path: root/linguistics/src/main/java/com/yahoo/language/simple/kstem/KStemmer.java

// Copyright Vespa.ai. Licensed under the terms of the Apache 2.0 license. See LICENSE in the project root.
/*
 * This algorithm is adapted from the Lucene code base which is Copyright 2008 Apache Software Foundation and Licensed
 * under the terms of the Apache License, Version 2.0, which was adapted from
 * the kstemmer code base which is Copyright 2003, CIIR University of Massachusetts
 * Amherst (http://ciir.cs.umass.edu) and Licensed under the terms of a modified old-style BSD license.
 */
package com.yahoo.language.simple.kstem;

/**
 * A stemmer implementing the Kstem algorithm by Bob Krovetz.
 */
public class KStemmer {

  static private final int MaxWordLen = 50;
  
  static private final String[] exceptionWords = {"aide", "bathe", "caste",
      "cute", "dame", "dime", "doge", "done", "dune", "envelope", "gage",
      "grille", "grippe", "lobe", "mane", "mare", "nape", "node", "pane",
      "pate", "plane", "pope", "programme", "quite", "ripe", "rote", "rune",
      "sage", "severe", "shoppe", "sine", "slime", "snipe", "steppe", "suite",
      "swinge", "tare", "tine", "tope", "tripe", "twine"};
  
  static private final String[][] directConflations = { {"aging", "age"},
      {"going", "go"}, {"goes", "go"}, {"lying", "lie"}, {"using", "use"},
      {"owing", "owe"}, {"suing", "sue"}, {"dying", "die"}, {"tying", "tie"},
      {"vying", "vie"}, {"aged", "age"}, {"used", "use"}, {"vied", "vie"},
      {"cued", "cue"}, {"died", "die"}, {"eyed", "eye"}, {"hued", "hue"},
      {"iced", "ice"}, {"lied", "lie"}, {"owed", "owe"}, {"sued", "sue"},
      {"toed", "toe"}, {"tied", "tie"}, {"does", "do"}, {"doing", "do"},
      {"aeronautical", "aeronautics"}, {"mathematical", "mathematics"},
      {"political", "politics"}, {"metaphysical", "metaphysics"},
      {"cylindrical", "cylinder"}, {"nazism", "nazi"},
      {"ambiguity", "ambiguous"}, {"barbarity", "barbarous"},
      {"credulity", "credulous"}, {"generosity", "generous"},
      {"spontaneity", "spontaneous"}, {"unanimity", "unanimous"},
      {"voracity", "voracious"}, {"fled", "flee"}, {"miscarriage", "miscarry"}};
  
  static private final String[][] countryNationality = {
      {"afghan", "afghanistan"}, {"african", "africa"},
      {"albanian", "albania"}, {"algerian", "algeria"},
      {"american", "america"}, {"andorran", "andorra"}, {"angolan", "angola"},
      {"arabian", "arabia"}, {"argentine", "argentina"},
      {"armenian", "armenia"}, {"asian", "asia"}, {"australian", "australia"},
      {"austrian", "austria"}, {"azerbaijani", "azerbaijan"},
      {"azeri", "azerbaijan"}, {"bangladeshi", "bangladesh"},
      {"belgian", "belgium"}, {"bermudan", "bermuda"}, {"bolivian", "bolivia"},
      {"bosnian", "bosnia"}, {"botswanan", "botswana"},
      {"brazilian", "brazil"}, {"british", "britain"},
      {"bulgarian", "bulgaria"}, {"burmese", "burma"},
      {"californian", "california"}, {"cambodian", "cambodia"},
      {"canadian", "canada"}, {"chadian", "chad"}, {"chilean", "chile"},
      {"chinese", "china"}, {"colombian", "colombia"}, {"croat", "croatia"},
      {"croatian", "croatia"}, {"cuban", "cuba"}, {"cypriot", "cyprus"},
      {"czechoslovakian", "czechoslovakia"}, {"danish", "denmark"},
      {"egyptian", "egypt"}, {"equadorian", "equador"},
      {"eritrean", "eritrea"}, {"estonian", "estonia"},
      {"ethiopian", "ethiopia"}, {"european", "europe"}, {"fijian", "fiji"},
      {"filipino", "philippines"}, {"finnish", "finland"},
      {"french", "france"}, {"gambian", "gambia"}, {"georgian", "georgia"},
      {"german", "germany"}, {"ghanian", "ghana"}, {"greek", "greece"},
      {"grenadan", "grenada"}, {"guamian", "guam"},
      {"guatemalan", "guatemala"}, {"guinean", "guinea"},
      {"guyanan", "guyana"}, {"haitian", "haiti"}, {"hawaiian", "hawaii"},
      {"holland", "dutch"}, {"honduran", "honduras"}, {"hungarian", "hungary"},
      {"icelandic", "iceland"}, {"indonesian", "indonesia"},
      {"iranian", "iran"}, {"iraqi", "iraq"}, {"iraqui", "iraq"},
      {"irish", "ireland"}, {"israeli", "israel"},
      {"italian", "italy"},
      {"jamaican", "jamaica"},
      {"japanese", "japan"},
      {"jordanian", "jordan"},
      {"kampuchean", "cambodia"},
      {"kenyan", "kenya"},
      {"korean", "korea"},
      {"kuwaiti", "kuwait"},
      {"lankan", "lanka"},
      {"laotian", "laos"},
      {"latvian", "latvia"},
      {"lebanese", "lebanon"},
      {"liberian", "liberia"},
      {"libyan", "libya"},
      {"lithuanian", "lithuania"},
      {"macedonian", "macedonia"},
      {"madagascan", "madagascar"},
      {"malaysian", "malaysia"},
      {"maltese", "malta"},
      {"mauritanian", "mauritania"},
      {"mexican", "mexico"},
      {"micronesian", "micronesia"},
      {"moldovan", "moldova"},
      {"monacan", "monaco"},
      {"mongolian", "mongolia"},
      {"montenegran", "montenegro"},
      {"moroccan", "morocco"},
      {"myanmar", "burma"},
      {"namibian", "namibia"},
      {"nepalese", "nepal"},
      // {"netherlands", "dutch"},
      {"nicaraguan", "nicaragua"}, {"nigerian", "nigeria"},
      {"norwegian", "norway"}, {"omani", "oman"}, {"pakistani", "pakistan"},
      {"panamanian", "panama"}, {"papuan", "papua"},
      {"paraguayan", "paraguay"}, {"peruvian", "peru"},
      {"portuguese", "portugal"}, {"romanian", "romania"},
      {"rumania", "romania"}, {"rumanian", "romania"}, {"russian", "russia"},
      {"rwandan", "rwanda"}, {"samoan", "samoa"}, {"scottish", "scotland"},
      {"serb", "serbia"}, {"serbian", "serbia"}, {"siam", "thailand"},
      {"siamese", "thailand"}, {"slovakia", "slovak"}, {"slovakian", "slovak"},
      {"slovenian", "slovenia"}, {"somali", "somalia"},
      {"somalian", "somalia"}, {"spanish", "spain"}, {"swedish", "sweden"},
      {"swiss", "switzerland"}, {"syrian", "syria"}, {"taiwanese", "taiwan"},
      {"tanzanian", "tanzania"}, {"texan", "texas"}, {"thai", "thailand"},
      {"tunisian", "tunisia"}, {"turkish", "turkey"}, {"ugandan", "uganda"},
      {"ukrainian", "ukraine"}, {"uruguayan", "uruguay"},
      {"uzbek", "uzbekistan"}, {"venezuelan", "venezuela"},
      {"vietnamese", "viet"}, {"virginian", "virginia"}, {"yemeni", "yemen"},
      {"yugoslav", "yugoslavia"}, {"yugoslavian", "yugoslavia"},
      {"zambian", "zambia"}, {"zealander", "zealand"},
      {"zimbabwean", "zimbabwe"}};
  
  static private final String[] supplementDict = {"aids", "applicator",
      "capacitor", "digitize", "electromagnet", "ellipsoid", "exosphere",
      "extensible", "ferromagnet", "graphics", "hydromagnet", "polygraph",
      "toroid", "superconduct", "backscatter", "connectionism"};
  
  static private final String[] properNouns = {"abrams", "achilles",
      "acropolis", "adams", "agnes", "aires", "alexander", "alexis", "alfred",
      "algiers", "alps", "amadeus", "ames", "amos", "andes", "angeles",
      "annapolis", "antilles", "aquarius", "archimedes", "arkansas", "asher",
      "ashly", "athens", "atkins", "atlantis", "avis", "bahamas", "bangor",
      "barbados", "barger", "bering", "brahms", "brandeis", "brussels",
      "bruxelles", "cairns", "camoros", "camus", "carlos", "celts", "chalker",
      "charles", "cheops", "ching", "christmas", "cocos", "collins",
      "columbus", "confucius", "conners", "connolly", "copernicus", "cramer",
      "cyclops", "cygnus", "cyprus", "dallas", "damascus", "daniels", "davies",
      "davis", "decker", "denning", "dennis", "descartes", "dickens", "doris",
      "douglas", "downs", "dreyfus", "dukakis", "dulles", "dumfries",
      "ecclesiastes", "edwards", "emily", "erasmus", "euphrates", "evans",
      "everglades", "fairbanks", "federales", "fisher", "fitzsimmons",
      "fleming", "forbes", "fowler", "france", "francis", "goering",
      "goodling", "goths", "grenadines", "guiness", "hades", "harding",
      "harris", "hastings", "hawkes", "hawking", "hayes", "heights",
      "hercules", "himalayas", "hippocrates", "hobbs", "holmes", "honduras",
      "hopkins", "hughes", "humphreys", "illinois", "indianapolis",
      "inverness", "iris", "iroquois", "irving", "isaacs", "italy", "james",
      "jarvis", "jeffreys", "jesus", "jones", "josephus", "judas", "julius",
      "kansas", "keynes", "kipling", "kiwanis", "lansing", "laos", "leeds",
      "levis", "leviticus", "lewis", "louis", "maccabees", "madras",
      "maimonides", "maldive", "massachusetts", "matthews", "mauritius",
      "memphis", "mercedes", "midas", "mingus", "minneapolis", "mohammed",
      "moines", "morris", "moses", "myers", "myknos", "nablus", "nanjing",
      "nantes", "naples", "neal", "netherlands", "nevis", "nostradamus",
      "oedipus", "olympus", "orleans", "orly", "papas", "paris", "parker",
      "pauling", "peking", "pershing", "peter", "peters", "philippines",
      "phineas", "pisces", "pryor", "pythagoras", "queens", "rabelais",
      "ramses", "reynolds", "rhesus", "rhodes", "richards", "robins",
      "rodgers", "rogers", "rubens", "sagittarius", "seychelles", "socrates",
      "texas", "thames", "thomas", "tiberias", "tunis", "venus", "vilnius",
      "wales", "warner", "wilkins", "williams", "wyoming", "xmas", "yonkers",
      "zeus", "frances", "aarhus", "adonis", "andrews", "angus", "antares",
      "aquinas", "arcturus", "ares", "artemis", "augustus", "ayers",
      "barnabas", "barnes", "becker", "bejing", "biggs", "billings", "boeing",
      "boris", "borroughs", "briggs", "buenos", "calais", "caracas", "cassius",
      "cerberus", "ceres", "cervantes", "chantilly", "chartres", "chester",
      "connally", "conner", "coors", "cummings", "curtis", "daedalus",
      "dionysus", "dobbs", "dolores", "edmonds"};
  
  static class DictEntry {
    boolean exception;
    String root;
    
    DictEntry(String root, boolean isException) {
      this.root = root;
      this.exception = isException;
    }
  }
  
  private static final CharArrayMap<DictEntry> dict_ht = initializeDictHash();


  private final OpenStringBuilder word = new OpenStringBuilder();
  private int j; /* index of final letter in stem (within word) */
  private int k; /*
                  * INDEX of final letter in word. You must add 1 to k to get
                  * the current length of word. When you want the length of
                  * word, use the method wordLength, which returns (k+1).
                  */
  
  /*
   * private void initializeStemHash() { if (maxCacheSize > 0) cache = new
   * CharArrayMap<String>(maxCacheSize,false); }
   ***/
  
  private char finalChar() {
    return word.charAt(k);
  }
  
  private char penultChar() {
    return word.charAt(k - 1);
  }
  
  private boolean isVowel(int index) {
    return !isCons(index);
  }
  
  private boolean isCons(int index) {
    char ch;
    
    ch = word.charAt(index);
    
    if ((ch == 'a') || (ch == 'e') || (ch == 'i') || (ch == 'o') || (ch == 'u')) return false;
    if ((ch != 'y') || (index == 0)) return true;
    else return (!isCons(index - 1));
  }
  
  private static CharArrayMap<DictEntry> initializeDictHash() {
    DictEntry defaultEntry;
    DictEntry entry;

    CharArrayMap<DictEntry> d = new CharArrayMap<>(1000, false);
    for (int i = 0; i < exceptionWords.length; i++) {
      if (!d.containsKey(exceptionWords[i])) {
        entry = new DictEntry(exceptionWords[i], true);
        d.put(exceptionWords[i], entry);
      } else {
        throw new RuntimeException("Warning: Entry [" + exceptionWords[i]
            + "] already in dictionary 1");
      }
    }
    
    for (int i = 0; i < directConflations.length; i++) {
      if (!d.containsKey(directConflations[i][0])) {
        entry = new DictEntry(directConflations[i][1], false);
        d.put(directConflations[i][0], entry);
      } else {
        throw new RuntimeException("Warning: Entry [" + directConflations[i][0]
            + "] already in dictionary 2");
      }
    }
    
    for (int i = 0; i < countryNationality.length; i++) {
      if (!d.containsKey(countryNationality[i][0])) {
        entry = new DictEntry(countryNationality[i][1], false);
        d.put(countryNationality[i][0], entry);
      } else {
        throw new RuntimeException("Warning: Entry [" + countryNationality[i][0]
            + "] already in dictionary 3");
      }
    }
    
    defaultEntry = new DictEntry(null, false);
    
    String[] array;
    array = KStemData1.data;
    
    for (int i = 0; i < array.length; i++) {
      if (!d.containsKey(array[i])) {
        d.put(array[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + array[i]
            + "] already in dictionary 4");
      }
    }
    
    array = KStemData2.data;
    for (int i = 0; i < array.length; i++) {
      if (!d.containsKey(array[i])) {
        d.put(array[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + array[i]
            + "] already in dictionary 4");
      }
    }
    
    array = KStemData3.data;
    for (int i = 0; i < array.length; i++) {
      if (!d.containsKey(array[i])) {
        d.put(array[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + array[i]
            + "] already in dictionary 4");
      }
    }
    
    array = KStemData4.data;
    for (int i = 0; i < array.length; i++) {
      if (!d.containsKey(array[i])) {
        d.put(array[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + array[i]
            + "] already in dictionary 4");
      }
    }
    
    array = KStemData5.data;
    for (int i = 0; i < array.length; i++) {
      if (!d.containsKey(array[i])) {
        d.put(array[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + array[i]
            + "] already in dictionary 4");
      }
    }
    
    array = KStemData6.data;
    for (int i = 0; i < array.length; i++) {
      if (!d.containsKey(array[i])) {
        d.put(array[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + array[i]
            + "] already in dictionary 4");
      }
    }
    
    array = KStemData7.data;
    for (int i = 0; i < array.length; i++) {
      if (!d.containsKey(array[i])) {
        d.put(array[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + array[i]
            + "] already in dictionary 4");
      }
    }
    
    for (int i = 0; i < KStemData8.data.length; i++) {
      if (!d.containsKey(KStemData8.data[i])) {
        d.put(KStemData8.data[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + KStemData8.data[i]
            + "] already in dictionary 4");
      }
    }
    
    for (int i = 0; i < supplementDict.length; i++) {
      if (!d.containsKey(supplementDict[i])) {
        d.put(supplementDict[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + supplementDict[i]
            + "] already in dictionary 5");
      }
    }
    
    for (int i = 0; i < properNouns.length; i++) {
      if (!d.containsKey(properNouns[i])) {
        d.put(properNouns[i], defaultEntry);
      } else {
        throw new RuntimeException("Warning: Entry [" + properNouns[i]
            + "] already in dictionary 6");
      }
    }
    
    return d;
  }
  
  private boolean isAlpha(char ch) {
    return ch >= 'a' && ch <= 'z'; // terms must be lowercased already
  }
  
  /* length of stem within word */
  private int stemLength() {
    return j + 1;
  };
  
  private boolean endsIn(char[] s) {
    if (s.length > k) return false;
    
    int r = word.length() - s.length; /* length of word before this suffix */
    j = k;
    for (int r1 = r, i = 0; i < s.length; i++, r1++) {
      if (s[i] != word.charAt(r1)) return false;
    }
    j = r - 1; /* index of the character BEFORE the posfix */
    return true;
  }
  
  private boolean endsIn(char a, char b) {
    if (2 > k) return false;
    // check left to right since the endings have often already matched
    if (word.charAt(k - 1) == a && word.charAt(k) == b) {
      j = k - 2;
      return true;
    }
    return false;
  }
  
  private boolean endsIn(char a, char b, char c) {
    if (3 > k) return false;
    if (word.charAt(k - 2) == a && word.charAt(k - 1) == b
        && word.charAt(k) == c) {
      j = k - 3;
      return true;
    }
    return false;
  }
  
  private boolean endsIn(char a, char b, char c, char d) {
    if (4 > k) return false;
    if (word.charAt(k - 3) == a && word.charAt(k - 2) == b
        && word.charAt(k - 1) == c && word.charAt(k) == d) {
      j = k - 4;
      return true;
    }
    return false;
  }
  
  private DictEntry wordInDict() {
    /***
     * if (matchedEntry != null) { if (dict_ht.get(word.getArray(), 0,
     * word.size()) != matchedEntry) {
     * System.out.println("Uh oh... cached entry doesn't match"); } return
     * matchedEntry; }
     ***/
    if (matchedEntry != null) return matchedEntry;
    DictEntry e = dict_ht.get(word.getArray(), 0, word.length());
    if (e != null && !e.exception) {
      matchedEntry = e; // only cache if it's not an exception.
    }
    // lookups.add(word.toString());
    return e;
  }
  
  /* Convert plurals to singular form, and '-ies' to 'y' */
  private void plural() {
    if (word.charAt(k) == 's') {
      if (endsIn('i', 'e', 's')) {
        word.setLength(j + 3);
        k--;
        if (lookup()) /* ensure calories -> calorie */
        return;
        k++;
        word.unsafeWrite('s');
        setSuffix("y");
        lookup();
      } else if (endsIn('e', 's')) {
        /* try just removing the "s" */
        word.setLength(j + 2);
        k--;
        
        /*
         * note: don't check for exceptions here. So, `aides' -> `aide', but
         * `aided' -> `aid'. The exception for double s is used to prevent
         * crosses -> crosse. This is actually correct if crosses is a plural
         * noun (a type of racket used in lacrosse), but the verb is much more
         * common
         */

        /****
         * YCS: this was the one place where lookup was not followed by return.
         * So restructure it. if ((j>0)&&(lookup(word.toString())) &&
         * !((word.charAt(j) == 's') && (word.charAt(j-1) == 's'))) return;
         *****/
        boolean tryE = j > 0
            && !((word.charAt(j) == 's') && (word.charAt(j - 1) == 's'));
        if (tryE && lookup()) return;
        
        /* try removing the "es" */

        word.setLength(j + 1);
        k--;
        if (lookup()) return;
        
        /* the default is to retain the "e" */
        word.unsafeWrite('e');
        k++;
        
        if (!tryE) lookup(); // if we didn't try the "e" ending before
        return;
      } else {
        if (word.length() > 3 && penultChar() != 's' && !endsIn('o', 'u', 's')) {
          /* unless the word ends in "ous" or a double "s", remove the final "s" */

          word.setLength(k);
          k--;
          lookup();
        }
      }
    }
  }
  
  private void setSuffix(String s) {
    setSuff(s, s.length());
  }
  
  /* replace old suffix with s */
  private void setSuff(String s, int len) {
    word.setLength(j + 1);
    for (int l = 0; l < len; l++) {
      word.unsafeWrite(s.charAt(l));
    }
    k = j + len;
  }
  
  /* Returns true if the word is found in the dictionary */
  // almost all uses of lookup() return immediately and are
  // followed by another lookup in the dict. Store the match
  // to avoid this double lookup.
  DictEntry matchedEntry = null;
  
  private boolean lookup() {
    matchedEntry = dict_ht.get(word.getArray(), 0, word.size());
    return matchedEntry != null;
  }
  
  // Set<String> lookups = new HashSet<>();
  
  /* convert past tense (-ed) to present, and `-ied' to `y' */
  private void pastTense() {
    /*
     * Handle words less than 5 letters with a direct mapping This prevents
     * (fled -> fl).
     */
    if (word.length() <= 4) return;
    
    if (endsIn('i', 'e', 'd')) {
      word.setLength(j + 3);
      k--;
      if (lookup()) /* we almost always want to convert -ied to -y, but */
      return; /* this isn't true for short words (died->die) */
      k++; /* I don't know any long words that this applies to, */
      word.unsafeWrite('d'); /* but just in case... */
      setSuffix("y");
      lookup();
      return;
    }
    
    /* the vowelInStem() is necessary so we don't stem acronyms */
    if (endsIn('e', 'd') && vowelInStem()) {
      /* see if the root ends in `e' */
      word.setLength(j + 2);
      k = j + 1;
      
      DictEntry entry = wordInDict();
      if (entry != null) if (!entry.exception) /*
                                                * if it's in the dictionary and
                                                * not an exception
                                                */
      return;
      
      /* try removing the "ed" */
      word.setLength(j + 1);
      k = j;
      if (lookup()) return;
      
      /*
       * try removing a doubled consonant. if the root isn't found in the
       * dictionary, the default is to leave it doubled. This will correctly
       * capture `backfilled' -> `backfill' instead of `backfill' ->
       * `backfille', and seems correct most of the time
       */

      if (doubleC(k)) {
        word.setLength(k);
        k--;
        if (lookup()) return;
        word.unsafeWrite(word.charAt(k));
        k++;
        lookup();
        return;
      }
      
      /* if we have a `un-' prefix, then leave the word alone */
      /* (this will sometimes screw up with `under-', but we */
      /* will take care of that later) */

      if ((word.charAt(0) == 'u') && (word.charAt(1) == 'n')) {
        word.unsafeWrite('e');
        word.unsafeWrite('d');
        k = k + 2;
        // nolookup()
        return;
      }
      
      /*
       * it wasn't found by just removing the `d' or the `ed', so prefer to end
       * with an `e' (e.g., `microcoded' -> `microcode').
       */

      word.setLength(j + 1);
      word.unsafeWrite('e');
      k = j + 1;
      // nolookup() - we already tried the "e" ending
      return;
    }
  }
  
  /* return TRUE if word ends with a double consonant */
  private boolean doubleC(int i) {
    if (i < 1) return false;
    
    if (word.charAt(i) != word.charAt(i - 1)) return false;
    return (isCons(i));
  }
  
  private boolean vowelInStem() {
    for (int i = 0; i < stemLength(); i++) {
      if (isVowel(i)) return true;
    }
    return false;
  }
  
  /* handle `-ing' endings */
  private void aspect() {
    /*
     * handle short words (aging -> age) via a direct mapping. This prevents
     * (thing -> the) in the version of this routine that ignores inflectional
     * variants that are mentioned in the dictionary (when the root is also
     * present)
     */

    if (word.length() <= 5) return;
    
    /* the vowelinstem() is necessary so we don't stem acronyms */
    if (endsIn('i', 'n', 'g') && vowelInStem()) {
      
      /* try adding an `e' to the stem and check against the dictionary */
      word.setCharAt(j + 1, 'e');
      word.setLength(j + 2);
      k = j + 1;
      
      DictEntry entry = wordInDict();
      if (entry != null) {
        if (!entry.exception) /* if it's in the dictionary and not an exception */
        return;
      }
      
      /* adding on the `e' didn't work, so remove it */
      word.setLength(k);
      k--; /* note that `ing' has also been removed */
      
      if (lookup()) return;
      
      /* if I can remove a doubled consonant and get a word, then do so */
      if (doubleC(k)) {
        k--;
        word.setLength(k + 1);
        if (lookup()) return;
        word.unsafeWrite(word.charAt(k)); /* restore the doubled consonant */
        
        /* the default is to leave the consonant doubled */
        /* (e.g.,`fingerspelling' -> `fingerspell'). Unfortunately */
        /* `bookselling' -> `booksell' and `mislabelling' -> `mislabell'). */
        /* Without making the algorithm significantly more complicated, this */
        /* is the best I can do */
        k++;
        lookup();
        return;
      }
      
      /*
       * the word wasn't in the dictionary after removing the stem, and then
       * checking with and without a final `e'. The default is to add an `e'
       * unless the word ends in two consonants, so `microcoding' ->
       * `microcode'. The two consonants restriction wouldn't normally be
       * necessary, but is needed because we don't try to deal with prefixes and
       * compounds, and most of the time it is correct (e.g., footstamping ->
       * footstamp, not footstampe; however, decoupled -> decoupl). We can
       * prevent almost all of the incorrect stems if we try to do some prefix
       * analysis first
       */

      if ((j > 0) && isCons(j) && isCons(j - 1)) {
        k = j;
        word.setLength(k + 1);
        // nolookup() because we already did according to the comment
        return;
      }
      
      word.setLength(j + 1);
      word.unsafeWrite('e');
      k = j + 1;
      // nolookup(); we already tried an 'e' ending
      return;
    }
  }
  
  /*
   * this routine deals with -ity endings. It accepts -ability, -ibility, and
   * -ality, even without checking the dictionary because they are so
   * productive. The first two are mapped to -ble, and the -ity is remove for
   * the latter
   */
  private void ityEndings() {
    int old_k = k;
    
    if (endsIn('i', 't', 'y')) {
      word.setLength(j + 1); /* try just removing -ity */
      k = j;
      if (lookup()) return;
      word.unsafeWrite('e'); /* try removing -ity and adding -e */
      k = j + 1;
      if (lookup()) return;
      word.setCharAt(j + 1, 'i');
      word.append("ty");
      k = old_k;
      /*
       * the -ability and -ibility endings are highly productive, so just accept
       * them
       */
      if ((j > 0) && (word.charAt(j - 1) == 'i') && (word.charAt(j) == 'l')) {
        word.setLength(j - 1);
        word.append("le"); /* convert to -ble */
        k = j;
        lookup();
        return;
      }
      
      /* ditto for -ivity */
      if ((j > 0) && (word.charAt(j - 1) == 'i') && (word.charAt(j) == 'v')) {
        word.setLength(j + 1);
        word.unsafeWrite('e'); /* convert to -ive */
        k = j + 1;
        lookup();
        return;
      }
      /* ditto for -ality */
      if ((j > 0) && (word.charAt(j - 1) == 'a') && (word.charAt(j) == 'l')) {
        word.setLength(j + 1);
        k = j;
        lookup();
        return;
      }
      
      /*
       * if the root isn't in the dictionary, and the variant *is* there, then
       * use the variant. This allows `immunity'->`immune', but prevents
       * `capacity'->`capac'. If neither the variant nor the root form are in
       * the dictionary, then remove the ending as a default
       */

      if (lookup()) return;
      
      /* the default is to remove -ity altogether */
      word.setLength(j + 1);
      k = j;
      // nolookup(), we already did it.
      return;
    }
  }
  
  /* handle -ence and -ance */
  private void nceEndings() {
    int old_k = k;
    char word_char;
    
    if (endsIn('n', 'c', 'e')) {
      word_char = word.charAt(j);
      if (!((word_char == 'e') || (word_char == 'a'))) return;
      word.setLength(j);
      word.unsafeWrite('e'); /* try converting -e/ance to -e (adherance/adhere) */
      k = j;
      if (lookup()) return;
      word.setLength(j); /*
                          * try removing -e/ance altogether
                          * (disappearance/disappear)
                          */
      k = j - 1;
      if (lookup()) return;
      word.unsafeWrite(word_char); /* restore the original ending */
      word.append("nce");
      k = old_k;
      // nolookup() because we restored the original ending
    }
    return;
  }
  
  /* handle -ness */
  private void nessEndings() {
    if (endsIn('n', 'e', 's', 's')) { /*
                                       * this is a very productive endings, so
                                       * just accept it
                                       */
      word.setLength(j + 1);
      k = j;
      if (word.charAt(j) == 'i') word.setCharAt(j, 'y');
      lookup();
    }
    return;
  }
  
  /* handle -ism */
  private void ismEndings() {
    if (endsIn('i', 's', 'm')) { /*
                                  * this is a very productive ending, so just
                                  * accept it
                                  */
      word.setLength(j + 1);
      k = j;
      lookup();
    }
    return;
  }
  
  /* this routine deals with -ment endings. */
  private void mentEndings() {
    int old_k = k;
    
    if (endsIn('m', 'e', 'n', 't')) {
      word.setLength(j + 1);
      k = j;
      if (lookup()) return;
      word.append("ment");
      k = old_k;
      // nolookup
    }
    return;
  }
  
  /* this routine deals with -ize endings. */
  private void izeEndings() {
    int old_k = k;
    
    if (endsIn('i', 'z', 'e')) {
      word.setLength(j + 1); /* try removing -ize entirely */
      k = j;
      if (lookup()) return;
      word.unsafeWrite('i');
      
      if (doubleC(j)) { /* allow for a doubled consonant */
        word.setLength(j);
        k = j - 1;
        if (lookup()) return;
        word.unsafeWrite(word.charAt(j - 1));
      }
      
      word.setLength(j + 1);
      word.unsafeWrite('e'); /* try removing -ize and adding -e */
      k = j + 1;
      if (lookup()) return;
      word.setLength(j + 1);
      word.append("ize");
      k = old_k;
      // nolookup()
    }
    return;
  }
  
  /* handle -ency and -ancy */
  private void ncyEndings() {
    if (endsIn('n', 'c', 'y')) {
      if (!((word.charAt(j) == 'e') || (word.charAt(j) == 'a'))) return;
      word.setCharAt(j + 2, 't'); /* try converting -ncy to -nt */
      word.setLength(j + 3);
      k = j + 2;
      
      if (lookup()) return;
      
      word.setCharAt(j + 2, 'c'); /* the default is to convert it to -nce */
      word.unsafeWrite('e');
      k = j + 3;
      lookup();
    }
    return;
  }
  
  /* handle -able and -ible */
  private void bleEndings() {
    int old_k = k;
    char word_char;
    
    if (endsIn('b', 'l', 'e')) {
      if (!((word.charAt(j) == 'a') || (word.charAt(j) == 'i'))) return;
      word_char = word.charAt(j);
      word.setLength(j); /* try just removing the ending */
      k = j - 1;
      if (lookup()) return;
      if (doubleC(k)) { /* allow for a doubled consonant */
        word.setLength(k);
        k--;
        if (lookup()) return;
        k++;
        word.unsafeWrite(word.charAt(k - 1));
      }
      word.setLength(j);
      word.unsafeWrite('e'); /* try removing -a/ible and adding -e */
      k = j;
      if (lookup()) return;
      word.setLength(j);
      word.append("ate"); /* try removing -able and adding -ate */
      /* (e.g., compensable/compensate) */
      k = j + 2;
      if (lookup()) return;
      word.setLength(j);
      word.unsafeWrite(word_char); /* restore the original values */
      word.append("ble");
      k = old_k;
      // nolookup()
    }
    return;
  }
  
  /*
   * handle -ic endings. This is fairly straightforward, but this is also the
   * only place we try *expanding* an ending, -ic -> -ical. This is to handle
   * cases like `canonic' -> `canonical'
   */
  private void icEndings() {
    if (endsIn('i', 'c')) {
      word.setLength(j + 3);
      word.append("al"); /* try converting -ic to -ical */
      k = j + 4;
      if (lookup()) return;
      
      word.setCharAt(j + 1, 'y'); /* try converting -ic to -y */
      word.setLength(j + 2);
      k = j + 1;
      if (lookup()) return;
      
      word.setCharAt(j + 1, 'e'); /* try converting -ic to -e */
      if (lookup()) return;
      
      word.setLength(j + 1); /* try removing -ic altogether */
      k = j;
      if (lookup()) return;
      word.append("ic"); /* restore the original ending */
      k = j + 2;
      // nolookup()
    }
    return;
  }
  
  private static char[] ization = "ization".toCharArray();
  private static char[] ition = "ition".toCharArray();
  private static char[] ation = "ation".toCharArray();
  private static char[] ication = "ication".toCharArray();
  
  /* handle some derivational endings */
  /*
   * this routine deals with -ion, -ition, -ation, -ization, and -ication. The
   * -ization ending is always converted to -ize
   */
  private void ionEndings() {
    int old_k = k;
    if (!endsIn('i', 'o', 'n')) {
      return;
    }
    
    if (endsIn(ization)) { /*
                            * the -ize ending is very productive, so simply
                            * accept it as the root
                            */
      word.setLength(j + 3);
      word.unsafeWrite('e');
      k = j + 3;
      lookup();
      return;
    }
    
    if (endsIn(ition)) {
      word.setLength(j + 1);
      word.unsafeWrite('e');
      k = j + 1;
      if (lookup()) /*
                     * remove -ition and add `e', and check against the
                     * dictionary
                     */
      return; /* (e.g., definition->define, opposition->oppose) */
      
      /* restore original values */
      word.setLength(j + 1);
      word.append("ition");
      k = old_k;
      // nolookup()
    } else if (endsIn(ation)) {
      word.setLength(j + 3);
      word.unsafeWrite('e');
      k = j + 3;
      if (lookup()) /* remove -ion and add `e', and check against the dictionary */
      return; /* (elmination -> eliminate) */
      
      word.setLength(j + 1);
      word.unsafeWrite('e'); /*
                              * remove -ation and add `e', and check against the
                              * dictionary
                              */
      k = j + 1;
      if (lookup()) return;
      
      word.setLength(j + 1);/*
                             * just remove -ation (resignation->resign) and
                             * check dictionary
                             */
      k = j;
      if (lookup()) return;
      
      /* restore original values */
      word.setLength(j + 1);
      word.append("ation");
      k = old_k;
      // nolookup()
      
    }
    
    /*
     * test -ication after -ation is attempted (e.g., `complication->complicate'
     * rather than `complication->comply')
     */

    if (endsIn(ication)) {
      word.setLength(j + 1);
      word.unsafeWrite('y');
      k = j + 1;
      if (lookup()) /*
                     * remove -ication and add `y', and check against the
                     * dictionary
                     */
      return; /* (e.g., amplification -> amplify) */
      
      /* restore original values */
      word.setLength(j + 1);
      word.append("ication");
      k = old_k;
      // nolookup()
    }
    
    // if (endsIn(ion)) {
    if (true) { // we checked for this earlier... just need to set "j"
      j = k - 3; // YCS
      
      word.setLength(j + 1);
      word.unsafeWrite('e');
      k = j + 1;
      if (lookup()) /* remove -ion and add `e', and check against the dictionary */
      return;
      
      word.setLength(j + 1);
      k = j;
      if (lookup()) /* remove -ion, and if it's found, treat that as the root */
      return;
      
      /* restore original values */
      word.setLength(j + 1);
      word.append("ion");
      k = old_k;
      // nolookup()
    }
    
    // nolookup(); all of the other paths restored original values
    return;
  }
  
  /*
   * this routine deals with -er, -or, -ier, and -eer. The -izer ending is
   * always converted to -ize
   */
  private void erAndOrEndings() {
    int old_k = k;
    
    if (word.charAt(k) != 'r') return; // YCS
    
    char word_char; /* so we can remember if it was -er or -or */
    
    if (endsIn('i', 'z', 'e', 'r')) { /*
                                       * -ize is very productive, so accept it
                                       * as the root
                                       */
      word.setLength(j + 4);
      k = j + 3;
      lookup();
      return;
    }
    
    if (endsIn('e', 'r') || endsIn('o', 'r')) {
      word_char = word.charAt(j + 1);
      if (doubleC(j)) {
        word.setLength(j);
        k = j - 1;
        if (lookup()) return;
        word.unsafeWrite(word.charAt(j - 1)); /* restore the doubled consonant */
      }
      
      if (word.charAt(j) == 'i') { /* do we have a -ier ending? */
        word.setCharAt(j, 'y');
        word.setLength(j + 1);
        k = j;
        if (lookup()) /* yes, so check against the dictionary */
        return;
        word.setCharAt(j, 'i'); /* restore the endings */
        word.unsafeWrite('e');
      }
      
      if (word.charAt(j) == 'e') { /* handle -eer */
        word.setLength(j);
        k = j - 1;
        if (lookup()) return;
        word.unsafeWrite('e');
      }
      
      word.setLength(j + 2); /* remove the -r ending */
      k = j + 1;
      if (lookup()) return;
      word.setLength(j + 1); /* try removing -er/-or */
      k = j;
      if (lookup()) return;
      word.unsafeWrite('e'); /* try removing -or and adding -e */
      k = j + 1;
      if (lookup()) return;
      word.setLength(j + 1);
      word.unsafeWrite(word_char);
      word.unsafeWrite('r'); /* restore the word to the way it was */
      k = old_k;
      // nolookup()
    }
    
  }
  
  /*
   * this routine deals with -ly endings. The -ally ending is always converted
   * to -al Sometimes this will temporarily leave us with a non-word (e.g.,
   * heuristically maps to heuristical), but then the -al is removed in the next
   * step.
   */
  private void lyEndings() {
    int old_k = k;
    
    if (endsIn('l', 'y')) {
      
      word.setCharAt(j + 2, 'e'); /* try converting -ly to -le */
      
      if (lookup()) return;
      word.setCharAt(j + 2, 'y');
      
      word.setLength(j + 1); /* try just removing the -ly */
      k = j;
      
      if (lookup()) return;
      
      if ((j > 0) && (word.charAt(j - 1) == 'a') && (word.charAt(j) == 'l')) /*
                                                                              * always
                                                                              * convert
                                                                              * -
                                                                              * ally
                                                                              * to
                                                                              * -
                                                                              * al
                                                                              */
      return;
      word.append("ly");
      k = old_k;
      
      if ((j > 0) && (word.charAt(j - 1) == 'a') && (word.charAt(j) == 'b')) { // always convert 'ably' to 'able'
        word.setCharAt(j + 2, 'e');
        k = j + 2;
        return;
      }
      
      if (word.charAt(j) == 'i') { /* e.g., militarily -> military */
        word.setLength(j);
        word.unsafeWrite('y');
        k = j;
        if (lookup()) return;
        word.setLength(j);
        word.append("ily");
        k = old_k;
      }
      
      word.setLength(j + 1); /* the default is to remove -ly */
      
      k = j;
      // nolookup()... we already tried removing the "ly" variant
    }
    return;
  }
  
  /*
   * this routine deals with -al endings. Some of the endings from the previous
   * routine are finished up here.
   */
  private void alEndings() {
    int old_k = k;
    
    if (word.length() < 4) return;
    if (endsIn('a', 'l')) {
      word.setLength(j + 1);
      k = j;
      if (lookup()) /* try just removing the -al */
      return;
      
      if (doubleC(j)) { /* allow for a doubled consonant */
        word.setLength(j);
        k = j - 1;
        if (lookup()) return;
        word.unsafeWrite(word.charAt(j - 1));
      }
      
      word.setLength(j + 1);
      word.unsafeWrite('e'); /* try removing the -al and adding -e */
      k = j + 1;
      if (lookup()) return;
      
      word.setLength(j + 1);
      word.append("um"); /* try converting -al to -um */
      /* (e.g., optimal - > optimum ) */
      k = j + 2;
      if (lookup()) return;
      
      word.setLength(j + 1);
      word.append("al"); /* restore the ending to the way it was */
      k = old_k;
      
      if ((j > 0) && (word.charAt(j - 1) == 'i') && (word.charAt(j) == 'c')) {
        word.setLength(j - 1); /* try removing -ical */
        k = j - 2;
        if (lookup()) return;
        
        word.setLength(j - 1);
        word.unsafeWrite('y');/* try turning -ical to -y (e.g., bibliographical) */
        k = j - 1;
        if (lookup()) return;
        
        word.setLength(j - 1);
        word.append("ic"); /* the default is to convert -ical to -ic */
        k = j;
        // nolookup() ... converting ical to ic means removing "al" which we
        // already tried
        // ERROR
        lookup();
        return;
      }
      
      if (word.charAt(j) == 'i') { /* sometimes -ial endings should be removed */
        word.setLength(j); /* (sometimes it gets turned into -y, but we */
        k = j - 1; /* aren't dealing with that case for now) */
        if (lookup()) return;
        word.append("ial");
        k = old_k;
        lookup();
      }
      
    }
    return;
  }
  
  /*
   * this routine deals with -ive endings. It normalizes some of the -ative
   * endings directly, and also maps some -ive endings to -ion.
   */
  private void iveEndings() {
    int old_k = k;
    
    if (endsIn('i', 'v', 'e')) {
      word.setLength(j + 1); /* try removing -ive entirely */
      k = j;
      if (lookup()) return;
      
      word.unsafeWrite('e'); /* try removing -ive and adding -e */
      k = j + 1;
      if (lookup()) return;
      word.setLength(j + 1);
      word.append("ive");
      if ((j > 0) && (word.charAt(j - 1) == 'a') && (word.charAt(j) == 't')) {
        word.setCharAt(j - 1, 'e'); /* try removing -ative and adding -e */
        word.setLength(j); /* (e.g., determinative -> determine) */
        k = j - 1;
        if (lookup()) return;
        word.setLength(j - 1); /* try just removing -ative */
        if (lookup()) return;
        
        word.append("ative");
        k = old_k;
      }
      
      /* try mapping -ive to -ion (e.g., injunctive/injunction) */
      word.setCharAt(j + 2, 'o');
      word.setCharAt(j + 3, 'n');
      if (lookup()) return;
      
      word.setCharAt(j + 2, 'v'); /* restore the original values */
      word.setCharAt(j + 3, 'e');
      k = old_k;
      // nolookup()
    }
    return;
  }
  
  public KStemmer() {}
  
  public String stem(String term) {
    boolean changed = stem(term.toCharArray(), term.length());
    if (!changed) return term;
    return asString();
  }
  
  /**
   * Returns the result of the stem (assuming the word was changed) as a String.
   */
  String asString() {
    String s = getString();
    if (s != null) return s;
    return word.toString();
  }
  
  CharSequence asCharSequence() {
    return result != null ? result : word;
  }

  String getString() {
    return result;
  }
  
  char[] getChars() {
    return word.getArray();
  }
  
  int getLength() {
    return word.length();
  }
  
  String result;
  
  private boolean matched() {
    /***
     * if (!lookups.contains(word.toString())) { throw new
     * RuntimeException("didn't look up "+word.toString()+" prev="+prevLookup);
     * }
     ***/
    // lookup();
    return matchedEntry != null;
  }
  
  /**
   * Stems the text in the token. Returns true if changed.
   */
  boolean stem(char[] term, int len) {
    
    result = null;
    
    k = len - 1;
    if ((k <= 1) || (k >= MaxWordLen - 1)) {
      return false; // don't stem
    }
    
    // first check the stemmer dictionaries, and avoid using the
    // cache if it's in there.
    DictEntry entry = dict_ht.get(term, 0, len);
    if (entry != null) {
      if (entry.root != null) {
        result = entry.root;
        return true;
      }
      return false;
    }
    
    /***
     * caching off is normally faster if (cache == null) initializeStemHash();
     * 
     * // now check the cache, before we copy chars to "word" if (cache != null)
     * { String val = cache.get(term, 0, len); if (val != null) { if (val !=
     * SAME) { result = val; return true; } return false; } }
     ***/
    
    word.reset();
    // allocate enough space so that an expansion is never needed
    word.reserve(len + 10);
    for (int i = 0; i < len; i++) {
      char ch = term[i];
      if (!isAlpha(ch)) return false; // don't stem
      // don't lowercase... it's a requirement that lowercase filter be
      // used before this stemmer.
      word.unsafeWrite(ch);
    }
    
    matchedEntry = null;
    /***
     * lookups.clear(); lookups.add(word.toString());
     ***/
    
    /*
     * This while loop will never be executed more than one time; it is here
     * only to allow the break statement to be used to escape as soon as a word
     * is recognized
     */
    while (true) {
      // YCS: extra lookup()s were inserted so we don't need to
      // do an extra wordInDict() here.
      plural();
      if (matched()) break;
      pastTense();
      if (matched()) break;
      aspect();
      if (matched()) break;
      ityEndings();
      if (matched()) break;
      nessEndings();
      if (matched()) break;
      ionEndings();
      if (matched()) break;
      erAndOrEndings();
      if (matched()) break;
      lyEndings();
      if (matched()) break;
      alEndings();
      if (matched()) break;
      entry = wordInDict();
      iveEndings();
      if (matched()) break;
      izeEndings();
      if (matched()) break;
      mentEndings();
      if (matched()) break;
      bleEndings();
      if (matched()) break;
      ismEndings();
      if (matched()) break;
      icEndings();
      if (matched()) break;
      ncyEndings();
      if (matched()) break;
      nceEndings();
      matched();
      break;
    }
    
    /*
     * try for a direct mapping (allows for cases like `Italian'->`Italy' and
     * `Italians'->`Italy')
     */
    entry = matchedEntry;
    if (entry != null) {
      result = entry.root; // may be null, which means that "word" is the stem
    }
    
    // no entry matched means result is "word"
    return true;
  }
  
}