package org.thdl.tib.text ; import java.lang.* ; import java.util.regex.Pattern ; import java.util.regex.Matcher ; import java.util.Vector ; import java.util.Enumeration ; import java.io.BufferedReader ; import java.io.InputStreamReader ; public class Pronounciation { public Pronounciation () throws Exception { this ( "THDL" ) ; } public Pronounciation ( String which ) throws Exception { setRules ( which ) ; lastWordOriginal = "" ; } /** * pronounciation standard enumerator */ static Enumeration getStandardEnumeration () { Vector vec = new Vector () ; vec.add ( "THDL" ) ; return vec.elements () ; } Rule [] rules ; protected int stateFlag ; protected int returnFlag ; protected String lastWordOriginal ; class Rule { /** * conditions */ public static final int STARTS_WITH = 0 ; public static final int ENDS_WITH = 1 ; public static final int ENDS_WITH_FOLLOWS_VOWEL = 2 ; public static final int ENDS_WITH_FOLLOWS_CONSONANT = 3 ; public static final int STARTS_WITH_BORDERS_VOWEL = 4 ; public static final int STARTS_WITH_BORDERS_CONSONANT = 5 ; public static final int EQUALS = 6 ; public static final int CONTAINS = 7 ; public static final int STARTS_WITH_BORDERS_CONSONANTS = 8 ; /** * actions */ public static final int REPLACE_MATCH = 0 ; public static final int SET_FLAG = 1 ; public static final int REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL = 2 ; public static final int REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT = 3 ; public static final int REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICELESS_CONSONANT = 4 ; public static final int SET_RETURN_FLAG = 5 ; /** * flags */ public static final int FLAG_ENDS_WITH_VOWEL = 1 ; public static final int FLAG_ENDS_WITH_CONSONANT_VOICED = 2 ; public static final int FLAG_ENDS_WITH_CONSONANT_VOICELESS = 4 ; /** * return flags */ public static final int DO_NOTHING = 0 ; public static final int DROP_SUFFIX_AND_NASALIZE = 1 ; private int condition ; private String condArg ; private int action ; private Object actionArg ; private Pattern pattern ; private static final String vowelSet = "aeiou" ; private static final String consonantSet = "bBcCdDfgGhjkKlmnNpPrsStTvwXzZ" ; /** * constructior */ Rule ( int condition, String condArg, int action, Object actionArg ) throws Exception { this.condition = condition ; this.action = action ; this.condArg = condArg ; this.actionArg = actionArg ; String patStr = "" ; switch ( condition ) { case CONTAINS : patStr = condArg ; break ; case EQUALS : patStr = "^" + condArg + "$" ; break ; case STARTS_WITH : patStr = "^" + condArg ; break ; case ENDS_WITH : patStr = condArg + "$" ; break ; case ENDS_WITH_FOLLOWS_VOWEL : patStr = "([" + vowelSet + "]{1,2})" + condArg + "$" ; if ( REPLACE_MATCH == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT == action ) this.actionArg = "$1" + (String)actionArg ; break ; case ENDS_WITH_FOLLOWS_CONSONANT : patStr = "([" + consonantSet + "])" + condArg + "$" ; if ( REPLACE_MATCH == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT == action ) this.actionArg = "$1" + (String)actionArg ; break ; case STARTS_WITH_BORDERS_VOWEL : patStr = "^" + condArg + "([" + vowelSet + "]{1,2})" ; if ( REPLACE_MATCH == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT == action ) this.actionArg = (String)actionArg + "$1" ; break ; case STARTS_WITH_BORDERS_CONSONANT: patStr = "^" + condArg + "(([" + consonantSet + "]))" ; if ( REPLACE_MATCH == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT == action ) this.actionArg = (String)actionArg + "$1" ; break ; case STARTS_WITH_BORDERS_CONSONANTS: patStr = "^" + condArg + "([" + consonantSet + "]{2})" ; if ( REPLACE_MATCH == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL == action || REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT == action ) this.actionArg = (String)actionArg + "$1" ; break ; default: throw new Exception ( "Invalid condition for a rule." ) ; } pattern = Pattern.compile ( patStr ) ; } /** * property access */ public Object getActionArg () { return actionArg ; } public String getConditionArg () { return condArg ; } public int getCondition () { return condition ; } public int getAction () { return action ; } public Pattern getPattern () { return pattern ; } } /** * setRules */ protected void setRules ( String phoneticSystemName ) throws Exception { if ( phoneticSystemName == "THDL" ) { setThdl () ; } else { throw new Exception ( "Invalid phonetic system." ) ; } } protected void setThdl () throws Exception { // // based on http://www.thdl.org/xml/showEssay.php?xml=/collections/langling/THDL_phonetics.xml&l=d1e671 // Rule [] thdlRules = { // // 6. When ba and bo appear as the final syllable of a word, they are transcribed as "wa" and "wo," respectively. // This also includes ba'i ( > wé, about which see rule 16 below) and bar ( > war) as final syllables, although the latter is // more evident in literary forms // new Rule ( Rule.EQUALS, "ba", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL, "wa" ), new Rule ( Rule.EQUALS, "ba", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT, "wa" ), new Rule ( Rule.EQUALS, "ba", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICELESS_CONSONANT, "pa" ), new Rule ( Rule.EQUALS, "bo", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL, "wo" ), new Rule ( Rule.EQUALS, "bo", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT, "wo" ), new Rule ( Rule.EQUALS, "bo", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICELESS_CONSONANT, "po" ), new Rule ( Rule.EQUALS, "bar", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL, "war" ), new Rule ( Rule.EQUALS, "bar", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT, "war" ), new Rule ( Rule.EQUALS, "bar", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICELESS_CONSONANT, "par" ), new Rule ( Rule.EQUALS, "bo", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL, "wo" ), new Rule ( Rule.EQUALS, "bo", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT, "wo" ), new Rule ( Rule.EQUALS, "bo", Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICELESS_CONSONANT, "po" ), // // 7. The consonant clusters py, phy and by are transcribed respectively as "ch," "ch," and "j." // new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "py", Rule.REPLACE_MATCH, "ch" ), new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "Py", Rule.REPLACE_MATCH, "ch" ), new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "by", Rule.REPLACE_MATCH, "j" ) , // // 8. The consonant cluster my is transcribed as "ny." // new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "my", Rule.REPLACE_MATCH, "ny" ) , // // 13. When the second syllable of a word begins with the prefix achung ('), nasalization occurs // A. An 'n' is inserted after the first syllable, and the suffix of the first syllable (if there is one) is elided // B. If the root letter of the second syllable is pha or ba, an 'm' is inserted after the first syllable, // and the suffix of the first syllable (if there is one) is elided new Rule ( Rule.STARTS_WITH, "'", Rule.SET_RETURN_FLAG, new Integer ( Rule.DROP_SUFFIX_AND_NASALIZE ) ), // // dirty workaround // new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANTS, "g", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANTS, "d", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANTS, "b", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANTS, "m", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANTS, "'", Rule.REPLACE_MATCH, "" ), // // 9. Consonant clusters with r subscripts (which are pronounced as retroflexes) are transcribed with an "r." // new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "kr", Rule.REPLACE_MATCH, "tr" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "Kr", Rule.REPLACE_MATCH, "tr" ), new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "gr", Rule.REPLACE_MATCH, "dr" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "nr", Rule.REPLACE_MATCH, "n" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "pr", Rule.REPLACE_MATCH, "tr" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "Pr", Rule.REPLACE_MATCH, "tr" ), new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "br", Rule.REPLACE_MATCH, "dr" ) , // // For all other consonant clusters in which the r subscript is not pronounced, // such as mr, sr, and so forth, THDL Simplified Phonetics simply drops the "r" // in accordance with the general principle // new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "mr", Rule.REPLACE_MATCH, "m" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "Sr", Rule.REPLACE_MATCH, "S" ), new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "sr", Rule.REPLACE_MATCH, "s" ) , // // 10. Consonant clusters containing a subscript la are transcribed as "l" // with the exception of zl, which is transcribed as "d." // new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "kl", Rule.REPLACE_MATCH, "l" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "gl", Rule.REPLACE_MATCH, "l" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "bl", Rule.REPLACE_MATCH, "l" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "rl", Rule.REPLACE_MATCH, "l" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "zl", Rule.REPLACE_MATCH, "d" ) , new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "sl", Rule.REPLACE_MATCH, "l" ) , // // 11. Consonant clusters with an l superscript and h root letter are rendered "lh." // new Rule ( Rule.STARTS_WITH, "lh", Rule.REPLACE_MATCH, "hl" ), // // 12. Consonant clusters with a d prefix and b root letter undergo transformations in the following way, // depending on whether the consonant cluster includes the subscripts y or r: // new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "db", Rule.REPLACE_MATCH, "w" ), new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "dby", Rule.REPLACE_MATCH, "y" ), new Rule ( Rule.STARTS_WITH_BORDERS_VOWEL, "dbr", Rule.REPLACE_MATCH, "r" ), // !!!!! TODO !!!!! // Note: there are some exceptions to the form the nasalization takes: // skyabs 'gro > kyamdro // rten 'brel > temdrel // lam 'bras > lamdré // // http://www.thdl.org/xml/showEssay.php?xml=/collections/langling/THDL_phonetics.xml&l=d1e294 // The THDL Simplified Phonetic system, in contrast to Wylie, drops all Tibetan letters not // pronounced in a given syllable. This includes the superscribed consonants r, l, and s ; new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "r", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "l", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "s", Rule.REPLACE_MATCH, "" ), // // .... the prefixes g, d, b, m, and ' ; // new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "g", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "d", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "b", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "m", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.STARTS_WITH_BORDERS_CONSONANT, "'", Rule.REPLACE_MATCH, "" ), // the suffixes d , ' , and s ; ( THEY ARE HANDLED ELSEWHERE ) // // .... and the post-suffixes s and d. // new Rule ( Rule.ENDS_WITH_FOLLOWS_CONSONANT, "s", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.ENDS_WITH_FOLLOWS_CONSONANT, "d", Rule.REPLACE_MATCH, "" ), // // 15. When two of the same vowels are connected by an achung, they are transcribed by dropping the achung and // combining the two vowels into one // new Rule ( Rule.CONTAINS, "a'a", Rule.REPLACE_MATCH, "a" ), new Rule ( Rule.CONTAINS, "e'e", Rule.REPLACE_MATCH, "e" ), new Rule ( Rule.CONTAINS, "i'i", Rule.REPLACE_MATCH, "i" ), new Rule ( Rule.CONTAINS, "o'o", Rule.REPLACE_MATCH, "o" ), new Rule ( Rule.CONTAINS, "u'u", Rule.REPLACE_MATCH, "u" ), new Rule ( Rule.ENDS_WITH, "as", Rule.REPLACE_MATCH, "é" ) , new Rule ( Rule.ENDS_WITH, "ad", Rule.REPLACE_MATCH, "é" ) , new Rule ( Rule.ENDS_WITH, "an", Rule.REPLACE_MATCH, "én" ) , new Rule ( Rule.ENDS_WITH, "al", Rule.REPLACE_MATCH, "él" ) , new Rule ( Rule.ENDS_WITH, "os", Rule.REPLACE_MATCH, "ö" ) , new Rule ( Rule.ENDS_WITH, "od", Rule.REPLACE_MATCH, "ö" ) , new Rule ( Rule.ENDS_WITH, "on", Rule.REPLACE_MATCH, "ön" ) , new Rule ( Rule.ENDS_WITH, "ol", Rule.REPLACE_MATCH, "öl" ) , new Rule ( Rule.ENDS_WITH, "u'", Rule.REPLACE_MATCH, "u" ) , new Rule ( Rule.ENDS_WITH, "us", Rule.REPLACE_MATCH, "ü" ) , new Rule ( Rule.ENDS_WITH, "ud", Rule.REPLACE_MATCH, "ü" ) , new Rule ( Rule.ENDS_WITH, "un", Rule.REPLACE_MATCH, "ün" ) , new Rule ( Rule.ENDS_WITH, "ul", Rule.REPLACE_MATCH, "ül" ) , new Rule ( Rule.ENDS_WITH, "es", Rule.REPLACE_MATCH, "e" ) , new Rule ( Rule.ENDS_WITH, "ed", Rule.REPLACE_MATCH, "e" ) , new Rule ( Rule.ENDS_WITH, "en", Rule.REPLACE_MATCH, "en" ) , new Rule ( Rule.ENDS_WITH, "el", Rule.REPLACE_MATCH, "el" ) , new Rule ( Rule.ENDS_WITH, "is", Rule.REPLACE_MATCH, "i" ) , new Rule ( Rule.ENDS_WITH, "id", Rule.REPLACE_MATCH, "i" ) , // // 5. The suffixes g and b are devoiced and rendered "k" and "p," respectively, // since this most closely approximates actual pronunciation. // new Rule ( Rule.ENDS_WITH_FOLLOWS_VOWEL, "g", Rule.REPLACE_MATCH, "k" ), new Rule ( Rule.ENDS_WITH_FOLLOWS_VOWEL, "b", Rule.REPLACE_MATCH, "p" ), // these rule must be *last* new Rule ( Rule.ENDS_WITH, "[aeiou]", Rule.SET_FLAG, new Integer(Rule.FLAG_ENDS_WITH_VOWEL) ), new Rule ( Rule.ENDS_WITH, "(b|d|g|l|m|n|N|r)", Rule.SET_FLAG, new Integer ( Rule.FLAG_ENDS_WITH_CONSONANT_VOICED ) ), new Rule ( Rule.ENDS_WITH, "(k|p|s|t)", Rule.SET_FLAG, new Integer ( Rule.FLAG_ENDS_WITH_CONSONANT_VOICELESS ) ), new Rule ( Rule.ENDS_WITH, "a'i", Rule.REPLACE_MATCH, "e" ) , new Rule ( Rule.ENDS_WITH, "e'i", Rule.REPLACE_MATCH, "e" ) , new Rule ( Rule.ENDS_WITH, "o'i", Rule.REPLACE_MATCH, "ö" ) , new Rule ( Rule.ENDS_WITH, "u'i", Rule.REPLACE_MATCH, "ü" ), // // 14. Multiple vowels that have discrete sounds and are connected by an achung (') are transcribed by dropping the achung // (at this point all achungs have been removed by previous rules) // new Rule ( Rule.CONTAINS, "'", Rule.REPLACE_MATCH, "" ), new Rule ( Rule.CONTAINS, "X", Rule.REPLACE_MATCH, "ts" ), new Rule ( Rule.CONTAINS, "T", Rule.REPLACE_MATCH, "t" ), new Rule ( Rule.CONTAINS, "P", Rule.REPLACE_MATCH, "p" ), } ; rules = thdlRules ; } /** * applyRule */ protected String applyRule ( Rule rule, String in ) throws Exception { switch ( rule.getAction () ) { // // Rule.REPLACE_MATCH - if text defined in condArg found, replace it with actionArg unconditionally // case Rule.REPLACE_MATCH : { Matcher matcher = rule.getPattern ().matcher ( in ) ; return matcher.replaceFirst ( (String)rule.getActionArg () ) ; } // // Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL - if text defined in condArg found, // AND last syllable ends with a vowel - replace it with actionArg // case Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOWEL : if ( 0 != ( stateFlag & Rule.FLAG_ENDS_WITH_VOWEL ) ) { Matcher matcher = rule.getPattern ().matcher ( in ) ; return matcher.replaceFirst ( (String)rule.getActionArg () ) ; } break ; // // Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT - if text defined in condArg found, // AND last syllable ends with a voiced consonant - replace it with actionArg // case Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICED_CONSONANT : if ( 0 != ( stateFlag & Rule.FLAG_ENDS_WITH_CONSONANT_VOICED ) ) { Matcher matcher = rule.getPattern ().matcher ( in ) ; return matcher.replaceFirst ( (String)rule.getActionArg () ) ; } break ; // // Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICELESS_CONSONANT - if text defined in condArg found, // AND last syllable ends with a voiceless consonant - replace it with actionArg // case Rule.REPLACE_IF_LAST_SYLLABLE_ENDED_WITH_VOICELESS_CONSONANT : if ( 0 != ( stateFlag & Rule.FLAG_ENDS_WITH_CONSONANT_VOICELESS ) ) { Matcher matcher = rule.getPattern ().matcher ( in ) ; return matcher.replaceFirst ( (String)rule.getActionArg () ) ; } break ; // // Rule.SET_FLAG - set the flag // case Rule.SET_FLAG : if ( rule.getPattern ().matcher ( in ).find () ) { stateFlag = ((Integer) rule.getActionArg ()).intValue () ; } break ; // // Rule.SET_RETURN_FLAG - set the return flag // case Rule.SET_RETURN_FLAG : if ( rule.getPattern ().matcher ( in ).find () ) { returnFlag = ((Integer) rule.getActionArg ()).intValue () ; } break ; // // shouldn't ever happen // default: throw new Exception ( "Invalid action." ) ; } return in ; } /** * processWord */ protected String processWord ( String in ) throws Exception { // // replace the roman digraphs // in = in.replaceAll ( "bh", "B" ) ; in = in.replaceAll ( "ch", "C" ) ; in = in.replaceAll ( "dz", "D" ) ; in = in.replaceAll ( "kh", "K" ) ; in = in.replaceAll ( "ng", "N" ) ; in = in.replaceAll ( "ph", "P" ) ; in = in.replaceAll ( "sh", "S" ) ; in = in.replaceAll ( "th", "T" ) ; in = in.replaceAll ( "ts", "X" ) ; in = in.replaceAll ( "tsh", "Q" ) ; in = in.replaceAll ( "zh", "Z" ) ; returnFlag = Rule.DO_NOTHING ; lastWordOriginal = in ; // // run the word through all rules // for ( int i = 0; i < rules.length; i++ ) { in = applyRule ( rules [i], in ) ; } // // put back the roman digraphs // in = in.replaceAll ( "B", "bh" ) ; in = in.replaceAll ( "C", "ch" ) ; in = in.replaceAll ( "D", "dz" ) ; in = in.replaceAll ( "K", "kh" ) ; in = in.replaceAll ( "N", "ng" ) ; in = in.replaceAll ( "P", "ph" ) ; in = in.replaceAll ( "S", "sh" ) ; in = in.replaceAll ( "T", "th" ) ; in = in.replaceAll ( "X", "ts" ) ; in = in.replaceAll ( "Q", "tsh" ) ; in = in.replaceAll ( "Z", "zh" ) ; return in ; } private static boolean isVowel ( char c ) { return ( -1 != "aeiouéöü".indexOf ( c ) ) ; } private static String dropSuffix ( String in ) { while ( !isVowel ( in.charAt ( in.length () - 1 ) ) ) in = in.substring ( 0, in.length () - 1 ) ; return in ; } private String nasalize ( String in, String followingSyllable ) { if ( followingSyllable.startsWith ( "'by" ) || followingSyllable.startsWith ( "'br" ) ) in = in + "n" ; else if ( followingSyllable.startsWith ( "'b" ) || lastWordOriginal.startsWith ( "'ph" ) ) in = in + "m" ; else in = in + "n" ; return in ; } /** * process */ public String process ( String in ) throws Exception { stateFlag = 0 ; String out = "" ; String [] matchWords = in.split ( "[^a-z']+" ) ; for ( int i = 0; i < matchWords.length; i++ ) { String newWord = processWord ( matchWords [i] ) ; if ( i > 0 && Rule.DROP_SUFFIX_AND_NASALIZE == returnFlag ) { out = dropSuffix ( out ) ; out = nasalize ( out, newWord ) ; } if ( out.length () > 0 ) out += " " ; out += newWord ; } return out ; } /** * main */ public static void main ( String [] args ) { try { Pronounciation p = new Pronounciation ( "THDL" ) ; BufferedReader br = new BufferedReader ( new InputStreamReader ( System.in ) ) ; while ( true ) { String in = br.readLine () ; if ( in == null ) break ; } } catch ( Exception e ) { System.err.println ( e.toString () ) ; } } };