This challenge is to find the longest chain of English words where the first 3 characters of the next word match the last 3 characters of the last word. You will use an common dictionary available in Linux distributions which can be downloaded here:


which has 99171 English words. If your local Linux /usr/share/dict/words is the same file (has md5sum == cbbcded3dc3b61ad50c4e36f79c37084), you can use that.

Words may only be used once in an answer.

EDIT: Letters must match exactly including upper/lower case, apostrophes, and accents.

An example of a valid answer is: idea deadpan panoramic micra craftsman mantra traffic fiche which would score 8.

The answer with the longest valid chain of words will be the winner. In the event of a tie, the earliest answer will win. Your answer should list the chain of words you have found and (of course) the program you wrote to do it.

  • \$\begingroup\$ What are the ways allowed to load these datas ? \$\endgroup\$
    – Hacketo
    Jan 25 '15 at 13:26
  • \$\begingroup\$ You should download the word list from the dropbox link above, then process it with your program to generate the best answer you can. \$\endgroup\$ Jan 25 '15 at 13:34
  • \$\begingroup\$ Do accents matter? Also, what about words that are shorter than three letters? \$\endgroup\$
    – KSFT
    Jan 25 '15 at 13:46
  • \$\begingroup\$ Words shorter than 3 letters can't meet the 3 letter match rule, so are excluded. Letters must match including accents and upper/lower case. \$\endgroup\$ Jan 25 '15 at 14:01
  • 1
    \$\begingroup\$ My analysis of the graph is that there's only one non-trivial strongly connected component, and the longest path in the compressed graph is of length 6. By considering the minimum for each three-letter sequence of words in the SCC with the sequence as a prefix and a suffix, I get an upper bound of 4655 on the longest chain of words, so there might still be a lot of room to improve on the current best of 1733. \$\endgroup\$ Jan 29 '15 at 11:00

Java, heuristic favouring the vertex which induces the largest graph: 1825 1855 1873

The code below runs in under 10 minutes and finds the following path:

[wad, wadis, dis, dismay, may, mayfly, flywheels, elsewhere, erecting, ingratiate, ateliers, ersatzes, zest, esthetic, tickled, ledger, germicide, idealizes, zestful, fulling, ingrains, institute, uterine, ineptness, essaying, ingeniously, slyness, essences, cessations, onshore, ores, resoundingly, glycerine, inertness, essay, say, saying, ingenuous, ousted, tediously, sly, slyest, estrogen, genuflecting, ingestion, ionizer, zeros, roses, sesames, mes, meshed, hedonist, isthmuses, sesame, amending, ingredient, entrapment, enthuses, session, ionosphere, erectness, essayist, isthmus, mustaches, hesitatingly, glycogen, generation, ions, onset, settable, blew, lewder, deriding, ingratiates, testicles, lessen, sensitization, ionization, ionizing, ingratiating, ingenious, ouster, terrorizing, ingest, estranges, gesticulating, ingrates, testis, tissue, sue, suede, edelweiss, issuing, ingraining, ingrown, owner, nerdiest, estimation, ionospheres, rescue, cue, cueing, ingesting, ingot, got, gotten, tensor, sorrowing, ingratiated, tedious, ousting, ingratiatingly, glycerin, ringside, identifiable, bleariest, ester, terminological, calibrator, torrent, entraps, apse, pseudonym, nymphomania, niacin, cinema, emailed, led, ledges, gesticulates, testicle, clement, entail, ail, ailment, enter, terrains, inspires, restaurateur, euros, rosiest, estimates, tester, termite, iterator, torture, urethras, raspiest, estimator, tore, oregano, anointment, enthuse, useful, fulfil, filmstrip, riposte, stereotyped, pedicure, urea, readmits, itself, elf, elfin, finagles, lesbians, answerable, bleat, eatery, erythrocytes, testosterone, one, ones, nest, esteemed, medicine, inextricable, blessed, sediment, entry, try, tryout, outsources, cesarians, answered, redressed, seducer, cervical, calumniates, test, establishment, entombment, enthusiastic, tickles, lessens, ensemble, blemishes, hesitant, antic, tick, ickiest, estimable, blemished, hedgehog, hogan, gantlet, letdown, own, ownership, hippest, estates, testates, testiest, establishes, hes, hesitates, testable, bleakest, esthetes, testament, entice, iceberg, erg, ergonomic, microscope, operatives, vestibules, lesser, serenade, adenoidal, dales, lest, estrangement, entrap, raptures, resourceful, fulsome, omen, menswear, earthliest, established, hedges, gestates, testy, styes, yeshivot, voter, terrible, blender, derides, descent, enticed, cedillas, lass, assailable, bleacher, hermit, mite, item, temperas, rash, ashtray, rayon, yonder, dermis, mismanage, agendas, dash, ashy, shy, shyster, terrapins, insatiable, bleeder, derives, vestment, entangle, glen, lengthens, ensconced, ceded, deduced, cedars, arsenic, nice, ice, iced, cedar, daredevil, villa, llamas, masseuse, use, useable, bleach, achievable, bleached, hedonistic, tic, ticker, kerchieves, vessel, sell, ell, elliptic, ticket, kettles, lessee, seeps, epsilon, longboat, oath, atherosclerosis, sisterhood, oodles, lesson, sonatas, tassel, selvage, age, agent, entranced, cedes, descender, deranges, gestures, restraint, interment, enthused, seduced, cedilla, llama, amalgam, gamut, mutable, blend, endear, earthy, thymus, mussel, seltzer, zero, erodes, despot, potful, fulfillment, enthrall, allot, lotus, tussle, sledgehammered, redolent, entrapped, pedestal, talk, alkalis, listen, tended, deductible, bleeped, pedigree, reentered, redistribute, uterus, rustproofed, fed, fedora, oranges, gesundheit, either, herdsman, manes, nestles, lessor, sorrowful, fullback, acknowledges, gestured, redoubtable, blended, deduces, cesareans, answer, werewolves, vesper, perseveres, restructures, reside, ideogram, rammed, meddlesome, omens, ensign, ignores, restrains, insolent, entanglement, entrenchment, enticement, entomological, calligraphy, physical, calico, iconoclast, astringent, entertainment, entrant, antennas, nasty, stymie, miens, enslave, averred, redefine, inexorable, blenched, hedgerow, rowboat, oat, oaten, tend, endears, arson, songwriter, terminable, blent, entreaty, atypical, calypso, psoriasis, sister, term, ermine, ineligible, bleaker, kerosene, enema, emancipator, tormentor, torrider, derailment, entertains, instil, tildes, destine, inelegant, anthropomorphic, hiccup, cupolas, lastingly, glycerol, rollback, acknowledgment, entombed, bedridden, denser, servicewomen, menopause, used, sedatives, vesicle, clearinghouse, user, servant, antipodes, descry, crystalline, inexpedient, enthusiast, astonishment, entirety, etymological, calendared, redbreast, astronomer, merinos, nosedove, overpay, pay, paymaster, termagant, antiaircraft, aftercare, ares, resentful, fulcrum, rumpus, pushcart, artiste, stethoscopes, pesetas, taste, steadfast, astride, ides, destitute, utensil, silvan, vanguard, ardent, entryway, waysides, despair, airdrop, ropes, pestered, redder, derangement, entered, redeemed, medullas, lasagnas, nasal, salsas, sashay, hay, haymow, mow, mowed, wedder, derringer, germane, anemic, microfilmed, media, diatom, tomboys, oyster, terminator, toreador, dorsal, salespeople, pleased, sedater, terabit, bitten, tentacle, clergyman, manifesto, stomach, achoo, hoopla, plaza, azalea, leaven, vendor, dormant, antiparticle, cleared, redraft, afterword, ordains, insufficient, entitlement, entomb, ombudsmen, men, mental, tallyhos, hospice, icecap, cape, aperitif, tiffed, fedoras, rasped, pediatric, rickshaw, hawker, keratin, tinctures, reset, setback, acknowledgement, enthronement, entwine, inexact, actor, torpedos, dosed, sedan, dancer, cerebrum, rumple, plea, leach, ache, cheaper, per, periscopes, pestilent, entreat, eater, terser, serape, ape, apes, pesky, skycap, capped, pederast, astuter, terrace, acetaminophen, henchmen, menopausal, saltcellar, lard, ardor, dormice, icebound, underbrush, ushered, redrew, rewound, underclass, assassin, sinew, newscast, astrologer, gerund, undertaken, ken, kens, ensnared, redcap, cappuccinos, nostrum, rum, rumored, redraw, rawhide, identical, calcine, inertia, tiara, arabesque, queerer, reruns, unsold, oldie, diesel, selectmen, mentored, redden, dental, talon, longhand, and, androgen, genome, omelet, lethal, hallucinogenic, nickname, amen, menhaden, denudes, despaired, redevelop, lope, operas, rasp, aspired, redskin, kindergartens, ensnares, resultant, anthropological, callus, lustful, fulcra, crammed, mediocre, crepes, pesticide, ideas, eastbound, under, derrières, respired, rediscovered, redundant, antihero, erode, ode, odes, described, bedevil, villager, gerrymander, deride, ideograph, aphid, hid, hides, describes, besides, despoil, oilskin, kingdom, dominant, ant, antipasti, stiffens, ensured, redeemer, merchant, antiwar, warped, pederasty, stylus, lush, usher, her, hereafter, terrapin, pinnacle, clerical, caliber, bereave, avenger, geriatric, rickshas, haste, stereoscopes, pester, termini, initiator, tortures, restorer, reran, ransomed, medulla, llanos, nostril, rill, illogical, calif, lifer, fervor, vortex, textures, resister, termed, medieval, valor, lord, ordered, rediscover, verbatim, times, mesdames, mescal, caliper, periscope, opera, erasures, restart, artichokes, kestrel, reliant, antebellum, lumbago, agog, goggle, gleeful, fulfill, illustrator, tor, torque, questionnaires, resumed, mediator, tort, orthodoxy, oxymora, oratorio, riot, iotas, taster, terrific, fiche, checkpoint, interloper, perfumes, mesas, sassafras, rasher, heraldry, drywall, all, allergens, ensnare, area, rearm, armchair, airman, manufactures, resurface, acerbic, bicycle, cleverer, rerun, runt, untidy, idyllic, lichens, ensures, resend, endemic, microchip, hippopotamus, muscatel, telecast, astronaut, autopilot, lot, loth, other, heros, rosin, single, gleamed, mediaeval, valet, lettered, redound, underside, ideological, calliper, perihelia, liaison, sonic, nicknames, messenger, germicides, descendant, antigen, genital, tall, allergen, gentleman, mangos, gossipped, pedicures, resistant, antlered, redeveloped, pedagogical, calligrapher, heroins, inside, idea, deafen, fen, fencer, cerebra, bravuras, rascal, calculus, lusher, herbivores, resins, instill, illicit, citric, ricochet, heterodoxy, oxygen, generic, rice, icebox, box, boxcar, cartography, physique, quell, ellipsis, sis, sisal, sallow, lowbrow, rowel, well, elliptical, calf, alfresco, scow, cow, cowboy, boy, boyfriend, end, endeared, red, redesign, ignoramus, musket, kettledrum, rump, umped, pedlar, larvas, vassal, salmonellas, last, astronomical, calfskin, kingfisher, hereupon, ponchos, hospital, talisman, mantel, telethon, honcho, chomped, pedant, antitoxins, instant, antipastos, tossup, superintend, endangered, redskins, instigator, torpor, portico, icon, conquistador, dormer, merganser, seraphic, hiccuped, pedagogue, guerrillas, laser, sera, eraser, seraph, aphasic, sickbed, bed, bedsores, resign, ignorant, anthropocentric, richer, herdsmen, menu, enures, resuscitator, tornado, ado, adobe, obeisant, anthill, illegal, gallon, longshoremen, menace, ace, acetylene, enemas, mas, mascot, cot, cotton, tonsures, restores, result, ultraviolet, letterbox, boxer, xerography, physiological, calmer, merchantmen, mentor, torus, russet, settee, teenager, gerbil, billfold, old, olden, denatures, resubmit, mitten, ten, tenon, nonchalant, antique, queasy, asymmetric, ricksha, shanghai, haircut, cutups, upsides, descriptor, torpid, pidgin, gins, instep, tepee, peeper, perturb, urbane, anemia, miasmas, mascaras, raspy, spy, spyglass, assures, resonator, tortilla, llano, anon, nontechnical, calabash, ashram, rampart, arthropod, podia, diagram, ramp, amp, amphitheatres, resistor, tortillas, lasagna, gnat, natal, talc, alcoholic, licensee, seemed, medical, calm, almanac, nacho, choreography, phylum, lumbar, barman, mannequins, insures, respires, resound, underarm, armatures, resides, desideratum, tumult, ultrasound, underdog, dogcatcher, herald, alderwoman, mandarins, insecticides, desires, respirator, torrid, rid, rides, descant, anticlimax, maximum, mum, mummer, meringue, guesser, sermon, monogram, ramrod, rodeo, deodorant, antelopes, peso, esophagus, gusset, setups, upshot, hotel, telescope, open, penicillin, lingos, gossip, sip, siphon, honor, normal, maltreat, eaten, tenet, nether, herpes, pesticides, descend, endow, downfall, alleyway, way, waylay, layman, manicures, reshuffle, flea, lea, leash, ashen, henchman, mandolin, linchpins, inscribes, bestow, townspeople, plectrum, rumbas, baste, sternum, numb, umbilici, icicle, cleaver, vertebra, brains, insouciant, antidepressant, anthem, hemoglobin, binocular, largos, gossamer, mermaid, aid, aides, desperado, adopt, opt, optima, imam, mambos, bosun, sun, sunspot, potpourris, risky, sky, skyscraper, perturbed, bedraggle, glee, lee, leech, echo, choreographer, heraldic, dictum, tumid, midday, day, daybed, bedsides, desktop, topknot, notepaper, periodical, calendar, dare, areas, easel, selfsame, amebas, basins, ins, insulin, linnet, nettlesome, omegas, gasp, aspartame, amend, endures, researcher, herbal, balsas, sass, assault, ultimatum, tumor, mortgagor, gores, resort, orthopaedic, dictatorship, hipper, person, sonar, narc, arc, archduke, ukelele, elegant, anther, hereabout, outfox, fox, foxtrot, rotogravures, restaurant, antechamber, beret, retriever, verbena, enamor, morsel, sellout, outmaneuver, vertical, call, allergenic, niche, chessman, mandolins, insipid, pidgins, install, allures, rescind, indignant, antithetical, calicos, cosmonaut, auto, utopia, piano, another, heretical, calk, alkali, alibi, ibis, bistro, troupe, upend, endorser, serviceman, mandarin, rind, inductee, teepee, pee, peekaboo, bootstrap, rape, apertures, resin, singular, larval, valiant, antiperspirant, antipasto, stop, topical, calisthenic, nicer, cervix, vixen, xenophobic, bicep, cephalic, licit, citizenship, hippopotami, amigos, gospel, pellet, letups, upstart, artificer, cerebellum, lumberman, manic, nicknamed, medic, dickie, kielbasas, sash, ash, ashcan, cannon, nonskid, kid, kidnaper, perjures, resolver, vernacular, larkspur, puree, reefer, ferret, retains, insofar, far, fart, artisan, sandbag, bagel, gelatin, tinsel, selectman, manacle, clever, versus, sustains, inscribed, bedpan, pandemic, microprocessor, sorbet, bet, betcha, char, harem, remodel, deli, elicit, citadel, deliver, verandas, dashikis, kisser, servicemen, menthol, holiday, daydreamer, merchantman, manikins, insane, anew, newsprint, interwove, overreach, achieve, even, venom, nomad, mad, madrigal, gala, alarm, armpit, pitchman, manor, northbound, underbid, bidet, detox, toxemia, miasma, smarten, tenderloins, insult, ultra, travel, velvet, veteran, random, domino, inopportune, uneconomic, microbes, bestir, tiro, ironware, arena, enamel, melodramas, mastodon, don, donut, nut, nutmeg, meg, megalopolis, lissom, sombre, breathe, therefrom, romper, performer, merman, mangrove, overshadow, downcast, astir, tiros, rostra, trachea, heaven, ventricle, clergywoman, maneuver, verbal, ballad, ladyship, hippie, pie, piebald, alderman, manatee, teethe, thereupon, poncho, choicer, ceramic, microscopic, picayune, uneaten, tendon, donor, northeast, astound, underpass, assessor, sorghum, hum, human, mantra, trainee, needlepoint, interplay, laywoman, mannikin, kinsman, mantillas, lassie, sieve, ever, verdigris, risen, sensor, sorrel, relabel, belabor, borsch, schlep, leprechauns, unsnap, nap, napkin, kin, kingpin, pinkeye, eyeglass, assemblyman, manikin, kingship, hip, hippos, postpartum, tumbrel, relic, lichee, heehaw, haw, hawser, servicewoman, many, anyhow, howsoever, vertex, text, extra, trap, rap, rapper, periwig, wigwag, wag, wagon, gonorrhea, heave, aver, vermin, minesweeper, perplex, lexicon, congas, gastronomic, microfiche, cheapen, pentathlon, longhair, air, aircraft, aft, aftertaste, stem, tempos, postwar, war, wart, article, clear, earshot, hotshot, hotbed, bedlam, lam, lambkin, kindergarten, tenser, serum, rumor, mortar, tarmac, macabre, breech, echos, hostel, telescopic, pickerel, relay, laypeople, pleas, east, astronomic, micra, crackpot, pot, potato, atom, tombed, bedbug, bugaboo, bootleg, leg, legato, atop, topple, plethora, orangutang, angora, orangutan, tan, tandem, democrat, rat, rattan, tang, angry, gryphon, honeybee, bee, beeswax, waxen, xenon, nonplus, lustre, trellis, lisle, sleepwear, earwig, wig, wigwam, wampum, pummel, melanomas, massacre, cretin, tin, tint, interviewee, wee, weeper, persimmon, monsignori, origin, gingham, ham, hamper, pericardia, diarrhea, heartthrob, rob, robes, besom, sombreros, rosebud, bud, budgerigar, garret, retrodden, denim, nimbus, bus, bushel, helmet, metaphor, horsefly, flypaper, peritonea, near, ear, earlobes, bestowal, wall, allay, layout, outlast, astrakhan, handicapper, perusal, saltpetre, tremor, moribund, undercut, cut, cutoff, off, offal, falcon, con, consul, sultan, tannin, ninepin, pinball, allegro, grommet, metro, trot, rot, rotten, tenpin, pineapple, plectra, transit, sitar, tar, taro, arousal, salmon, moneybag, bagpipe, ipecac, cache, checkout, outrun, runaround, undersea, sea, sear, earache, cherub, rub, rubicund, underpin, pin, pint, intagli, glib, lib, libel, bellyache, cherubim, bimbos, bosuns, unsound, undertow, tow, towel, wellington, ton, tonsil, silicon, concoct, octet, tetrahedra, drachmae, maestri, tripos, possum, sum, sumac, macro, crocus, custom, tom, tomcat, catsup, sup, superstar, tarpaulin, linchpin, pinpoint, intercom, comet, met, metacarpus, pussycat, catastrophe, phenomenon, nonverbal, ballpoint, interurban, bani, animal, malt, altar, tartar, tarot, rotund, undergrad, radio, diocesan, sandbar, bar, barren, renewal, walkout, outstrip, ripen, pen, pencil, cilantro, trout, outran, rancor, corncob, cob, cobra, bra, brag, rag, ragas, gas, gasohol, holdout, output, put, putsch, schwas, was, waste, stereo, reoccur, cur, curb, urban, ban, bantam, tam, tamp, ampul, pullout, outwit, wit, withal, halo, alohas, hasp, asparagus, gusto, stove, overlap, lapel, pelvis, visit, sit, sitcom, compendia, diadem, demigod, god, goddam, dam, dampen, pennon, non, noncom, compel, pelican, cancan, can, cancel, celesta, starlit, lit, litmus, muscat, cat, catnap, naphtha, than, handcar, carpel, pellagra, grammar, mar, mariachi, chichi, chi, chimp, imp, impel, pelvic, vicar, car, caribou, bourgeoisie, siesta, stab, tab, tabu, abut, but, butterfat, fat, fathom, homespun, pun, puns, unsheathe, the, theorem, remove, overtax, tax, taxicab, cab, cabal, balsam, sambas, basal, salamis, missal, salt, altho, tho, thou, housebound, underground, underclassman, man, mannikins, insectivores, resonant, antelope, operator, torn, ornamental, tallow, low, lowered, reddens, enshrine, inefficient, entertainer, nerves, vestiges, gesturing, ingested, tediousness, essentials]

Core ideas

In Approximating longest directed paths and cycles, Bjorklund, Husfeldt, and Khanna, Lecture Notes in Computer Science (2004), 222-233, the authors propose that in sparse expander graphs a long path can be found by a greedy search which selects at each step the neighbour of the current tail of the path which spans the largest subgraph in G', where G' is the original graph with the vertices in the current path deleted. I'm not sure of a good way to test whether a graph is an expander graph, but we're certainly dealing with a sparse graph, and since its core is about 20000 vertices and it has a diameter of only 15 it must have good expansion properties. So I adopt this greedy heuristic.

Given a graph G(V, E), we can find how many vertices are reachable from each vertex using Floyd-Warshall in Theta(V^3) time, or using Johnson's algorithm in Theta(V^2 lg V + VE) time. However, I know that we're dealing with a graph which has a very large strongly connected component (SCC), so I take a different approach. If we identify SCCs using Tarjan's algorithm then we also get a topological sort for the compressed graph G_c(V_c, E_c), which will be much smaller, in O(E) time. Since G_c is a DAG, we can compute reachability in G_c in O(V_c^2 + E_c) time. (I have subsequently discovered that this is hinted at in exercise 26-2.8 of CLR).

Since the dominant factor in the running time is E, I optimise it by inserting dummy nodes for the prefixes/suffixes. So rather than 151 * 64 = 9664 edges from words ending -res to words starting res- I have 151 edges from words ending -res to #res# and 64 edges from #res# to words starting res-.

And finally, since each search takes about 4 minutes on my old PC, I try to combine the results with previous long paths. This is much faster, and turned up my current best solution.



package org.cheddarmonk.math.graph;

import java.util.Set;

public interface Graph<V> {
    public Set<V> getAdjacent(V node);
    public double getWeight(V from, V to);


package org.cheddarmonk.math.graph;

import java.util.*;

public class MutableGraph<V> implements Graph<V> {
    private Map<V, Map<V, Double>> edgesBySource = new HashMap<V, Map<V, Double>>();

    public void addEdge(V from, V to, double weight) {
        if (!edgesBySource.containsKey(to)) edgesBySource.put(to, new HashMap<V, Double>());
        Map<V, Double> e = edgesBySource.get(from);
        if (e == null) edgesBySource.put(from, e = new HashMap<V, Double>());
        if (e.containsKey(to)) throw new IllegalArgumentException("There is already an edge between the vertices");
        e.put(to, weight);

    public Set<V> getAdjacent(V node) {
        Map<V, Double> e = edgesBySource.get(node);
        if (e == null) throw new IllegalArgumentException("node doesn't appear to be in the graph");
        return Collections.unmodifiableSet(e.keySet());

    public double getWeight(V from, V to) {
        Map<V, Double> e = edgesBySource.get(from);
        if (e == null) throw new IllegalArgumentException("from doesn't appear to be in the graph");
        if (!edgesBySource.containsKey(to)) throw new IllegalArgumentException("to doesn't appear to be in the graph");

        Double c = e.get(to);
        return c == null ? 0 : c.doubleValue();


package org.cheddarmonk.math.graph;

import java.util.*;

* A helper class for finding the strongly connected components of a graph using Tarjan's algorithm.
* http://en.wikipedia.org/wiki/Tarjan%27s_strongly_connected_components_algorithm
public class StronglyConnectedComponents<V> {
    private final Graph<V> g;
    private List<Set<V>> topologicallySortedSccs = new ArrayList<Set<V>>();

    private final LinkedList<V> S = new LinkedList<V>();
    private final Set<V> S2 = new HashSet<V>();
    private final Map<V, Integer> index = new HashMap<V, Integer>();
    private final Map<V, Integer> lowlink = new HashMap<V, Integer>();

    private StronglyConnectedComponents(Graph<V> g) {
        this.g = g;

    private void strongConnect(V v) {
        int idx = index.size();
        index.put(v, idx);
        lowlink.put(v, idx);


        for (V w : g.getAdjacent(v)) {
            if (!index.containsKey(w)) {
                if (lowlink.get(w) < lowlink.get(v)) {
                    lowlink.put(v, lowlink.get(w));
            else if (S2.contains(w)) {
                if (index.get(w) < lowlink.get(v)) {
                    lowlink.put(v, index.get(w));

        if (lowlink.get(v).equals(index.get(v))) {
            Set<V> scc = new HashSet<V>();
            V w;
            do {
                w = S.pop();
            } while (!w.equals(v));


    public static <V> List<Set<V>> analyse(Graph<V> g, Set<V> sources) {
        if (g == null) throw new IllegalArgumentException("g");

        StronglyConnectedComponents<V> scc = new StronglyConnectedComponents<V>(g);
        for (V v : sources) {
            if (!scc.index.containsKey(v)) {

        return scc.topologicallySortedSccs;


package org.cheddarmonk.ppcg;

import java.io.*;
import java.util.*;
import org.cheddarmonk.math.graph.*;

public class PPCG44922 {
    private static final String path = "/usr/share/dict/words";
    private static Set<String> allWords;
    private static Graph<String> fullGraph;

    public static void main(String[] args) {

        Random rnd = new Random();
        List<String> a = search(rnd);
        List<String> b = search(rnd);
        List<String> chain = spliceChains(a, b);

    private static List<String> search(Random rnd) {
        List<String> chain = new ArrayList<String>();
        chain.add(selectOptimalReachabilityCount(fullGraph, allWords, rnd));
        while (true) {
            String tail = chain.get(chain.size() - 1);
            FilteredGraph g = new FilteredGraph(chain);

            // We know that tail only has one successor, so skip ahead.
            Set<String> candidates = new HashSet<String>(fullGraph.getAdjacent(suffix(tail)));
            if (candidates.isEmpty()) break;

            chain.add(selectOptimalReachabilityCount(g, candidates, rnd));

        Iterator<String> it = chain.iterator();
        while (it.hasNext()) {
            if (it.next().charAt(0) == '#') it.remove();
        return chain;

    private static List<String> spliceChains(List<String> a, List<String> b) {
        Set<String> intersect = new HashSet<String>(b);
        if (intersect.isEmpty()) return null;

        // Splice the longest bits. To avoid cycles, we look for intersection points which have the same set of reached intersection points.
        // Thus to get from one to the next we can take either route without violating the unique occurrence of each element in the spliced chain.
        Set<String> left = new HashSet<String>();
        Set<String> right = new HashSet<String>();
        List<String> newChain = new ArrayList<String>();

        // NB We assume that either a(0) and b(0) are the same or neither is in intersect.
        // This is a safe assumption in practice because they're both "wad".
        int idxA = 0, idxB = 0, nextA = 0, nextB = 0;
        while (idxA < a.size()) {
            while (nextA < a.size() && !intersect.contains(a.get(nextA))) nextA++;
            String tailA = nextA < a.size() ? a.get(nextA) : "";

            while (nextB < b.size() && !intersect.contains(b.get(nextB))) nextB++;
            String tailB = nextB < b.size() ? b.get(nextB) : "";

            if (left.equals(right) && tailA.equals(tailB)) {
                // We take the longer of idxA to nextA-1 or idxB to nextB - 1.
                if (nextA - idxA > nextB - idxB) newChain.addAll(a.subList(idxA, nextA));
                else newChain.addAll(b.subList(idxB, nextB));

                idxA = nextA;
                idxB = nextB;

        if (new HashSet<String>(newChain).size() == newChain.size()) return newChain;
        throw new IllegalStateException();

    private static void loadGraph() {
        Set<String> words = new HashSet<String>();
        Set<String> prefixes = new HashSet<String>();
        Set<String> suffixes = new HashSet<String>();
        try {
            BufferedReader br = new BufferedReader(new InputStreamReader(new FileInputStream(path), "UTF-8"));
            String line;
            while ((line = br.readLine()) != null) {
                if (line.length() >= 3) {
        catch (IOException ioe) {
            throw new RuntimeException(ioe);

        // Filter down to a core with decent reachability.
        MutableGraph<String> g = new MutableGraph<String>();
        Iterator<String> it = words.iterator();
        while (it.hasNext()) {
            String line = it.next();
            if (prefixes.contains(prefix(line)) && prefixes.contains(suffix(line))) {
                // In the interests of keeping the number of edges down, I insert fake vertices.
                g.addEdge(prefix(line), line, 1);
                g.addEdge(line, suffix(line), 1);
            else it.remove();

        fullGraph = g;
        allWords = Collections.unmodifiableSet(words);

    private static String prefix(String word) {
        return "#" + word.substring(0, 3) + "#";

    private static String suffix(String word) {
        return "#" + word.substring(word.length() - 3, word.length()) + "#";

    private static <V> Map<V, Integer> reachabilityCount(Graph<V> g, Set<V> sources) {
        List<Set<V>> sccs = StronglyConnectedComponents.analyse(g, sources);
        int n = sccs.size();

        // Within a strongly connected component, each vertex can reach each other vertex.
        // Then we need to also take into account the other SCCs which they can reach.
        // We can exploit the fact that we already have a topological sort of the DAG of SCCs to do this efficiently.
        Map<V, Integer> index = new HashMap<V, Integer>();
        for (int i = 0; i < n; i++) {
            for (V v : sccs.get(i)) index.put(v, i);

        BitSet[] reachableSccs = new BitSet[n];
        Map<V, Integer> reachabilityCounts = new HashMap<V, Integer>();
        for (int i = 0; i < n; i++) {
            Set<V> scc = sccs.get(i);
            reachableSccs[i] = new BitSet(n);
            for (V v : scc) {
                for (V w : g.getAdjacent(v)) {
                    int j = index.get(w);
                    if (j < i) reachableSccs[i].or(reachableSccs[j]);

            int count = 0;
            for (int j = reachableSccs[i].nextSetBit(0); j >= 0; j = reachableSccs[i].nextSetBit(j+1)) {
                count += sccs.get(j).size();
            for (V v : scc) {
                reachabilityCounts.put(v, count);

        return reachabilityCounts;

    private static <V extends Comparable<? super V>> V selectOptimalReachabilityCount(Graph<V> g, Set<V> candidates, Random rnd) {
        Map<V, Integer> r = reachabilityCount(g, candidates);

        int max = 0;
        List<V> attaining = new ArrayList<V>();
        for (V candidate : candidates) {
            int score = r.get(candidate);
            if (score > max) {
                max = score;
            if (score == max) attaining.add(candidate);

        return selectRandom(attaining, rnd);

    private static <T extends Comparable<? super T>> T selectRandom(Collection<T> elts, Random rnd) {
        List<T> deterministic = new ArrayList<T>(elts);
        Collections.shuffle(deterministic, rnd);
        return deterministic.get(0);

    private static class FilteredGraph implements Graph<String> {
        private final Set<String> filteredVertices;

        public FilteredGraph(Collection<String> filteredVertices) {
            this.filteredVertices = new HashSet<String>(filteredVertices);

        public Set<String> getAdjacent(String node) {
            if (filteredVertices.contains(node)) return Collections.emptySet();

            Set<String> adj = new HashSet<String>(fullGraph.getAdjacent(node));
            return adj;

        public double getWeight(String from, String to) {
            throw new RuntimeException("Not used");
  • \$\begingroup\$ "erecting" hmm.. \$\endgroup\$ Feb 8 '17 at 14:06

Ruby, 1701

"Del" -> "ersatz's" (full sequence)

Trying to find the optimal solution proved too costly in time. So why not pick random samples, cache what we can and hope for the best?

First a Hash is built that maps prefixes to full worlds starting with that prefix (e.g. "the" => ["the", "them", "their", ...]). Then for every word in the list the method sequence is called. It gets the words that could possibly follow from the Hash, takes a sample of 100 and calls itself recursively. The longest is taken and proudly displayed. The seed for the RNG (Random::DEFAULT) and the length of the sequence are also displayed.

I had to run the program a few times to get a good result. This particular result was generated with seed 328678850348335483228838046308296635426328678850348335483228838046308296635426.


require "json"

def prefix(word); word[0, 3];  end
def suffix(word); word[-3, 3]; end

def sequence(word, prefixes, skip)
  if SUBS.key?(word) && (SUBS[word] - skip) == SUBS[word]
    return SUBS[word]

  skip         += [word] 
  suffix        = suffix(word)
  possibilities = (prefixes[suffix] ||= []) - skip

  if possibilities.empty?
    return [word]
    sequences = possibilities.sample(100).map do |possibility|
      sequence(possibility, prefixes, skip)

    return SUBS[word] = [word] + sequences.max_by(&:size)

def correct?(sequence)
  once_only = sequence.all? { |y| sequence.count(y) == 1 }
  following = sequence.each_cons(2).all? { |a,b| a[-3,3] == b[0,3] }

  return once_only && following

words = open("words.txt", "r", &:read).split.select { |word| word.size >= 3 }

SUBS     = {}

# Built a Hash that maps prefixes to an Array of words starting with the prefix.
words.each do |word|
  prefix = prefix(word)

  PREFIXES[prefix] ||= []
  PREFIXES[prefix] << word

longest = [1]

words.each do |word|

  sequence = sequence(word, PREFIXES, [word])

  if sequence.size > longest.size
    longest = sequence

puts longest.inspect
puts "Generated with seed: #{Random::DEFAULT.seed}"
puts "Length: #{longest.size}"
puts "Correct: #{correct?(longest)}"
  • \$\begingroup\$ I didn't think of randomly sampling the possibilities for the next word! I'm using that idea! \$\endgroup\$
    – KSFT
    Jan 26 '15 at 20:22
  • \$\begingroup\$ How long did this take to run? \$\endgroup\$
    – KSFT
    Jan 26 '15 at 20:25
  • \$\begingroup\$ I didn't time it, but I estimate about 15-20 minutes (I let it run over dinner). \$\endgroup\$
    – britishtea
    Jan 26 '15 at 20:44
  • \$\begingroup\$ My Python one is still building the dict (hash)... \$\endgroup\$
    – KSFT
    Jan 26 '15 at 20:45
  • \$\begingroup\$ Building the dictionary should be fast (a single iteration trough all words). Ruby reports 0.0996 seconds. \$\endgroup\$
    – britishtea
    Jan 26 '15 at 20:57

Score: 1631 1662 words

['Aachen', 'hen', 'henceforward', 'ardor', 'dorsal', 'salmon', 'monolog', 'log', 'logjam', 
'jam', 'jamb', 'ambassador', 'dormouse', 'useable', 'bleeding', 'ingenious', 'ouster', 
'terminable', 'bleakness', 'essay', 'say', 'saying', 'ingress', 'essences', 'cession', 
'ionosphere', 'ere', 'erecting', 'ingratiating', 'ingrate', 'ate', 'ateliers', "ersatz's"]

You can find the whole sequence here: http://pastebin.com/TfAvhP9X

I don't have the complete source code. I was trying out different approaches. But here are some code snippets, that should be able to generate a sequence of about the same length. Sorry, it's not very beautiful.

Code (Python):

First some preprocessing of the data.

from collections import defaultdict

with open('words') as f:
    words = [line.strip() for line in f]
words = [word for word in words if len(word)>=3 and word[-2:]!="'s"]

word_connections = defaultdict(list)
for word in words:

Then I defined a recursive function (Depth first search).

global m
def find_chain(chain):
    s = set(word_connections[chain[-1][-3:]])-set(chain)
    if len(s)== 0:
        global m
        if len(chain) > m:
            print(len(chain), chain)
        for w in s:
            if w not in chain:
                find_chain(chain + [w])

for word in words:

Of course this takes way too long. But after some time it found a sequence with 1090 elements, and I stopped.

The next thing to do, is a local search. For each two neighbors n1,n2 in the sequence, I try to find a sequence starting at n1 and ending at n2. If such a sequence exists, I insert it.

def tryimpove(chain, length, startvalue=0):
    for i in range(startvalue,len(chain)-1):
        for sub in sorted(short_sequences([chain[i]],length,chain[i+1]),key=len,reverse=True):

            if len(s & set(sub))==1:
                return tryimpove(chain,length,i)
    return chain

def short_sequences(chain,length,end):
    if 2 <= len(chain):
        if chain[-1][-3:]==end[:3]:
            yield chain
    if len(chain) < length:
        s = set(word_connections[chain[-1][-3:]])-set(chain)
        for w in s:
            for i in short_sequences(chain + [w],length,end):
                yield i

for m in range(5, 100):
    seq = tryimpove(seq,m)

Of course I also had to stop the program manually.

  • \$\begingroup\$ How long it takes to get your result ? \$\endgroup\$
    – Hacketo
    Jan 26 '15 at 9:50
  • \$\begingroup\$ About an hour to generate the 1090 sequence, and another hour to do the local search. \$\endgroup\$
    – Jakube
    Jan 26 '15 at 9:52

PHP, 1742 1795

I've been messing about with PHP on this. The trick is definitely culling the list to the around 20k that are actually valid, and just throwing the rest away. My program does this iteratively (some words it throws away in the first iteration mean others are no longer valid) at the start.

My code is horrible, it uses a number of global variables, it uses way too much memory (it keeps a copy of the entire prefix table for every iteration) and it took literally days to come up with my current best, but it still manages to be winning - for now. It starts off pretty quickly but gets slower & slower as time goes on.


  function show_list($list)
      foreach ($list as $item => $blah)
      $st.="$item ";
      return rtrim($st);

  function mysort($a,$b)
      global $parts;
      return $b_count-$a_count;

  function recurse($line,$list,$parts)
    global $lines; 
    global $max;
    global $finished;
    global $best;
    if ($finished)
    if (isset($parts[$end]))
        if ($maxp<0)
        for ($a=1;$a<3;$a++)

        foreach ($parts[$end] as $key => $endpart)

            if (!isset($list[$endpart]))

    if ($count>$max)
        //echo "current best: $count\n";
        file_put_contents('best.txt',show_list($list) . "\n",FILE_APPEND);

  function cull_lines(&$lines)
      global $parts;
          foreach ($lines as $line)
              if (strlen($line)<3)
              if (strlen($line)>3)
          foreach ($lines as $key => $line)
              if (strlen($line)<3 || !isset($parts[$end]) || !isset($end_parts[substr($line,0,3)] ) )
          foreach ($parts as $part)
            if (count($part)==1)
                $this_part = substr($part[0],0,3);
                $this_min = 10000;
                $this_key = 0;
                if (strlen($part[$source_part])==3)
                    foreach ($lines as $key => $line)
                    if ($line == $part[$source_part])
                elseif (isset($end_parts[$this_part]))
                    foreach ($end_parts[$this_part] as $key => $end_part)
                        if (isset($parts[substr($end_part,0,3)]))
                            if ($n<$this_min)

                    foreach ($lines as $key => $line)
                    if ($line == $part[$source_part])

                    elseif ($line == $end_parts[$this_part][$this_key])
                        $lines[$key].=' ' . $part[$source_part];

          echo "$wordcount words left\n";
      while ($wordcount!=count($lines));

  $lines = explode("\n",file_get_contents('words.txt'));
  foreach ($parts as $key=>$part)

  foreach ($lines as $rand => $blah)
      if (mt_rand(0,$n)==1)
  echo "Starting with $rand...\n";


An obvious improvement would be using an orphan word for the start & finish.

Anyway, I really don't know why my pastebin list was moved to a comment here, it's back as a link to pastebin as I've now included my code.


  • \$\begingroup\$ I included the code from the Pastebin in your answer, then we don't have to rely on an external domain; in case Pastebin goes down, we can still see your code here. \$\endgroup\$
    – ProgramFOX
    Jan 29 '15 at 16:30
  • \$\begingroup\$ Well, I was happy to have the winning answer for a short while at least. Nice work Peter Taylor... \$\endgroup\$ Jan 31 '15 at 8:06
  • \$\begingroup\$ Oh, now I see that I shouldn't have added the list to your answer; I was a bit confused and I mixed up code and word list, and included the list in your answer. I'm sorry for that. \$\endgroup\$
    – ProgramFOX
    Jan 31 '15 at 15:46

Python : 1702 - 1704 - 1733 words

I used a Dict to map all the prefixes to all the words , as

    "AOL" : [
        "AOL", "AOL's"
    "oct" : [

Edit little improvement: removing all useless words at the beginning, if their suffixes aren't in the prefixes list (would obviously be a end word)

Then take a word in the list and browse the prefix map like a tree Node

import sys, fileinput
def p(w):return w[:3]

class Computing:
    def __init__(_):
        _.wordData = []; _.prefixes = {}
        _.seq = []; _.bestAttempt = []
        _.stop = False
        for l in fileinput.input():
            word = l.strip()
            if len(word) > 2:_.wordData.append(_.addPfx(word, p(word)))
        _.fromI = 0; _.toI = len(_.wordData)

    def rmUseless(_):
        for w in _.wordData:
            if not w[-3:] in _.prefixes: _.wordData.remove(_.rmPfx(w,p(w)))

    def addPfx(_, w, px):
        if not px in _.prefixes:_.prefixes[px] = []
        return w
    def rmPfx(_,w,px):
        if len(_.prefixes[px]) == 0:del _.prefixes[px];
        return w

    def findBestSequence(_):
        def pickItem():
            r = None
            if _.fromI < _.toI:r = _.wordData[_.toI-1];_.toI -= 1;
            return r
        while not _.stop:
            w = pickItem()
            if not w:break;
            _.seq = [_.rmPfx(w,p(w))]
            _.addPfx(w, p(w))

        print " ".join(_.bestAttempt)

    def checkForNextWord(_):
        _.stop = len(_.seq) >= 1733
        cw = _.seq[-1]
        if not _.stop:
            if cw[-3:] in _.prefixes:
                lW = []
                for word in _.prefixes[cw[-3:]]:
                    if not word[-3:] in lW:
                        if _.stop :break;
                        _.addPfx(_.seq.pop(), p(word))
                    if _.stop :break;
        if len(_.bestAttempt) < len(_.seq):_.bestAttempt = _.seq[:];


The program need a number of words to know when stop, can be find as 1733 in the method checkForNextWord̀

The program need the file path as argument

Not very pythonic but I tried.

It took less than 2 min to compute this sequence : full output:

études desalinate ate ateliers ... blest esteeming ingenuous ousters


Score: 249 500 1001

Aachen --> hen --> henceforward --> ardent --> entail --> ail --> ailed --> led --> ledger --> gerbil --> bilateral --> rallying --> ingenious --> ousted --> tedious --> ouster --> terabit --> bit --> bitched --> hedgehog --> hog --> hogan --> gander --> derail --> ailerons --> onset --> set --> setback --> acknowledgement --> entailed --> ledgers --> ersatzes --> zest --> established --> hedgerow --> row --> rowboat --> oat --> oaten --> ten --> tenable --> bleach --> ache --> cheapen --> pen --> penalizes --> zestful --> fulcra --> crab --> rabbinical --> calabash --> ash --> ashamed --> medal --> dale --> ale --> alerted --> tediously --> sly --> slyest --> establishes --> hes --> hesitant --> ant --> antacid --> cider --> derailed --> ledges --> gestated --> tediousness --> essay --> say --> saying --> ingeniously --> slyness --> essaying --> ingenuous --> ousting --> ingenuousness --> essayist --> isthmus --> muscat --> cat --> cataclysmic --> mice --> ice --> iceberg --> erg --> ergonomic --> micra --> crabbed --> bed --> bedazzles --> lesbians --> answer --> were --> ere --> erecting --> ingest --> establishing --> ingesting --> ingestion --> ion --> ionization --> ionizer --> zero --> erode --> ode --> odes --> desalinates --> test --> establishment - -> entailing --> ingot --> got --> gotten --> tenant --> antagonist --> isthmuses --> sesame --> amebas --> basal --> salaamed --> medallion --> ionizing --> ingraining --> ingrains --> ins --> insane --> anecdotal --> talc --> alcohol --> hold --> old --> olden --> den --> denature --> urea --> reach --> ached --> hedges --> gestates --> testable --> bleached --> hedging --> ingrates --> testament --> entangle --> gleamed --> medallions --> onshore --> ore --> oregano --> anodes --> desalinating --> ingratiates --> testates --> tester --> terabits --> its --> itself --> elf --> elfin --> fin --> finagle --> gleaming --> ingratiating --> ingratiatingly --> glycerin --> rind --> indebtedness --> essences --> cesareans --> answerable --> bleacher --> her --> herald --> alder --> derailing -- > ingredient --> entanglement --> entangles --> lesion --> ionosphere --> erection --> ionospheres --> resale --> alerting --> ingresses --> sesames --> mes --> mesas --> sash --> ashcan --> can --> canard --> ardor --> dorkiest --> estates --> testes --> testicle --> cleaner --> nerdiest --> esteemed --> meddles --> lessee --> see --> seeded --> dedicates --> testicles --> lessen --> senates --> testiest --> esteeming --> ingrown --> own --> owner --> nerves --> vesicle --> cleanest --> ester --> terabytes --> testis --> tissue --> sue --> suede --> edema --> emaciates --> testosterone --> one --> ones --> nest --> esthetes --> testy --> sty --> styes --> yes --> yeshivas --> vascular --> larches --> hesitatingly --> glycerine --> inedible --> bleaches --> hesitates --> testifying --> ingrate --> ate --> ateliers

Here's my code:

import sys
w=[i for i in open("words.txt").read().split("\n") if len(i)>=3 and "'" not in i]
b=[i[:3] for i in w]
e=[i[-3:] for i in w]
a=[i for i in zip(*(w,b,e))]
def pathfrom(i,a,l):
    for j in a:
        if j[1]==i[2]:
            r=pathfrom(j,[m for m in a if m!=j],l+1)
            if r[1]:
                return path,r[1]
            if len(path)>len(longpath):
    if l>=250:
        return longpath,True
    return longpath,False
for i in a[:2]:
    print i
    p=pathfrom(i,[j for j in a if i!=j],1)
    if len(p)>len(chain_):
        print p
    print p

Edit: 1001:


Edit: 500:

Aachen --> hen --> henceforward --> ardent --> entail --> ail --> ailed --> led --> ledger --> gerbil --> bilateral --> rallying --> ingenious --> ousted --> tedious --> ouster --> terabit --> bit --> bitched --> hedgehog --> hog --> hogan --> gander --> derail --> ailerons --> onset --> set --> setback --> acknowledgement --> entailed --> ledgers --> ersatzes --> zest --> established --> hedgerow --> row --> rowboat --> oat --> oaten --> ten --> tenable --> bleach --> ache --> cheapen --> pen --> penalizes --> zestful --> fulcra --> crab --> rabbinical --> calabash --> ash --> ashamed --> medal --> dale --> ale --> alerted --> tediously --> sly --> slyest --> establishes --> hes --> hesitant --> ant --> antacid --> cider --> derailed --> ledges --> gestated --> tediousness --> essay --> say --> saying --> ingeniously --> slyness --> essaying --> ingenuous --> ousting --> ingenuousness --> essayist --> isthmus --> muscat --> cat --> cataclysmic --> mice --> ice --> iceberg --> erg --> ergonomic --> micra --> crabbed --> bed --> bedazzles --> lesbians --> answer --> were --> ere --> erecting --> ingest --> establishing --> ingesting --> ingestion --> ion --> ionization --> ionizer --> zero --> erode --> ode --> odes --> desalinates --> test --> establishment - -> entailing --> ingot --> got --> gotten --> tenant --> antagonist --> isthmuses --> sesame --> amebas --> basal --> salaamed --> medallion --> ionizing --> ingraining --> ingrains --> ins --> insane --> anecdotal --> talc --> alcohol --> hold --> old --> olden --> den --> denature --> urea --> reach --> ached --> hedges --> gestates --> testable --> bleached --> hedging --> ingrates --> testament --> entangle --> gleamed --> medallions --> onshore --> ore --> oregano --> anodes --> desalinating --> ingratiates --> testates --> tester --> terabits --> its --> itself --> elf --> elfin --> fin --> finagle --> gleaming --> ingratiating --> ingratiatingly --> glycerin --> rind --> indebtedness --> essences --> cesareans --> answerable --> bleacher --> her --> herald --> alder --> derailing -- > ingredient --> entanglement --> entangles --> lesion --> ionosphere --> erection --> ionospheres --> resale --> alerting --> ingresses --> sesames --> mes --> mesas --> sash --> ashcan --> can --> canard --> ardor --> dorkiest --> estates --> testes --> testicle --> cleaner --> nerdiest --> esteemed --> meddles --> lessee --> see --> seeded --> dedicates --> testicles --> lessen --> senates --> testiest --> esteeming --> ingrown --> own --> owner --> nerves --> vesicle --> cleanest --> ester --> terabytes --> testis --> tissue --> sue --> suede --> edema --> emaciates --> testosterone --> one --> ones --> nest --> esthetes --> testy --> sty --> styes --> yes --> yeshivas --> vascular --> larches --> hesitatingly --> glycerine --> inedible --> bleaker --> keratin --> tin --> tincture --> urethras --> rascal --> calamine --> ineducable --> bleakest --> esthetic --> tic --> tick --> ickiest --> estimable --> bleariest --> estimator --> tor --> torched --> hedonistic --> ticker --> kerchieves --> vesicles --> lessens --> ensconced --> cedar --> dare --> are --> area --> reachable --> bleat --> eat --> eatable --> bleeder --> derailment --> enter --> term --> ermine --> ineffable --> bleeped --> pedagog --> goggle --> gleans --> answered --> red --> redbreast --> aster --> termagant --> antagonistic --> ticket --> kettledrum --> rum --> rumbas --> basalt --> altar --> tar --> tarantulas --> lasagna --> gnarliest --> estrangement --> entered --> redcap --> cap --> capable --> bleeps --> epsilon --> loneliest --> estranges --> gestured --> redcaps --> apse --> pseudonym --> nymphomania --> niacin --> cinchonas --> nasal --> salable --> blend --> end --> endanger --> geriatric --> rice --> icebound --> undeceives --> vesper --> per --> perambulator --> tore --> ores --> resales --> lesser --> sera --> era --> eras --> rash --> ashen --> henchman --> man --> manacle --> cleanliest --> estrogen --> gendarmes --> mescal --> calcine --> inefficient --> entertainingly --> glycerol --> role --> oleander --> derangement --> entertainment --> entertains --> insatiable --> blended --> deduced --> cedars --> arsenic --> nice --> icebox -- > box --> boxcar --> car --> caracul --> cullender --> deranges --> gestures --> reschedules --> lesson --> son --> sonar --> narc --> arc --> arcade --> adenoidal --> dales --> lessor --> sorbet --> bet --> betaken --> ken --> kens --> ensemble --> blender --> deride --> idea --> deacon --> con --> concave --> avenger --> germane --> anemia --> miaowed --> wed --> wedded --> deducible --> blent - -> enthrall --> all --> allay --> lay --> layaway --> way --> wayfarer --> reran --> ran --> rancher --> heralded --> deductible --> blessed --> sedan --> danced --> cedes --> descant --> anteater --> termed --> meddlesome --> omegas --> gas --> gash --> ashram --> ram --> ramble --> blew --> lewder --> derides --> descend --> endangered --> redcoat --> oath --> atherosclerosis --> sis --> sisal - -> salad --> lad --> ladder --> derivatives --> vessel --> seldom --> domains -- > inscribed --> bedbug --> bug --> bugaboo --> boo --> boobed --> bedder --> derives --> vestiges --> gesundheit --> either --> heraldic --> dice --> icebreaker --> kerosene --> enema --> email --> ailment --> enthronement --> enthuse --> use --> used --> sedater --> terminator --> toreador --> dormant --> antebellum - -> lumbago --> ago --> agonizingly --> glycogen --> gender --> dermis --> misadventures --> rescind --> indecent --> enthused --> sedatives --> vestment --> enthusiast --> astir --> tirades --> descendant --> antecedent --> entice --> icecap --> capacitor --> torment --> enticed --> cedilla --> llama --> amalgam --> gambit --> bitches --> hesitates --> testifying --> ingrate --> ate --> ateliers


Mathematica 1482 1655

Something to get started...


Links are the intersection prefixes and suffixes for words.

linkableWords=(wds=RandomSample@Select[words,MemberQ[links,StringTake[#,3]]\[And]MemberQ[links,StringTake[#,-3]]& ])/.
w_String:> {w,StringTake[w,3],StringTake[w,-3]}

Edges are all the directed links from a word to other words:

edges[{w_,fr_,fin_}]:= Cases[linkableWords,{w1_,fin,_}:> (w\[DirectedEdge]w1)]

Find a path between "mend" and "zest".

FindPath[g, "begin", "end", {1480, 3000}, 1][[1]]

Result (1655 words)

{"mend", "endorser", "server", "vertebral", "rallying", "ingrains", 
"insurrectionist", "isthmus", "mussels", "elsewhere", "erection", 
"ionizes", "zestful", "fullness", "essaying", "ingeniously", 
"slyest", "estimator", "tornados", "doses", "sesame", "amebic", 
"bicycled", "ledges", "gestation", "ionizing", "ingratiates", 
"testifying", "ingesting", "inglorious", "ouster", "terminated", 
"tediousness", "essayist", "isthmuses", "session", "ion", 
"ionization", "ionospheres", "resubmitted", "tedious", "ousting", 
"ingest", "ester", "terminates", "testicle", "cleanliness", "essay", 
"say", "saying", "ingratiating", "ingratiatingly", "glycerine", 
"inefficient", "entrances", "cesarians", "answering", "ingenious", 
"ousted", "tediously", "sly", "slyness", "essences", "cesareans", 
"answer", "were", "erecting", "ingredient", "enterprises", 
"sessions", "onshore", "oregano", "anorak", "raking", "ingraining", 
"ingrown", "owner", "nerdiest", "estranging", "ingot", "gotten", 
"tendonitis", "tissue", "suede", "edelweiss", "issuing", "ingestion", 
"ionosphere", "erections", "onset", "settles", "lesion", "ionizer", 
"zeroing", "ingresses", "sesames", "mesmerizing", "ingrates", 
"testes", "testiest", "estrangement", "entail", "ail", "ailment", 
"entice", "icecap", "captivates", "testy", "sty", "stylistic", 
"tickles", "lessee", "seeded", "deductibles", "lesser", 
"servicewoman", "many", "anymore", "ores", "resourceful", "fullback", 
"acknowledgment", "entertainer", "nerves", "vest", "esteemed", 
"mediates", "testament", "entered", "redbreast", "astonishes", 
"hesitatingly", "glycogen", "genera", "eras", "rashes", "hesitates", 
"testicles", "lest", "establishment", "entwines", "nest", "estates", 
"testates", "testosterone", "oneself", "elf", "elfin", "fingered", 
"redcaps", "apse", "pseudonym", "nymphomania", "niacin", "cinemas", 
"masochistic", "tickled", "led", "ledger", "geriatric", "rice", 
"icebreaker", "kerosine", "inexperienced", "ceded", "deductible", 
"blew", "lewder", "derivable", "blemished", "hedgerow", "rowel", 
"welfare", "arena", "enamel", "melded", "dedicates", "tester", 
"terabit", "bitmap", "mapped", "pedicures", "restored", "redeemer", 
"merchantman", "manipulator", "torpedos", "dosed", "seduced", 
"cedilla", "llano", "another", "heretic", "tic", "ticker", "keratin", 
"tinctures", "restaurateur", "euros", "rosettes", "testable", 
"bleaker", "kerosene", "energizer", "zero", "eroded", "deduced", 
"cedar", "dare", "ares", "respondent", "entranced", "cedillas", 
"lasagnas", "nastiest", "esthetic", "ticket", "ketches", "hes", 
"hesitant", "antipasto", "stoppered", "redounded", "deducible", 
"bleeped", "pedant", "antimatter", "terminable", "blent", "enthuse", 
"user", "serenade", "adenoidal", "dales", "lessen", "sentimental", 
"talker", "kerchieves", "vestry", "try", "tryout", "outdone", "ones", 
"nestles", "lesson", "songwriter", "terrapin", "pinched", 
"hedonistic", "tick", "ickiest", "established", "hedgehog", "hogan", 
"gander", "derringer", "gerbil", "billboard", "ardor", "dorkiest", 
"estrogen", "gent", "entirety", "etymological", "calk", "alkalis", 
"lissome", "omegas", "gasolene", "enema", "emaciates", "test", 
"estranges", "gestured", "redeemed", "medic", "diced", "cedars", 
"arsenic", "nice", "iceberg", "erg", "ergonomic", "microcomputer", 
"terser", "sergeant", "antipastos", "tost", "osteopathy", "thy", 
"thymus", "mussiest", "estimable", "blend", "endeavored", "redound", 
"undercover", "verbal", "balk", "alkali", "alibi", "ibis", "bison", 
"sonar", "narcosis", "sister", "terraced", "cede", "edema", 
"emancipator", "torpor", "portraiture", "urea", "reassign", 
"ignoble", "blenched", "hedges", "gesture", "urethras", "raspy", 
"spyglass", "ass", "assailant", "antiquarians", "answered", 
"reduced", "cedes", "despair", "airfares", "resumed", "medicine", 
"ineffable", "bleacher", "herdsmen", "menhaden", "dent", 
"entitlement", "enticement", "entangle", "gleamed", "medullas", 
"lassie", "sieve", "even", "vender", "derivatives", "vessel", 
"selectmen", "mentor", "toreador", "dormer", "meringue", "guerrilla", 
"llanos", "nosedove", "overact", "actionable", "bleeps", "epsilon", 
"longhorn", "ornament", "entreaty", "atypical", "calendar", "dares", 
"resurgent", "entreat", "eater", "term", "ermine", "inedible", 
"bleeder", "derrières", "resentful", "fulcra", "crabbed", 
"bedevilment", "entwine", "inelegant", "antitoxins", "inspired", 
"redder", "derides", "descendant", "antihistamine", "inequitable", 
"bleat", "eaten", "tenured", "redcap", "capstans", "answerable", 
"blender", "deranges", "gestures", "restart", "arteriosclerosis", 
"sis", "sisal", "saltpeter", "terrifyingly", "glycerin", "rink", 
"inkwell", "ellipsis", "sisterhood", "oodles", "lessor", "sorrowed", 
"wedges", "gesundheit", "either", "hereafter", "termite", "iterator", 
"tornado", "adobes", "bespoken", "ken", "kens", "ensnare", "area", 
"rear", "earful", "fulfil", "fillet", "letdown", "ownership", 
"hipped", "pediatric", "richer", "heretical", "calculus", "lusher", 
"heraldic", "dice", "icebound", "underscored", "redskins", "instant", 
"antiperspirant", "anthropomorphic", "hiccup", "cup", "cups", 
"upstage", "agendas", "dashingly", "glycerol", "role", "oleo", 
"leonine", "ineluctable", "blessed", "sedatives", "vesicles", 
"lessens", "ensured", "redefine", "inextinguishable", "bleach", 
"achoo", "hooch", "ocher", "hero", "erode", "ode", "odes", "desktop", 
"topple", "pleasured", "redeveloped", "pediment", "entrapped", 
"pederasty", "stylus", "lush", "usher", "hermaphrodite", "item", 
"tempos", "postpaid", "aide", "ideogram", "rampart", "artisan", 
"sandhog", "hog", "hogwash", "ash", "ashram", "rammed", "mediocre", 
"crestfallen", "lend", "endow", "downcast", "astronomer", 
"merriment", "entrant", "antiwar", "warden", "dentures", "restful", 
"fulfillment", "entrapment", "enthrall", "allay", "layout", 
"outbound", "underclassman", "manhole", "oleander", "dermis", 
"misused", "sedater", "terrific", "fiche", "cheapens", "ensnares", 
"restrains", "insolent", "entombed", "bedraggle", "gleeful", 
"fulfilment", "entrenchment", "entrap", "rapper", "persistent", 
"enthronement", "enthusiast", "astute", "uterus", "rustproofed", 
"fedora", "orangeades", "despised", "seducer", "ceramic", 
"microscopic", "picnic", "nicotine", "inexpedient", "entomb", 
"ombudsman", "mantel", "teletypewriter", "terminological", "calif", 
"lifetimes", "mescaline", "inertia", "tiaras", "raster", "terrace", 
"acetaminophen", "henchmen", "menhadens", "enslaves", "vesper", 
"peroxide", "ideograph", "aphid", "hides", "desideratum", "tumor", 
"mortgagee", "geegaw", "gawk", "awkward", "ardent", "enthused", 
"sediment", "enter", "termed", "mediaeval", "valentine", "inexact", 
"actives", "vestment", "entourage", "agent", "entryway", "wayside", 
"idea", "dear", "earache", "checkups", "upsides", "descent", 
"entertainment", "entomological", "calicos", "cosign", "ignored", 
"redcoat", "oaten", "tensed", "sedan", "dank", "anklet", "lettered", 
"redskin", "kingpin", "pinups", "ups", "upshot", "hotbed", 
"bedtimes", "mes", "messenger", "germicides", "destitute", "utensil", 
"silencer", "cervix", "vixens", "ensign", "ignorant", "antipasti", 
"stimulus", "lusty", "stymie", "miens", "enslave", "averred", 
"redrew", "rewritten", "tenpins", "instructor", "torrent", 
"entertains", "insult", "ultrasound", "undersides", "despoil", 
"oilcloth", "other", "hereupon", "pondered", "redundant", "anthill", 
"ill", "illicit", "citizens", "ensnared", "rediscovered", "redesign", 
"ignoramus", "muskmelon", "longer", "gerrymander", "deride", "ideas", 
"easy", "asylum", "lumbermen", "mendicant", "antlered", "redevelop", 
"lopes", "pester", "terrapins", "instil", "tildes", "deserves", 
"vesicle", "cleave", "avenger", "germane", "anemia", "miasmas", 
"mash", "ashy", "shy", "shyster", "termagant", "antiaircraft", 
"afterglow", "lowland", "and", "androgen", "genitalia", "liars", 
"arson", "sonatas", "taste", "stepsister", "termini", "initiator", 
"tor", "torn", "ornamental", "tallow", "lowered", "red", "redraft", 
"aft", "aftertaste", "stereotypes", "pesky", "skyrocket", 
"kettledrum", "rummer", "merciful", "fulsome", "omens", "ensures", 
"resultant", "antennas", "nasal", "saleswoman", "mane", "anemometer", 
"terrains", "insightful", "fulcrum", "rumbas", "baseman", 
"mannikins", "insures", "resound", "underpass", "assassins", "inset", 
"settee", "teethe", "theological", "calf", "alfresco", "scornful", 
"fulfill", "illustrator", "torpid", "pidgin", "gins", "instal", 
"talc", "alcove", "overtakes", "kestrel", "relabel", "beleaguered", 
"redraw", "rawhide", "identical", "caliber", "beret", "retrace", 
"acetylene", "enemas", "massacred", "redeploys", "oyster", 
"terminator", "tortillas", "last", "astronomical", "calliope", 
"operator", "tort", "orthographic", "hiccups", "upstart", 
"artificer", "cervical", "callus", "lustre", "trend", "endeavor", 
"vortex", "textures", "researcher", "heroins", "instill", "illegal", 
"galloped", "pedagogical", "callipered", "rediscover", "vertebra", 
"brasher", "herbicides", "descry", "cryptogram", "ramrod", "rodeo", 
"deodorizer", "zeros", "rosebush", "ushered", "redden", "denatures", 
"reset", "setups", "upside", "ides", "describes", "besides", 
"desperado", "adores", "reshuffle", "flea", "leaflet", "lethal", 
"halibut", "but", "button", "tonic", "niche", "cherubim", "bimbos", 
"bosun", "sunk", "unkind", "indentures", "resend", "endures", 
"restorer", "reran", "rang", "anger", "germicide", "ideological", 
"calabash", "ashamed", "medical", "caloric", "rickshas", "hasten", 
"tendon", "donkey", "keyword", "ordains", "insecticides", "desires", 
"resin", "sins", "inspector", "torrid", "rid", "rides", "despot", 
"potpie", "piebald", "aldermen", "menace", "ace", "acerbic", "bicep", 
"cephalic", "lichen", "hennas", "nasty", "styes", "yesterday", "day", 
"daybed", "bedridden", "dental", "talisman", "mankind", "indignant", 
"antique", "questionnaires", "resubmit", "mitten", "tenfold", "old", 
"olden", "denudes", "design", "ignores", "resumes", "mesdames", 
"mesas", "sass", "assemblywoman", "mangle", "glee", "leeway", 
"waylay", "laywomen", "menswear", "ear", "earldom", "domains", "ins", 
"instrumental", "tall", "all", "allegorical", "calm", "almanac", 
"nacre", "credit", "dittos", "tossup", "superman", "mandolin", 
"linesman", "manacle", "cleverer", "rerun", "runaway", "way", 
"wayfarer", "reruns", "unshaven", "ventures", "resell", "elliptical", 
"calmer", "mercuric", "ricochet", "heterodoxy", "oxymora", 
"orangutang", "angina", "inapt", "apt", "aptitudes", "descend", 
"endear", "earlobes", "bestowal", "walleyes", "yes", "yeshivas", 
"vassal", "saltcellar", "larval", "valiant", "anthropological", 
"calfskin", "kind", "inductee", "tee", "teenager", "gerund", 
"underclass", "assemblyman", "manservant", "antelopes", "peso", 
"esoteric", "rickshaw", "hawser", "servicewomen", "mental", 
"tallyhos", "hospital", "talon", "longshoremen", "men", "menthol", 
"holography", "phylum", "lumberman", "manikin", "kingpins", 
"install", "allures", "resuscitator", "tortilla", "llamas", 
"massacres", "resistor", "tormentor", "torque", "queasy", 
"asymmetric", "ricksha", "sharped", "pedlar", "largos", "gossamer", 
"merganser", "service", "icebox", "boxer", "xerography", "physical", 
"calculator", "tortures", "resonant", "anticlimax", "maxima", "imam", 
"mammon", "monograph", "aphelia", "liaison", "sonic", "nicknamed", 
"media", "diametrical", "calliper", "performed", "medulla", "llama", 
"amalgam", "gamins", "insulin", "lineman", "mantra", "transplant", 
"antigen", "genres", "respires", "resold", "oldie", "diesel", 
"seldom", "domed", "medieval", "valor", "lordship", "hipper", "per", 
"perspires", "restores", "restructures", "resort", "orthodoxy", 
"oxygen", "gentlemen", "menopausal", "saltpetre", "treacle", 
"cleaver", "verdigris", "risen", "send", "end", "endemic", 
"microfiche", "checkout", "outclass", "assault", "ultraviolet", 
"let", "letterbox", "boxcar", "carom", "roman", "manifesto", 
"stovepipes", "pesticides", "described", "bedsides", "descant", 
"anthem", "hempen", "penguins", "insignificant", "antebellum", 
"lumbar", "barracudas", "dash", "ashcan", "cannonball", "allover", 
"verbena", "enamor", "morgue", "guerrillas", "lash", "ashen", 
"henchman", "mandolins", "inspires", "resistant", "antechamber", 
"bereave", "aver", "vermin", "minim", "nimbus", "bus", "businessman", 
"mantras", "rasp", "asphalt", "altogether", "her", "hereabout", 
"outcast", "astrological", "calisthenic", "nicknames", "mescal", 
"calliopes", "pesetas", "tassel", "selectman", "mannikin", 
"kinswoman", "man", "manic", "nicer", "cerebra", "bravado", "adobe", 
"obeisant", "antiparticle", "clever", "versus", "sushi", "shirr", 
"irrelevant", "antelope", "open", "pentagon", "gonad", "nadir", 
"directorship", "hippopotami", "amid", "midwifed", "fedoras", 
"rasher", "herbal", "ball", "allot", "lot", "lotus", "tussle", 
"sledgehammer", "merchant", "ant", "antidepressant", "anther", 
"heraldry", "drywall", "allegros", "rosebud", "budgerigar", 
"garbageman", "manikins", "inscribes", "bestow", "townsmen", "menu", 
"enures", "restaurant", "antithetical", "calico", "icon", "confound", 
"underbid", "bidden", "denser", "seraphic", "hiccuped", "pedigree", 
"reeve", "ever", "vertical", "caliper", "perusal", "salami", "amir", 
"mires", "restraint", "interstellar", "larkspur", "puritanical", 
"calligrapher", "herdsman", "manatee", "teepee", "peeve", "everyday", 
"daydreamer", "meres", "result", "ultimatum", "tumbril", "rill", 
"illogical", "calligraphy", "physic", "sickbed", "bedsores", 
"resolver", "vertebras", "rascal", "call", "allergenic", "nickname", 
"amebas", "baste", "stepson", "son", "sonnet", "net", "nether", 
"heros", "rosins", "insular", "larvas", "vast", "astrakhan", 
"handyman", "manicures", "resins", "instep", "tepid", "pidgins", 
"inscribed", "bedbug", "bug", "bugbear", "earwax", "waxen", 
"xenophobia", "biathlon", "longhair", "airstrip", "ripple", "pleas", 
"eastbound", "underachiever", "verbatim", "timbre", "brew", 
"rewound", "underplay", "laywoman", "mandarins", "insofar", "farm", 
"armpit", "pitcher", "herald", "alderman", "mangos", "gossip", 
"sipped", "pedagogue", "guerillas", "laser", "serape", "aped", 
"pederast", "astound", "underground", "underpins", "insane", 
"anemic", "micra", "crane", "anew", "new", "newscast", "astir", 
"tiro", "ironware", "are", "areas", "east", "astronomic", 
"microchip", "hippopotamus", "mustache", "chervil", "villas", "lass", 
"assassin", "sinew", "newsman", "mangrove", "overtax", "taxicab", 
"cabana", "anathemas", "mast", "astronaut", "author", "horoscope", 
"opera", "eraser", "serfdom", "dominos", "nostrum", "rumpus", "pus", 
"pushcart", "arthropod", "podia", "diatom", "tomboy", "boycott", 
"ottoman", "manhunt", "untidy", "idyllic", "licensee", "seethe", 
"thereabout", "outplay", "layoff", "officer", "cerebrum", "rum", 
"rumple", "plethora", "oracle", "clergyman", "maneuver", "verandas", 
"dashikis", "kisser", "serum", "rumor", "morbid", "bidet", "detach", 
"achiever", "vertex", "text", "extremer", "merino", "inopportune", 
"uneaten", "tensor", "sort", "orthopedic", "dickie", "kielbasas", 
"sashay", "hayloft", "often", "ten", "tenpin", "pinkeye", "eyeball", 
"allegro", "grout", "outfox", "fox", "foxtrot", "rot", "rotund", 
"underwear", "earshot", "hot", "hotshot", "hotel", "telex", 
"lexicon", "congresswoman", "manor", "northbound", "undertow", 
"township", "hippos", "possessor", "sorbet", "betcha", "chart", 
"art", "article", "clear", "earwig", "wigwam", "wampum", "pummel", 
"melodic", "dictum", "tumbrel", "relic", "licit", "citadel", "delay", 
"lay", "laypeople", "plectra", "traumas", "mascot", "cotyledon", 
"donor", "nor", "normal", "malt", "altar", "tart", "artiste", 
"stencil", "cilantro", "trouper", "pericardia", "diadem", "democrat", 
"rattan", "tang", "angstrom", "romper", "perturb", "urban", "bang", 
"angel", "gelatin", "tint", "intros", "rostra", "trapper", 
"persimmon", "monsignori", "origin", "ginkgos", "gospel", "pelvis", 
"visor", "sorghum", "humid", "midair", "air", "airfoil", "oil", 
"oilskin", "kin", "kindergarten", "tentacle", "cleanser", "sermon", 
"monolog", "logarithmic", "microbes", "bestir", "tiros", "rosin", 
"sin", "singleton", "tonsil", "silicon", "con", "constraint", 
"intagli", "glint", "interwove", "overshadow", "downtrodden", 
"dentin", "tin", "tinsel", "sellout", "out", "output", "put", 
"putsch", "schoolmarm", "arm", "armor", "moribund", "underpin", 
"pint", "interloper", "periwig", "wig", "wigwag", "wagon", 
"gonorrhea", "hearten", "tenon", "nonverbal", "balsam", "samovar", 
"varmint", "interviewee", "weeper", "perturbed", "bed", "bedpan", 
"panache", "chestnut", "nut", "nutmeg", "meg", "megalopolis", 
"lissom", "somersault", "ultra", "tram", "ramp", "amputee", "teeth", 
"ethos", "hos", "hostel", "telescopic", "picayune", "uneven", 
"vendor", "dorsal", "salad", "ladybug", "bugaboo", "boomerang", 
"angora", "orangutan", "tandem", "demagogry", "gryphon", 
"honeycombed", "bedlam", "lamb", "ambergris", "risky", "sky", 
"skycap", "capstan", "tannin", "ninepin", "pinpoint", "interpret", 
"retiree", "reefer", "fer", "ferret", "returnee", "needlepoint", 
"interurban", "bantam", "tamp", "ampul", "pullout", "outrun", 
"runabout", "outstrip", "rip", "ripen", "pennon", "nonfat", "fathom", 
"homespun", "puns", "unsubscribes", "besom", "sombre", "breathe", 
"theatre", "tremor", "mortar", "tarpaulin", "lintel", "telethon", 
"honeydew", "dewlap", "lap", "lapel", "pelvic", "victim", "timpani", 
"animus", "muscat", "cat", "catsup", "sup", "superstar", "taro", 
"arousal", "salamis", "misprint", "interwoven", "venom", "nomad", 
"madam", "dam", "dampen", "penicillin", "lint", "intercom", 
"compound", "underpay", "pay", "payoff", "off", "offal", "fallout", 
"outwit", "withal", "halt", "altho", "tho", "thou", "housebound", 
"undergrad", "radio", "diocesan", "sanserif", "riffraff", 
"affidavit", "vitamin", "minicam", "campus", "pussycat", "catamaran", 
"rancor", "cornucopia", "piano", "anon", "non", "nonpartisan", 
"sandbar", "bar", "barren", "renewal", "walkout", "outruns", 
"unsnap", "naphtha", "thalamus", "musky", "skydove", "overrun", 
"run", "runs", "unsheathe", "the", "theorem", "remove", "overreach", 
"ache", "cherub", "rubes", "beseech", "echo", "chosen", "sensor", 
"sorrel", "relay", "layman", "mantillas", "lasagna", "gnat", 
"natures", "resonator", "torus", "russet", "set", "setback", 
"acknowledgement", "entanglement", "entombment", "entourages", 
"gestates", "testing", "ingratiate", "ate", "ateliers", "ersatzes", 

Python, 90

upended deductibles lessee seesawed wed wedded deduces cesarians answerable blesses sesames meshes hesitation ionosphere erection ionizer zero erode ode odes deserter terminations onshore ores reside idealistic tic tickles lesson song ongoing ingrains instill illiberal rallying ingest estranges gestures responses sesame amend endives vestment entertainer nervously slyest esters ersatzes zestful fullback acknowledges gesture urethras rashes hesitant anticlimax maxima imagine inebriates tested tediously slyness essences cesareans answer werewolves vest esteemed meddlesome omelet lettuces cession ionization ionizing ingestion ionospheres rescue cueing ingrown ownership hip hippie piety etymologist isthmuses sessions onset settled ledgers

First i clean up the list manually by deleting all

  • words with a Capital letter
  • words with apostrophe's
  • words with éêèáâàö
  • 1- and 2-letter words

this costs me at most 2 points because those words can only be in the beginning or the end of the chain but it reduces the wordlist by 1/3 and i dont have to deal with unicode.

Next I construct a list of all pre- and suffixes, find the overlap and discard all words unless both the pre- and suffix is in the overlap set. Again, this shaves off at most 2 points off my maximal achievable score, but it reduces the wordlist to a third of original size (try running your algorithm on the short_list for a possible speedup) and the remaining words are highly connected (except some 3-letter words which are connected only to themselves). In fact, almost any word can be reached from any other word through a path with an average of 4 edges.

I store the number of links in an adjacency matrix which simplifies all operations and lets me do cool stuff like looking n steps ahead or counting cycles... at least in theory, since it takes around 15 seconds to square the matrix I dont actually do this during the search. Instead i start at a random prefix and walk around randomly, either picking an ending uniformly, favoring those that occur often (like '-ing') or those that occur less often.
All three variants suck equally and produce chains in 20-40 range, but at least its fast. Guess I gotta add recursion after all.

from numpy import *
f = open('words_short.txt')
words = f.read().split() # 62896

prefix = [w[:3] for w in words]     # 2292
suffix = [w[-3:] for w in words]    # 2262
common = set(prefix) & set(suffix)  # 1265

PSW = [(p,s,w) for (p,s,w) in zip(prefix, suffix, words) if p in common and s in common] # 28673
common = list(common)
mapping = dict(zip(common, range(len(common)))) # enumerate trigrams

M = zeros((len(common), len(common)), dtype=int) # for fast processing
W = [[[] for i in range(len(common))] for j in range(len(common))] # for reconstruction
for p,s,w in PSW: # build adjacency matrix
    M[mapping[p], mapping[s]] += 1

def chain(A, rho=0):
    B = array(A)
    links = []
    start = random.randint(len(B))
    while 1:
        nextpos = where(B[links[-1],:]>0)[0]
        if len(nextpos)==0: return links
        nextnum = B[links[-1],nextpos]

        p = ones(len(nextnum))/len(nextnum) # pick uniformly
        if rho>0: p = nextnum*1./sum(nextnum) # prioritize many links
        if rho>1: p = 1/p; p = p/sum(p) # prioritize few links

        chosen = random.choice(nextpos, p=p)
        B[links[-1], chosen] -= 1

def chain2words(L):
    # can only be used once because of .pop()
    z = zip(L[:-1],L[1:])
    res = []
    for p,s in z:
    return res

chains = [chain(M) for i in range(100)]
bestchain = chains[argmax(map(len, chains))]
print ' '.join(chain2words(bestchain))

Originally I wanted to try something similar to this but since this is a directed graph with cycles, none of the standard algorithms for Topological Sorting, Longest Path, Largest Eulerian Path or Chinese Postman Problem work without heavy modifications.

And just because it looks nice, here is a picture of the adjacency matrix M, M^2 and M^infinity (infinity = 32, it doesnt change afterwards) with white = nonzero entries
enter image description here

  • \$\begingroup\$ So your score is 90 ? While we already have 1700+ entries .. Anything I am missing ? \$\endgroup\$
    – Optimizer
    Jan 28 '15 at 10:56
  • 1
    \$\begingroup\$ First of all I'm still working on it, but apart from that - It seemed like a good idea, I tried it, it failed. If anything, this will stop people from wasting time using the same approach \$\endgroup\$
    – DenDenDo
    Jan 28 '15 at 11:16
  • \$\begingroup\$ Heh :) Keep up the positive attitude :) Hope to see this get better results. \$\endgroup\$
    – Optimizer
    Jan 28 '15 at 11:18
  • 2
    \$\begingroup\$ "those words can only be in the beginning or the end of the chain" is incorrect. The largest connected component in the graph includes words such as boutonnières which have an accented character, but not in the prefix or suffix. It only affects a dozen or so words, but one of them might be a key link. \$\endgroup\$ Jan 28 '15 at 17:06

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.