     firebrand      countermand
beforehand     overhand       secondhand     fairyland
contraband     overland       taxistand      tableland
fatherland     reprimand      borderland     timberland

A word with more than one pronunciation gets rhymes for each:

$ ry -I -y 1 -L 35 live
8 rhymes for 'live' (L AY1 V or L IH1 V)

live (L AY1 V)

1 syllable
dive      drive     five      hive      strive    thrive

live (L IH1 V)

1 syllable
give     sieve

The rest work as in lx: an expression to narrow the results (ry cat 'b.*'), and -s, -l, -i, -e, -t, -b, -p, -P, -u, -U, -n, -I and -x.

Near rhymes have the same number of syllables after the stress and the same or similar vowels, with consonants that sound alike: voiced or not ("love", "rough"), similar ("cat", "cap"), or one more or less ("cat", "cast"). A word with plenty of close near rhymes gets only those; one with few gets more distant ones too. Plurals and other forms of perfect rhymes ("cats" for "cat") come last, dimmed. A word with fewer than 10 perfect rhymes gets near rhymes after them, unless you've narrowed the results down (with an expression, -y, -l, -i, -e or -t) or are piping them. A word with no perfect rhymes gets only near rhymes:

$ ry -L 50 orange
(no perfect rhymes for 'orange'; showing near rhymes)
2 syllables
forage      florin      foraged     pouring     shorten     warrant
foreign     boring      forum       quorum      snoring     warrens
porridge    coarsen     goring      roaring     soaring     warring
sporran     cordon      lozenge     scoring     storing     florins
storage     coring      oaring      shoring     torrent     portage
warren      flooring    poring      shortage    warden      sporing

3 syllables
abhorrence     decorum        ignoring       foreshorten
abhorrent      deploring      imploring      importune
abhorring      encoring       restoring      offshoring
adoring        exploring      outpouring     reportage

4 syllables
underscoring    lachrymal

You can rhyme a phrase, where little words like "you", "the" and "it" go unstressed, as they would in speech. A phrase that ends in them also gets words that rhyme with the rest, followed by the same little words. Those come last, and when they follow "you", "me", "it" and the like, only verbs are used. T or D followed by "you" runs together, as in "virtue":

$ ry -L 40 'hurt you'
2 syllables
virtue

near rhymes
curfew      bureaux     burrow      nurture     virtues
borough     burro       furrow      thorough

… you
alert you         convert you       flirt you         skirt you
assert you        desert you        insert you        spurt you
avert you         disconcert you    invert you        squirt you
blurt you         divert you        pervert you       subvert you
concert you       exert you         revert you

With -N you get near rhymes for both: "curfew" and "purview", then "curse you", "nurse you", "work you" and so on.

Wrenched rhymes (-w) end with the same sounds, stressed in one word and not the other: "be" and "philosophy", "window" and "go". Songs are full of them, since singing can put the stress where it likes. ry -w be -y 4 gives "philosophy", "democracy", "biology" and so on, and ry -w window gives "go", "though", "below", and "shadow", "meadow", "willow". With -N or -A the sounds can differ as they do for near rhymes and assonance: ry -w -A -y 2 be gives "always", "carried", "babies", "coffees".

You can also rhyme a pronunciation in ARPAbet (ry "K AE1 T"). A word ry doesn't know is pronounced like the known word whose spelling ends the same way most closely ("blorange" like "orange"), and it says so.

Pronunciations come from the CMU Pronouncing Dictionary, which is American English. Of the words up to level 60, it has 56%. Another 30% are worked out from words it has: plurals, past tenses, British spellings and compounds. The rest (14%) are guessed by a model trained on it, so a few of those are wrong.

Examples

Word games

NYT Spelling Bee. Words of 4 or more letters, using only the seven letters (each as often as you like), that contain the center letter "h" (-s h). Spelling Bee doesn't accept contractions like "he'll".

$ lx -I -l 4- -rf eihnglw -s h
38 matches for /h/ from 'eihnglw' length 4-

ghee         high         wheelie
ghillie      hill         wheen
he'll        hinge        when
heel         hinging      whew
heeling      neigh        while
heinie       neighing     whiling
hell         nigh         whin
henge        nihil        whine
hewing       weigh        whining
hewn         weighing     whinge
hieing       whee         whingeing
higgle       wheel        whinging
higgling     wheeling

Spelling Bee pangrams. -ra finds words that use every one of the letters, each as often as you like.

$ lx -ra eihnglw
wheeling

Wordle. One -w per guess: g for green, y for yellow, . for grey. The best next guesses come first, with how much each would tell you, in bits. The starred words can't be the answer, but would narrow things down more than any word that could.

$ lx -w crane:.g...
pouts 4.88*    trims 4.19     drugs 3.74     frisk 3.36     brood 3.06
plots 4.82*    profs 4.14     troop 3.73     trout 3.33     groom 3.04
typos 4.78*    prism 4.14     proud 3.72     prior 3.30     brusk 3.00
posit 4.76*    frost 4.13     drift 3.69     proof 3.29     wroth 2.90
moist 4.70*    proms 4.10     group 3.67     brisk 3.28     privy 2.87
omits 4.70*    dript 4.09     trust 3.64     promo 3.21     troth 2.86
plods 4.68*    prows 4.08     trump 3.64     growl 3.19     drill 2.86
pilot 4.67*    gript 4.08     drums 3.62     drily 3.18     broom 2.86
tipsy 4.62*    writs 3.97     fruit 3.61     drool 3.17     grill 2.84
foist 4.60*    prosy 3.96     orbit 3.56     frump 3.17     droll 2.83
trips 4.48     grids 3.95     grubs 3.54     brush 3.17     proxy 2.82
trios 4.41     wrist 3.95     brows 3.53     troll 3.15     druid 2.79
prods 4.33     tromp 3.87     drubs 3.53     grimy 3.15     frill 2.67
prigs 4.30     brigs 3.86     droop 3.51     froth 3.14     truth 2.60
drops 4.30     frogs 3.82     truss 3.51     truly 3.13     brook 2.43
drips 4.27     trots 3.79     dross 3.48     trill 3.12     gruff 2.43
grits 4.25     props 3.75     prowl 3.44     broil 3.11     dryly 2.17
grist 4.22     grows 3.74     gross 3.43     broth 3.07     frizz 1.91
grips 4.20     brims 3.74     tryst 3.40     primp 3.07     wryly 1.78

After a second guess, the best you can do is a guess that might be right:

$ lx -w crane:..y.g -w toils:.y...
abode 2.00
above 2.00
adobe 2.00
awoke 0.81

Wordle, by hand. The same kind of search as a regex: "r" is second and "n" is fourth, there's an "e" somewhere, and "o", "s" and "t" are out. (?=.*e) is a lookahead: "contains an e".

$ lx '(?=.*e)[^ost]r[^ost]n[^ost]'
arena
brine
crane
crine
frena
prune
urine

Scrabble hooks. Letters that make a word when put in front of "at", and after it.

$ lx --hooks at
bcefghklmoprstvw  at  e

Q without U.

$ lx -i q -e u
faqir        qindarka
faqirs       qindarkas
niqab        qintar
qadi         qintars
qadis        qoph
qi           qophs
qibla        qwerty
qiblas       sheqalim
qiblah       sheqel

Word ladders. Every word one letter away from "love".

$ lx --change love
cove
dove
hove
lave
live
lobe
lode
loge
lone
lope
lore
lose
move
rove
wove

Adding a letter.

$ lx --plus cat
cant
cart
cast
cate
cats
chat
coat
scat

Crossword fill. Five letters, B?A?K, most common first.

$ lx -o common 'b.a.k'
black
blank

Scrabble bingos. Every anagram of a rack.

$ lx -a retains
antsier
nastier
retains
retinas
retsina
stainer
stearin

Words from a rack. Five-letter words from "aeilnrst", most common first.

$ lx -o common -l 5 -f aeilnrst | head
aisle
alert
alien
aline
alter
aren't
arise
earls
earns
inert

Blank tiles. . stands for any one letter. Six-letter words or longer from "abcdef" plus a blank:

$ lx -l 6- -f abcdef.
abduce
bached
backed
barfed
braced
cabbed
cabled
chafed
decaff
decafs
deface
fabled
facade

and the anagrams that use every tile:

$ lx -a zebr.
braze
zebra

Exploring English

Words ending in "icide".

$ lx ".*icide"
aborticide      herbicide       patricide
acaricide       homicide        pesticide
algicide        infanticide     regicide
bactericide     insecticide     rodenticide
deicide         larvicide       sororicide
feticide        liberticide     spermicide
filicide        matricide       suicide
foeticide       menticide       tyrannicide
fratricide      miticide        uxoricide
fungicide       parasiticide    vaticide
germicide       parricide       vermicide

Words containing "rdb", with the info header.

$ lx -Is rdb
18 matches for /rdb/

birdbath       hardbitten
birdbaths      hardboard
birdbrain      hardbound
birdbrains     swordbill
birdbrained    swordbills
cardboard      wordbook
hardback       wordbooks
hardbacks      yardbird
hardball       yardbirds

Words shaped like "rarer": the same letter first, third and last.

$ lx -t a.a.a
abaca
abaya
afara
agama
asana
duded
ninon
ovolo
popup
rarer
sises

Seven-letter palindromes with four different letters, mirrored.

$ lx -t abcdcba
deified
racecar
reviver
rotator

No vowels, six letters or more.

$ lx -e aeiou -l 6-
crwths     rhythms
crypts     spryly
flybys     sylphs
ghylls     synchs
glyphs     synths
nymphs     syzygy
psychs     trysts
rhythm     tsktsk

All five vowels, in order.

$ lx '[^aeiou]*a[^aeiou]*e[^aeiou]*i[^aeiou]*o[^aeiou]*u[^aeiou]*'
abstemious
abstemiously
arsenious
facetious
facetiously

The longest words you can type on the top row of a keyboard.

$ lx -o length '[qwertyuiop]+' | tail -3
proprietor
repertoire
typewriter

How many words end in "tion"?

$ lx -n ".*tion"
2299

(Take this with a grain of salt: it depends on what you count as a word, and which technical or niche jargons are included. Try -L 95 to include obscure words.)

Words ending in "woman" versus words ending in "man" but not "woman". (?<!wo) is a lookbehind: "not preceded by wo".

$ lx -n ".*woman"
72

$ lx -n '.*(?<!wo)man'
323

The same double letter twice, except vowels and "s" (to skip "-lessness"), including proper words. \1 repeats whatever (...) matched, and -b leaves out plurals like "hillbillies".

$ lx -Ibus "([^aeious])\1.*\1\1"
50 matches for /([^aeious])\1.*\1\1/ (8 proper, 42 common)

Allhallowmas
Allhallows
Allhallowtide
Boccaccio
Chancellorsville
Connellsville
Llewellyn
Ulladulla

acciaccatura        hullaballoo         pralltriller
allegorically       idyllically         razzamatazz
allergically        illegally           razzmatazz
ballistically       illiberally         riffraff
bellpull            illogically         rollerball
chiffchaff          intellectually      scuttlebutt
collaterally        jellyroll           skillfully
colloquially        kinnikinnick        syllabically
dillydally          metallically        syllogistically
dullsville          millennially        unskillfully
elliptically        monosyllabically    villanelle
flibbertigibbet     pellmell            volleyball
granddaddy          phallically         whippersnapper
hillbilly           pizzazz             willfully

Acronyms of five or more letters: proper words only (-U), in capitals (-c).

$ lx -Uc '[A-Z]{5,}' | head
ALGOL
ANZUS
ASCII
ASEAN
ASPCA
AWACS
BAFTA
BASIC
BOGOF
CAPTCHA

Phrases and compounds ending in "cake".

$ lx -p ".*cake" | head
angel cake
angel food cake
beefcake
buckwheat cake
cake
cheesecake
coffee cake
coffeecake
cotton cake
cupcake

Only phrases. -P leaves out single words.

$ lx -P 'sperm.*'
sperm bank
sperm cell
sperm oil
sperm whale
spermatic cord

Using Lexitron from Python

>>> from lexitron import search
>>> search('.*icide', sort='common')['common'][:3]
['fungicide', 'germicide', 'homicide']

search() takes the same options as the command line; see its docstring.

Changes in version 4.1

• Patterns (-t) understand .*, .+ and .? as in regex: ab.*ab matches "enlighten" as well as six-letter words.
• -w with many candidates (after a poor opener) scores guesses against a sample, so it stays quick; with -u, proper and common candidates are scored together.
• -o best stops with an error on lists over 5000 words instead of running for minutes; -n no longer sorts, so counts are faster.
• Errors for Wordle guesses of different lengths and for --probes with -o, which used to be ignored silently.

Changes in version 4.0

• New filters: -i (must contain these letters), -e (must not contain them) and -t (pattern: abba matches "deed", a.a.a matches "rarer").
• New word-game options: -w for Wordle (sorted best guess first), --plus, --minus and --change for adding, removing and changing letters, and --hooks for Scrabble hooks.
• New sort order -o best: best Wordle guess first. In a terminal, -w shows the bits of information each guess would give, with better guesses that can't be the answer first; --probes lists only those.
• -i now means --include. The info header is -I.
• Blank tiles in -f and -a letters are . only, matching . in expressions and patterns; ? is no longer accepted.
• The expression can be left out when another option says what to search for.
• Shorter help, with options in a more sensible order, and a note on what . means in expressions, letters and patterns.
• The old option names --only-phrases and --only-uppercase are gone.

Changes in version 3.3.1

• Long option names are consistent: --only-phrases and --only-uppercase are now --phrases-only and --uppercase-only, like --base-only and --number-only. The old names still work.

Changes in version 3.3

• Phrases with spaces are listed in their base form only: "sperm whale" but not "sperm whales". SCOWL lists inflected forms for only a few phrases, so they're left out for consistency. Hyphenated words keep theirs ("x-rayed").

Changes in version 3.2

• New option -P searches only phrases and hyphenated entries, like "sperm whale"; -p still adds them to the usual results.

Changes in version 3.1

• New default sort, family: alphabetical, with inflected forms right after their base form ("run ran running runs"). -o alpha gives the old order.
• Terminal output is styled: inflected forms listed under their base form are dimmed, and other words are colored by how common they are, from bold green to red. Set NO_COLOR to turn this off; piped output is never styled.

Changes in version 3.0

• New word list: SCOWL replaces AGID, adding everyday words AGID was missing ("a", "of", "from"), British and other spellings, and obscurity levels (-L). The default level leaves out very obscure words; use -L 95 to search everything.
• Expressions now match the whole word by default. In 2.x, lx cat matched any word containing "cat"; to get that behavior now, use lx -s cat.
• Inflected forms (plurals, conjugations, etc) are now searched too; use -b to search base forms only, as in 2.x.
• New options: letters (-f, -a, -r, with blank tiles), length (-l), sorting (-o), phrases (-p) and case-sensitive search (-c).
• Options can go anywhere on the command line.
• Output no longer crashes when piped or redirected; it falls back to one word per line.
• -i with -n prints the header line; -u and -U can no longer be combined.
• Invalid regular expressions give a short error message instead of a traceback.
• Quotation marks in the expression are no longer stripped, so you can search for words containing apostrophes.
• Lexitron can be used as a Python library.
• Requires Python 3.9 or later.

Acknowledgements

For its dictionary, Lexitron uses the English Speller Database (ESDB, previously known as SCOWL) by Kevin Atkinson. See wordlist.aspell.net and lexitron/SCOWL-Copyright.txt for its copyright and license.

Versions before 3.0 used the Automatically Generated Inflection Database (AGID), also by Kevin Atkinson.

License

Lexitron is licensed under GNU GPL Version 2.

Contact

Questions, bug reports, and feature requests can be filed on the Github issues tracker.