edu.stanford.nlp.process.PTBLexer.flex Maven / Gradle / Ivy
package edu.stanford.nlp.process;
// Stanford English Tokenizer -- a deterministic, fast high-quality tokenizer
// Copyright (c) 2002-2009 The Board of Trustees of
// The Leland Stanford Junior University. All Rights Reserved.
//
// This program is free software; you can redistribute it and/or
// modify it under the terms of the GNU General Public License
// as published by the Free Software Foundation; either version 2
// of the License, or (at your option) any later version.
//
// This program is distributed in the hope that it will be useful,
// but WITHOUT ANY WARRANTY; without even the implied warranty of
// MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
// GNU General Public License for more details.
//
// You should have received a copy of the GNU General Public License
// along with this program; if not, write to the Free Software
// Foundation, Inc., 59 Temple Place - Suite 330, Boston, MA 02111-1307, USA.
//
// For more information, bug reports, fixes, contact:
// Christopher Manning
// Dept of Computer Science, Gates 1A
// Stanford CA 94305-9010
// USA
// [email protected]
// http://nlp.stanford.edu/software/
import java.io.Reader;
import java.util.logging.Logger;
import java.util.Map;
import java.util.Map.Entry;
import java.util.Properties;
import java.util.Set;
import edu.stanford.nlp.ling.CoreLabel;
import edu.stanford.nlp.ling.CoreAnnotations.AfterAnnotation;
import edu.stanford.nlp.ling.CoreAnnotations.BeforeAnnotation;
import edu.stanford.nlp.ling.CoreAnnotations.OriginalTextAnnotation;
import edu.stanford.nlp.util.StringUtils;
/** Provides a tokenizer or lexer that does a pretty good job at
* deterministically tokenizing English according to Penn Treebank conventions.
* The class is a scanner generated by
* JFlex (1.4.3) from the specification
* file
* PTBLexer.flex. As well as copying what is in the Treebank,
* it now contains some extensions to deal with modern text and encoding
* issues, such as recognizing URLs and common Unicode characters, and a
* variety of options for doing or suppressing certain normalizations.
* Although they shouldn't really be there, it also interprets certain of the
* characters between U+0080 and U+009F as Windows CP1252 characters.
*
* Fine points: Output normalized tokens should not contain spaces,
* providing the normalizeSpace option is true. The space will be turned
* into a non-breaking space (U+00A0). Otherwise, they can appear in
* a couple of token classes (phone numbers, fractions).
* The original
* PTB tokenization (messy) standard also escapes certain other characters,
* such as * and /, and normalizes things like " to `` or ''. By default,
* this tokenizer does all of those things. However, you can turn them
* off by using the ptb3Escaping=false option, or, parts of it on or off,
* or unicode
* character alternatives on with different options. You can also build an
* invertible tokenizer, with which you can still access the original
* character sequence and the non-token whitespace around it in a CoreLabel.
* And you can ask for newlines to be tokenized.
*
* Character entities: For legacy reasons, this file will parse and
* interpret some simply SGML/XML/HTML character entities. For modern formats
* like XML, you are better off doing XML parsing, and then running the
* tokenizer on CDATA elements. But we and others frequently work with simple
* SGML text corpora that are not XML (like LDC text collections). In practice,
* they only include very simple markup and a few simple entities, and the
* combination of the -parseInside option and the minimal character entity
* support in this file is enough to handle them. So we leave this functionality
* in, even though it could conceivably mess with a correct XML file if the
* output of decoding had things that look like character entities. In general,
* handled symbols are changed to ASCII/Unicode forms, but handled accented
* letters are just left as character entities in words.
*
* Character support: PTBLexer works works for a large subset of
* Unicode Base Multilingual Plane characters (only). It recognizes all
* characters that match the JFlex/Java [:letter:] and [:digit] character
* class (but, unfortunately, JFlex does not support most
* other Unicode character classes available in Java regular expressions).
* It also matches all defined characters in the Unicode ranges U+0000-U+07FF
* excluding control characters except the ones very standardly found in
* plain text documents. Finally select other characters commonly found in
* English unicode text are included.
*
* Implementation note: The scanner is caseless, but note, if adding
* or changing regexps, that caseless does not expand inside character
* classes. From the manual: "The %caseless option does not change the
* matched text and does not effect character classes. So [a] still only
* matches the character a and not A, too." Note that some character
* classes still deliberately don't have both cases, so the scanner's
* operation isn't completely case-independent, though it mostly is.
*
* Implementation note: This Java class is automatically generated
* from PTBLexer.flex using jflex. DO NOT EDIT THE JAVA SOURCE. This file
* has now been updated for JFlex 1.4.2+. (This required code changes: this
* version only works right with JFlex 1.4.2+; the previous version only works
* right with JFlex 1.4.1.)
*
* @author Tim Grow
* @author Christopher Manning
* @author Jenny Finkel
*/
%%
%class PTBLexer
%unicode
%function next
%type Object
%char
%caseless
%state YyStrictlyTreebank3 YyTraditionalTreebank3
%{
/**
* Constructs a new PTBLexer. You specify the type of result tokens with a
* LexedTokenFactory, and can specify the treatment of tokens by boolean
* options given in a comma separated String
* (e.g., "invertible,normalizeParentheses=true").
* If the String is null or empty, you get the traditional
* PTB3 normalization behaviour (i.e., you get ptb3Escaping=false). If you
* want no normalization, then you should pass in the String
* "ptb3Escaping=false". The known option names are:
*
* - invertible: Store enough information about the original form of the
* token and the whitespace around it that a list of tokens can be
* faithfully converted back to the original String. Valid only if the
* LexedTokenFactory is an instance of CoreLabelTokenFactory. The
* keys used in it are TextAnnotation for the tokenized form,
* OriginalTextAnnotation for the original string, BeforeAnnotation and
* AfterAnnotation for the whitespace before and after a token, and
* perhaps BeginPositionAnnotation and EndPositionAnnotation to record
* token begin/after end offsets, if they were specified to be recorded
* in TokenFactory construction. (Like the String class, begin and end
* are done so end - begin gives the token length.)
*
- tokenizeNLs: Whether end-of-lines should become tokens (or just
* be treated as part of whitespace)
*
- ptb3Escaping: Enable all traditional PTB3 token transforms
* (like -LRB-, -RRB-). This is a macro flag that sets or clears all the
* options below.
*
- americanize: Whether to rewrite common British English spellings
* as American English spellings
*
- normalizeSpace: Whether any spaces in tokens (phone numbers, fractions
* get turned into U+00A0 (non-breaking space). It's dangerous to turn
* this off for most of our Stanford NLP software, which assumes no
* spaces in tokens.
*
- normalizeAmpersandEntity: Whether to map the XML & to an
* ampersand
*
- normalizeCurrency: Whether to do some awful lossy currency mappings
* to turn common currency characters into $, #, or "cents", reflecting
* the fact that nothing else appears in the old PTB3 WSJ. (No Euro!)
*
- normalizeFractions: Whether to map certain common composed
* fraction characters to spelled out letter forms like "1/2"
*
- normalizeParentheses: Whether to map round parentheses to -LRB-,
* -RRB-, as in the Penn Treebank
*
- normalizeOtherBrackets: Whether to map other common bracket characters
* to -LCB-, -LRB-, -RCB-, -RRB-, roughly as in the Penn Treebank
*
- asciiQuotes Whether to map quote characters to the traditional ' and "
*
- latexQuotes: Whether to map to ``, `, ', '' for quotes, as in Latex
* and the PTB3 WSJ (though this is now heavily frowned on in Unicode).
* If true, this takes precedence over the setting of unicodeQuotes;
* if both are false, no mapping is done.
*
- unicodeQuotes: Whether to map quotes to the range U+2018 to U+201D,
* the preferred unicode encoding of single and double quotes.
*
- ptb3Ellipsis: Whether to map ellipses to ..., the old PTB3 WSJ coding
* of an ellipsis. If true, this takes precedence over the setting of
* unicodeEllipsis; if both are false, no mapping is done.
*
- unicodeEllipsis: Whether to map dot and optional space sequences to
* U+2026, the Unicode ellipsis character
*
- ptb3Dashes: Whether to turn various dash characters into "--",
* the dominant encoding of dashes in the PTB3 WSJ
*
- escapeForwardSlashAsterisk: Whether to put a backslash escape in front
* of / and * as the old PTB3 WSJ does for some reason (something to do
* with Lisp readers??).
*
- untokenizable: What to do with untokenizable characters (ones not
* known to the tokenizers). Six options combining whether to log a
* warning for none, the first, or all, and whether to delete them or
* to include them as single character tokens in the output: noneDelete,
* firstDelete, allDelete, noneKeep, firstKeep, allKeep.
* The default is "firstDelete".
*
- strictTreebank3: PTBTokenizer deliberately deviates from strict PTB3
* WSJ tokenization in two cases. Setting this improves compatibility
* for those cases. They are: (i) When an acronym is followed by a
* sentence end, such as "U.S." at the end of a sentence, the PTB3
* has tokens of "U.S" and ".", while by default PTBTokenzer duplicates
* the period returning tokens of "U.S." and ".", and (ii) PTBTokenizer
* will return numbers with a whole number and a fractional part like
* "5 7/8" as a single token (with a non-breaking space in the middle),
* while the PTB3 separates them into two tokens "5" and "7/8".
*
*
* @param r The Reader to tokenize text from
* @param tf The LexedTokenFactory that will be invoked to convert
* each substring extracted by the lexer into some kind of Object
* (such as a Word or CoreLabel).
*/
public PTBLexer(Reader r, LexedTokenFactory tf, String options) {
this(r);
this.tokenFactory = tf;
if (options == null) {
options = "";
}
Properties prop = StringUtils.stringToProperties(options);
Set> props = prop.entrySet();
for (Map.Entry item : props) {
String key = (String) item.getKey();
String value = (String) item.getValue();
boolean val = Boolean.valueOf(value);
if ("".equals(key)) {
// allow an empty item
} else if ("invertible".equals(key)) {
invertible = val;
} else if ("tokenizeNLs".equals(key)) {
tokenizeNLs = val;
} else if ("ptb3Escaping".equals(key)) {
americanize = val;
normalizeSpace = val;
normalizeAmpersandEntity = val;
normalizeCurrency = val;
normalizeFractions = val;
normalizeParentheses = val;
normalizeOtherBrackets = val;
latexQuotes = val;
unicodeQuotes = val;
asciiQuotes = val;
ptb3Ellipsis = val;
unicodeEllipsis = val;
ptb3Dashes = val;
escapeForwardSlashAsterisk = val;
} else if ("americanize".equals(key)) {
americanize = val;
} else if ("normalizeSpace".equals(key)) {
normalizeSpace = val;
} else if ("normalizeAmpersandEntity".equals(key)) {
normalizeAmpersandEntity = val;
} else if ("normalizeCurrency".equals(key)) {
normalizeCurrency = val;
} else if ("normalizeFractions".equals(key)) {
normalizeFractions = val;
} else if ("normalizeParentheses".equals(key)) {
normalizeParentheses = val;
} else if ("normalizeOtherBrackets".equals(key)) {
normalizeOtherBrackets = val;
} else if ("latexQuotes".equals(key)) {
latexQuotes = val;
} else if ("unicodeQuotes".equals(key)) {
unicodeQuotes = val;
if (val) {
latexQuotes = false; // need to override default
}
} else if ("asciiQuotes".equals(key)) {
asciiQuotes = val;
if (val) {
latexQuotes = false; // need to override default
unicodeQuotes = false;
}
} else if ("ptb3Ellipsis".equals(key)) {
ptb3Ellipsis = val;
} else if ("unicodeEllipsis".equals(key)) {
unicodeEllipsis = val;
} else if ("ptb3Dashes".equals(key)) {
ptb3Dashes = val;
} else if ("escapeForwardSlashAsterisk".equals(key)) {
escapeForwardSlashAsterisk = val;
} else if ("untokenizable".equals(key)) {
if (value.equals("noneDelete")) {
untokenizable = UntokenizableOptions.NONE_DELETE;
} else if (value.equals("firstDelete")) {
untokenizable = UntokenizableOptions.FIRST_DELETE;
} else if (value.equals("allDelete")) {
untokenizable = UntokenizableOptions.ALL_DELETE;
} else if (value.equals("noneKeep")) {
untokenizable = UntokenizableOptions.NONE_KEEP;
} else if (value.equals("firstKeep")) {
untokenizable = UntokenizableOptions.FIRST_KEEP;
} else if (value.equals("allKeep")) {
untokenizable = UntokenizableOptions.ALL_KEEP;
} else {
throw new IllegalArgumentException("PTBLexer: Invalid option value in constructor: " + key + ": " + value);
}
} else if ("strictTreebank3".equals(key)) {
strictTreebank3 = val;
} else {
throw new IllegalArgumentException("PTBLexer: Invalid options key in constructor: " + key);
}
}
// this.seenUntokenizableCharacter = false; // unnecessary, it's default initialized
if (invertible) {
if ( ! (tf instanceof CoreLabelTokenFactory)) {
throw new IllegalArgumentException("PTBLexer: the invertible option requires a CoreLabelTokenFactory");
}
prevWord = (CoreLabel) tf.makeToken("", 0, 0);
prevWordAfter = new StringBuilder();
}
if (strictTreebank3) {
yybegin(YyStrictlyTreebank3);
} else {
yybegin(YyTraditionalTreebank3);
}
}
private static final Logger LOGGER = Logger.getLogger(PTBLexer.class.getName());
private LexedTokenFactory tokenFactory;
private CoreLabel prevWord;
private StringBuilder prevWordAfter;
private boolean seenUntokenizableCharacter;
private enum UntokenizableOptions { NONE_DELETE, FIRST_DELETE, ALL_DELETE, NONE_KEEP, FIRST_KEEP, ALL_KEEP }
private UntokenizableOptions untokenizable = UntokenizableOptions.FIRST_DELETE;
/* Flags begin with historical ptb3Escaping behavior */
private boolean invertible;
private boolean tokenizeNLs;
private boolean americanize = true;
private boolean normalizeSpace = true;
private boolean normalizeAmpersandEntity = true;
private boolean normalizeCurrency = true;
private boolean normalizeFractions = true;
private boolean normalizeParentheses = true;
private boolean normalizeOtherBrackets = true;
private boolean latexQuotes = true;
private boolean unicodeQuotes;
private boolean asciiQuotes;
private boolean ptb3Ellipsis = true;
private boolean unicodeEllipsis;
private boolean ptb3Dashes = true;
private boolean escapeForwardSlashAsterisk = true;
private boolean strictTreebank3 = false;
/*
* This has now been extended to cover the main Windows CP1252 characters,
* at either their correct Unicode codepoints, or in their invalid
* positions as 8 bit chars inside the iso-8859 control region.
*
* ellipsis 85 0133 2026 8230
* single quote curly starting 91 0145 2018 8216
* single quote curly ending 92 0146 2019 8217
* double quote curly starting 93 0147 201C 8220
* double quote curly ending 94 0148 201D 8221
* en dash 96 0150 2013 8211
* em dash 97 0151 2014 8212
*/
public static final String opendblquote = "``";
public static final String closedblquote = "''";
public static final String openparen = "-LRB-";
public static final String closeparen = "-RRB-";
public static final String openbrace = "-LCB-";
public static final String closebrace = "-RCB-";
public static final String ptbmdash = "--";
public static final String ptb3EllipsisStr = "...";
public static final String unicodeEllipsisStr = "\u2026";
/** For tokenizing carriage returns. (JS) */
public static final String NEWLINE_TOKEN = "*NL*";
private Object normalizeFractions(final String in) {
String out = in;
if (normalizeFractions) {
if (escapeForwardSlashAsterisk) {
out = out.replaceAll("\u00BC", "1\\\\/4");
out = out.replaceAll("\u00BD", "1\\\\/2");
out = out.replaceAll("\u00BE", "3\\\\/4");
out = out.replaceAll("\u2153", "1\\\\/3");
out = out.replaceAll("\u2153", "2\\\\/3");
} else {
out = out.replaceAll("\u00BC", "1/4");
out = out.replaceAll("\u00BD", "1/2");
out = out.replaceAll("\u00BE", "3/4");
out = out.replaceAll("\u2153", "1/3");
out = out.replaceAll("\u2153", "2/3");
}
}
// System.err.println("normalizeFractions="+normalizeFractions+", escapeForwardSlashAsterisk="+escapeForwardSlashAsterisk);
// System.err.println("Mapped |"+in+"| to |" + out + "|.");
return getNext(out, in);
}
private static String removeSoftHyphens(String in) {
String result = in.replaceAll("\u00AD", "");
if (result.length() == 0) {
return "-";
} else {
return result;
}
}
private static String normalizeCurrency(String in) {
String s1 = in;
s1 = s1.replaceAll("\u00A2", "cents");
s1 = s1.replaceAll("\u00A3", "#"); // historically used for pound in PTB3
s1 = s1.replaceAll("[\u0080\u00A4\u20A0\u20AC]", "\\$"); // Euro (ECU, generic currency) -- no good translation!
return s1;
}
private static String latexQuotes(String in, boolean probablyLeft) {
String s1 = in;
if (probablyLeft) {
s1 = s1.replaceAll("'|'", "`");
s1 = s1.replaceAll("\"|"", "``");
} else {
s1 = s1.replaceAll("'|'", "'");
s1 = s1.replaceAll("\"|"", "''");
}
s1 = s1.replaceAll("[\u0091\u2018\u201B\u2039]", "`");
s1 = s1.replaceAll("[\u0092\u2019\u203A]", "'");
s1 = s1.replaceAll("[\u0093\u201C\u00AB]", "``");
s1 = s1.replaceAll("[\u0094\u201D\u00BB]", "''");
return s1;
}
private static String asciiQuotes(String in, boolean probablyLeft) {
String s1 = in;
s1 = s1.replaceAll("'|[\u0091\u2018\u0092\u2019\u201A\u201B\u2039\u203A']", "'");
s1 = s1.replaceAll(""|[\u0093\u201C\u0094\u201D\u201E\u00AB\u00BB\"]", "\"");
return s1;
}
private static String unicodeQuotes(String in, boolean probablyLeft) {
String s1 = in;
if (probablyLeft) {
s1 = s1.replaceAll("'|'", "\u2018");
s1 = s1.replaceAll("\"|"", "\u201c");
} else {
s1 = s1.replaceAll("'|'", "\u2019");
s1 = s1.replaceAll("\"|"", "\u201d");
}
s1 = s1.replaceAll("[\u0091\u2018]", "\u2018");
s1 = s1.replaceAll("[\u0092\u2019]", "\u2019");
s1 = s1.replaceAll("[\u0093\u201C]", "\u201c");
s1 = s1.replaceAll("[\u0094\u201D]", "\u201d");
return s1;
}
private Object handleQuotes(String tok, boolean probablyLeft) {
String normTok;
if (latexQuotes) {
normTok = latexQuotes(tok, probablyLeft);
} else if (unicodeQuotes) {
normTok = unicodeQuotes(tok, probablyLeft);
} else if (asciiQuotes) {
normTok = asciiQuotes(tok, probablyLeft);
} else {
normTok = tok;
}
return getNext(normTok, tok);
}
private Object handleEllipsis(final String tok) {
if (ptb3Ellipsis) {
return getNext(ptb3EllipsisStr, tok);
} else if (unicodeEllipsis) {
return getNext(unicodeEllipsisStr, tok);
} else {
return getNext(tok, tok);
}
}
/** This quotes a character with a backslash, but doesn't do it
* if the character is already preceded by a backslash.
*/
private static String delimit(String s, char c) {
int i = s.indexOf(c);
while (i != -1) {
if (i == 0 || s.charAt(i - 1) != '\\') {
s = s.substring(0, i) + '\\' + s.substring(i);
i = s.indexOf(c, i + 2);
} else {
i = s.indexOf(c, i + 1);
}
}
return s;
}
private static String normalizeAmp(final String in) {
return in.replaceAll("(?i:&)", "&");
}
private Object getNext() {
final String txt = yytext();
return getNext(txt, txt);
}
/** Make the next token.
* @param txt What the token should be
* @param originalText The original String that got transformed into txt
*/
private Object getNext(String txt, String originalText) {
if (invertible) {
String str = prevWordAfter.toString();
prevWordAfter.setLength(0);
CoreLabel word = (CoreLabel) tokenFactory.makeToken(txt, yychar, yylength());
word.set(OriginalTextAnnotation.class, originalText);
word.set(BeforeAnnotation.class, str);
prevWord.set(AfterAnnotation.class, str);
prevWord = word;
return word;
} else {
return tokenFactory.makeToken(txt, yychar, yylength());
}
}
private Object getNormalizedAmpNext() {
final String txt = yytext();
if (normalizeAmpersandEntity) {
return getNext(normalizeAmp(txt), txt);
} else {
return getNext();
}
}
%}
/* Don't allow SGML to cross lines, even though it can...
Really SGML shouldn't be here at all, it's kind of legacy. */
SGML = <\/?[A-Za-z!?][^>\r\n]*>
SPMDASH = &(MD|mdash|ndash);|[\u0096\u0097\u2013\u2014\u2015]
SPAMP = &
SPPUNC = &(HT|TL|UR|LR|QC|QL|QR|odq|cdq|#[0-9]+);
SPLET = &[aeiouAEIOU](acute|grave|uml);
/* \u3000 is ideographic space */
SPACE = [ \t\u00A0\u2000-\u200A\u3000]
SPACES = {SPACE}+
NEWLINE = \r|\r?\n|\u2028|\u2029|\u000B|\u000C|\u0085
SPACENL = ({SPACE}|{NEWLINE})
SENTEND = {SPACENL}({SPACENL}|([A-Z]|{SGML}))
DIGIT = [:digit:]|[\u07C0-\u07C9]
DATE = {DIGIT}{1,2}[\-\/]{DIGIT}{1,2}[\-\/]{DIGIT}{2,4}
NUM = {DIGIT}+|{DIGIT}*([.:,\u00AD\u066B\u066C]{DIGIT}+)+
/* Now don't allow bracketed negative numbers! They have too many uses (e.g.,
years or times in parentheses), and having them in tokens messes up
treebank parsing.
NUMBER = [\-+]?{NUM}|\({NUM}\) */
NUMBER = [\-+]?{NUM}
SUBSUPNUM = [\u207A\u207B\u208A\u208B]?([\u2070\u00B9\u00B2\u00B3\u2074-\u2079]+|[\u2080-\u2089]+)
/* Constrain fraction to only match likely fractions */
FRAC = ({DIGIT}{1,4}[- \u00A0])?{DIGIT}{1,4}(\\?\/|\u2044){DIGIT}{1,4}
FRACSTB3 = ({DIGIT}{1,4}-)?{DIGIT}{1,4}(\\?\/|\u2044){DIGIT}{1,4}
FRAC2 = [\u00BC\u00BD\u00BE\u2153-\u215E]
DOLSIGN = ([A-Z]*\$|#)
/* These are cent and pound sign, euro and euro, and Yen, Lira */
DOLSIGN2 = [\u00A2\u00A3\u00A4\u00A5\u0080\u20A0\u20AC\u060B\u0E3F\u20A4\uFFE0\uFFE1\uFFE5\uFFE6]
/* not used DOLLAR {DOLSIGN}[ \t]*{NUMBER} */
/* |\( ?{NUMBER} ?\)) # is for pound signs */
/* For some reason U+0237-U+024F (dotless j) isn't in [:letter:]. Recent additions? */
WORD = ([:letter:]|{SPLET}|[\u00AD\u0237-\u024F\u02C2-\u02C5\u02D2-\u02DF\u02E5-\u02FF\u0300-\u036F\u0370-\u037D\u0384\u0385\u03CF\u03F6\u03FC-\u03FF\u0483-\u0487\u04CF\u04F6-\u04FF\u0510-\u0525\u055A-\u055F\u0591-\u05BD\u05BF\u05C1\u05C2\u05C4\u05C5\u05C7\u0615-\u061A\u063B-\u063F\u064B-\u065E\u0670\u06D6-\u06EF\u06FA-\u06FF\u070F\u0711\u0730-\u074F\u0750-\u077F\u07A6-\u07B1\u07CA-\u07F5\u07FA\u0900-\u0903\u093C\u093E-\u094E\u0951-\u0955\u0962-\u0963\u0981-\u0983\u09BC-\u09C4\u09C7\u09C8\u09CB-\u09CD\u09D7\u09E2\u09E3\u0A01-\u0A03\u0A3C\u0A3E-\u0A4F\u0A81-\u0A83\u0ABC-\u0ACF\u0B82\u0BBE-\u0BC2\u0BC6-\u0BC8\u0BCA-\u0BCD\u0C01-\u0C03\u0C3E-\u0C56\u0D3E-\u0D44\u0D46-\u0D48\u0E30-\u0E3A\u0E47-\u0E4E\u0EB1-\u0EBC\u0EC8-\u0ECD])+
/* The $ was for things like New$ */
/* WAS: only keep hyphens with short one side like co-ed */
/* But treebank just allows hyphenated things as words! */
THING = ([dDoOlL]{APOSETCETERA}([:letter:]|[:digit:]))?([:letter:]|[:digit:])+({HYPHEN}([dDoOlL]{APOSETCETERA}([:letter:]|[:digit:]))?([:letter:]|[:digit:])+)*
THINGA = [A-Z]+(([+&]|{SPAMP})[A-Z]+)+
THING3 = [A-Za-z0-9]+(-[A-Za-z]+){0,2}(\\?\/[A-Za-z0-9]+(-[A-Za-z]+){0,2}){1,2}
APOS = ['\u0092\u2019]|'
/* Includes extra ones that may appear inside a word, rightly or wrongly */
APOSETCETERA = {APOS}|[\u0091\u2018\u201B]
HTHING = [A-Za-z0-9][A-Za-z0-9.,\u00AD]*(-([A-Za-z0-9\u00AD]+|{ACRO}\.))+
REDAUX = {APOS}([msdMSD]|re|ve|ll)
/* For things that will have n't on the end. They can't end in 'n' */
/* \u00AD is soft hyphen */
SWORD = [A-Za-z\u00AD]*[A-MO-Za-mo-z](\u00AD)*
SREDAUX = n{APOSETCETERA}t
/* Tokens you want but already okay: C'mon 'n' '[2-9]0s '[eE]m 'till?
[Yy]'all 'Cause Shi'ite B'Gosh o'clock. Here now only need apostrophe
final words. */
/* Note that Jflex doesn't support {2,} form. Only {2,k}. */
/* [yY]' is for Y'know, y'all and I for I. So exclude from one letter first */
/* Rest are for French borrowings. n allows n'ts in "don'ts" */
/* Arguably, c'mon should be split to "c'm" + "on", but not yet. */
APOWORD = {APOS}n{APOS}?|[lLdDjJ]{APOS}|Dunkin{APOS}|somethin{APOS}|ol{APOS}|{APOS}em|[A-HJ-XZn]{APOSETCETERA}[:letter:]{2}[:letter:]*|{APOS}[2-9]0s|{APOS}till?|[:letter:][:letter:]*[aeiouyAEIOUY]{APOSETCETERA}[aeiouA-Z][:letter:]*|{APOS}cause|cont'd\.?|'twas|nor'easter|c'mon|e'er|s'mores|ev'ry|li'l|nat'l
APOWORD2 = y{APOS}
FULLURL = https?:\/\/[^ \t\n\f\r\"<>|()]+[^ \t\n\f\r\"<>|.!?(){},-]
LIKELYURL = ((www\.([^ \t\n\f\r\"<>|.!?(){},]+\.)+[a-zA-Z]{2,4})|(([^ \t\n\f\r\"`'<>|.!?(){},-_$]+\.)+(com|net|org|edu)))(\/[^ \t\n\f\r\"<>|()]+[^ \t\n\f\r\"<>|.!?(){},-])?
EMAIL = [a-zA-Z0-9][^ \t\n\f\r\"<>|()\u00A0]*@([^ \t\n\f\r\"<>|().\u00A0]+\.)*([^ \t\n\f\r\"<>|().\u00A0]+)
/* Technically, names should be capped at 15 characters. However, then
you get into weirdness with what happens to the rest of the characters. */
TWITTER_NAME = @[a-zA-Z_][a-zA-Z_0-9]*
TWITTER_CATEGORY = #{WORD}
TWITTER = {TWITTER_NAME}|{TWITTER_CATEGORY}
/* Abbreviations - induced from 1987 WSJ by hand */
ABMONTH = Jan|Feb|Mar|Apr|Jun|Jul|Aug|Sep|Sept|Oct|Nov|Dec
/* Jun and Jul barely occur, but don't seem dangerous */
ABDAYS = Mon|Tue|Tues|Wed|Thu|Thurs|Fri
/* Sat. and Sun. barely occur and can easily lead to errors, so we omit them */
/* In caseless, |a\.m|p\.m handled as ACRO, and this is better as can often
be followed by capitalized. */
/* Ma. or Me. isn't included as too many errors, and most sources use Mass. etc. */
/* Fed. is tricky. Usually sentence end, but not before "Governor" or "Natl. Mtg. Assn." */
/* Make some states case sensitive, since they're also reasonably common words */
ABSTATE = Ala|Ariz|[A]rk|Calif|Colo|Conn|Dak|Del|Fla|Ga|[I]ll|Ind|Kans?|Ky|La|[M]ass|Md|Mich|Minn|[M]iss|Mo|Mont|Neb|Nev|Okla|[O]re|Pa|Penn|Tenn|Tex|Va|Vt|[W]ash|Wisc?|Wyo
/* In the caseless world S.p.A. "Società Per Azioni (Italian: shared company)" is got as a regular acronym */
ACRO = [A-Za-z](\.[A-Za-z])+|(Canada|Sino|Korean|EU|Japan|non)-U\.S|U\.S\.-(U\.K|U\.S\.S\.R)
ABTITLE = Mr|Mrs|Ms|[M]iss|Drs?|Profs?|Sens?|Reps?|Attys?|Lt|Col|Gen|Messrs|Govs?|Adm|Rev|Maj|Sgt|Cpl|Pvt|Mt|Capt|Ste?|Ave|Pres|Lieut|Hon|Brig|Co?mdr|Pfc|Spc|Supts?|Det
ABPTIT = Jr|Sr|Bros|(Ed|Ph)\.D|Blvd|Rd|Esq
/* Bhd is Malaysian companies! */
ABCOMP = Inc|Cos?|Corp|Pp?tys?|Ltd|Plc|Bancorp|Dept|Bhd|Assn|Univ|Intl|Sys
ABCOMP2 = Invt|Elec|Natl|M[ft]g
/* Don't included fl. oz. since Oz turns up too much in caseless tokenizer. */
ABNUM = Nos?|Prop|Ph|tel|est|ext|sq|ft
/* p used to be in ABNUM list, but it can't be any more, since the lexer
is now caseless. We don't want to have it recognized for P. Both
p. and P. are now under ABBREV4. ABLIST also went away as no-op [a-e] */
/* est. is "estimated" -- common in some financial contexts. ext. is extension */
/* ABBREV1 abbreviations are normally followed by lower case words. If
they're followed by an uppercase one, we assume there is also a
sentence boundary */
ABBREV1 = ({ABMONTH}|{ABDAYS}|{ABSTATE}|{ABCOMP}|{ABNUM}|{ABPTIT}|etc|al|seq)\.
/* ABRREV2 abbreviations are normally followed by an upper case word. We
assume they aren't used sentence finally */
/* ACRO Is a bad case -- can go either way! */
ABBREV4 = [A-Za-z]|{ABTITLE}|vs|Alex|Wm|Jos|Cie|a\.k\.a|cf|TREAS|{ACRO}|{ABCOMP2}
ABBREV2 = {ABBREV4}\.
ACRONYM = ({ACRO})\.
/* Cie. is used by French companies sometimes before and sometimes at end as in English Co. But we treat as allowed to have Capital following without being sentence end. Cia. is used in Spanish/South American company abbreviations, which come before the company name, but we exclude that and lose, because in a caseless segmenter, it's too confusable with CIA. */
/* in the WSJ Alex. is generally an abbreviation for Alex. Brown, brokers! */
/* Added Wm. for William and Jos. for Joseph */
/* In tables: Mkt. for market Div. for division of company, Chg., Yr.: year */
/* phone numbers. keep multi dots pattern separate, so not confused with decimal numbers. */
PHONE = (\([0-9]{2,3}\)[ \u00A0]?|(\+\+?)?([0-9]{2,4}[\- \u00A0])?[0-9]{2,4}[\- \u00A0])[0-9]{3,4}[\- \u00A0]?[0-9]{3,5}|((\+\+?)?[0-9]{2,4}\.)?[0-9]{2,4}\.[0-9]{3,4}\.[0-9]{3,5}
/* Fake duck feet appear sometimes in WSJ, and aren't likely to be SGML, less than, etc., so group. */
FAKEDUCKFEET = <<|>>
OPBRAC = [<\[]|<
CLBRAC = [>\]]|>
HYPHEN = [-_\u058A\u2010\u2011]
HYPHENS = \-+
LDOTS = \.{3,5}|(\.[ \u00A0]){2,4}\.|[\u0085\u2026]
ATS = @+
UNDS = _+
ASTS = \*+|(\\\*){1,3}
HASHES = #+
FNMARKS = {ATS}|{HASHES}|{UNDS}
INSENTP = [,;:\u3001]
QUOTES = {APOS}|''|[`\u2018\u2019\u201A\u201B\u201C\u201D\u0091\u0092\u0093\u0094\u201E\u201F\u2039\u203A\u00AB\u00BB]{1,2}
DBLQUOT = \"|"
/* Cap'n for captain, c'est for french */
TBSPEC = -(RRB|LRB|RCB|LCB|RSB|LSB)-|C\.D\.s|pro-|anti-|S(&|&)P-500|S(&|&)Ls|Cap{APOS}n|c{APOS}est
TBSPEC2 = {APOS}[0-9][0-9]
/* U+2200-U+2BFF has a lot of the various mathematical, etc. symbol ranges */
MISCSYMBOL = [+%&~\^|\\¦\u00A7¨\u00A9\u00AC\u00AE¯\u00B0-\u00B3\u00B4-\u00BA\u00D7\u00F7\u0387\u05BE\u05C0\u05C3\u05C6\u05F3\u05F4\u0600-\u0603\u0606-\u060A\u060C\u0614\u061B\u061E\u066A\u066D\u0703-\u070D\u07F6\u07F7\u07F8\u0964\u0965\u0E4F\u1FBD\u2016\u2017\u2020-\u2023\u2030-\u2038\u203B\u203E-\u2042\u2044\u207A-\u207F\u208A-\u208E\u2100-\u214F\u2190-\u21FF\u2200-\u2BFF\u3012\u30FB\uFF01-\uFF0F\uFF1A-\uFF20\uFF3B-\uFF40\uFF5B-\uFF65\uFF65]
/* \uFF65 is Halfwidth katakana middle dot; \u30FB is Katakana middle dot */
/* Math and other symbols that stand alone: °²× ∀ */
// Consider this list of bullet chars: 2219, 00b7, 2022, 2024
%%
cannot { yypushback(3) ; return getNext(); }
{SGML} { final String origTxt = yytext();
String txt = origTxt;
if (normalizeSpace) {
txt = txt.replaceAll(" ", "\u00A0"); // change to non-breaking space
}
return getNext(txt, origTxt);
}
{SPMDASH} { if (ptb3Dashes) {
return getNext(ptbmdash, yytext()); }
else {
return getNext();
}
}
{SPAMP} { return getNormalizedAmpNext(); }
{SPPUNC} { return getNext(); }
{WORD}/{REDAUX} { final String origTxt = yytext();
String tmp = removeSoftHyphens(origTxt);
if (americanize) {
tmp = Americanize.americanize(tmp);
}
return getNext(tmp, origTxt);
}
{SWORD}/{SREDAUX} { final String txt = yytext();
return getNext(removeSoftHyphens(txt),
txt); }
{WORD} { final String origTxt = yytext();
String tmp = removeSoftHyphens(origTxt);
if (americanize) {
tmp = Americanize.americanize(tmp);
}
return getNext(tmp, origTxt);
}
{APOWORD} { return getNext(); }
{APOWORD2}/[:letter:] { return getNext(); }
{FULLURL} { String txt = yytext();
if (escapeForwardSlashAsterisk) {
txt = delimit(txt, '/');
txt = delimit(txt, '*');
}
return getNext(txt, yytext()); }
{LIKELYURL} { String txt = yytext();
if (escapeForwardSlashAsterisk) {
txt = delimit(txt, '/');
txt = delimit(txt, '*');
}
return getNext(txt, yytext()); }
{EMAIL} { return getNext(); }
{TWITTER} { return getNext(); }
{REDAUX}/[^A-Za-z] { return handleQuotes(yytext(), false);
}
{SREDAUX} { return handleQuotes(yytext(), false);
}
{DATE} { String txt = yytext();
if (escapeForwardSlashAsterisk) {
txt = delimit(txt, '/');
}
return getNext(txt, yytext());
}
{NUMBER} { return getNext(removeSoftHyphens(yytext()),
yytext()); }
{SUBSUPNUM} { return getNext(); }
{FRAC} { String txt = yytext();
if (escapeForwardSlashAsterisk) {
txt = delimit(txt, '/');
}
if (normalizeSpace) {
txt = txt.replaceAll(" ", "\u00A0"); // change to non-breaking space
}
return getNext(txt, yytext());
}
{FRACSTB3} { String txt = yytext();
if (escapeForwardSlashAsterisk) {
txt = delimit(txt, '/');
}
if (normalizeSpace) {
txt = txt.replaceAll(" ", "\u00A0"); // change to non-breaking space
}
return getNext(txt, yytext());
}
{FRAC2} { return normalizeFractions(yytext()); }
{TBSPEC} { return getNormalizedAmpNext(); }
{THING3} { if (escapeForwardSlashAsterisk) {
return getNext(delimit(yytext(), '/'), yytext());
} else {
return getNext();
}
}
{DOLSIGN} { return getNext(); }
{DOLSIGN2} { if (normalizeCurrency) {
return getNext(normalizeCurrency(yytext()), yytext()); }
else {
return getNext();
}
}
/* Any acronym can be treated as sentence final iff followed by this list of words (pronouns, determiners, and prepositions, etc.). "U.S." is the single big source of errors. Character classes make this rule case sensitive! (This is needed!!) */
{ACRONYM}/({SPACES})([A]bout|[A]ccording|[A]dditionally|[A]fter|[A]n|[A]|[A]s|[A]t|[B]ut|[E]arlier|[H]e|[H]er|[H]ere|[H]owever|[I]f|[I]n|[I]t|[L]ast|[M]any|[M]ore|[M]r\.|[M]s\.|[N]ow|[O]nce|[O]ne|[O]ther|[O]ur|[S]he|[S]ince|[S]o|[S]ome|[S]uch|[T]hat|[T]he|[T]hen|[T]here|[T]hese|[T]hey|[T]his|[W]e|[W]hen|[W]hile|[W]hat|[Y]et|[Y]ou){SPACE} {
String s;
// try to fight around an apparent jflex bug where it gets a space at the token
// end by getting wrong the length of the trailing context.
while (yylength() > 0) {
char last = yycharat(yylength()-1);
if (last == ' ' || last == '\t') {
yypushback(1);
} else {
break;
}
}
if (strictTreebank3) {
yypushback(1); // return a period for next time
s = yytext();
} else {
s = yytext();
yypushback(1); // return a period for next time
}
return getNext(s, yytext());
}
{ABBREV1}/{SENTEND} {
String s;
if (strictTreebank3) {
yypushback(1); // return a period for next time
s = yytext();
} else {
s = yytext();
yypushback(1); // return a period for next time
}
return getNext(s, yytext()); }
{ABBREV1} { return getNext(); }
{ABBREV2} { return getNext(); }
{ABBREV4}/{SPACE} { return getNext(); }
{ACRO}/{SPACENL} { return getNext(); }
{TBSPEC2}/{SPACENL} { return getNext(); }
{WORD}\./{INSENTP} { return getNext(removeSoftHyphens(yytext()),
yytext()); }
{PHONE} { String txt = yytext();
if (normalizeSpace) {
txt = txt.replaceAll(" ", "\u00A0"); // change to non-breaking space
}
if (normalizeParentheses) {
txt = txt.replaceAll("\\(", openparen);
txt = txt.replaceAll("\\)", closeparen);
}
return getNext(txt, yytext());
}
{DBLQUOT}/[A-Za-z0-9$] { return handleQuotes(yytext(), true); }
{DBLQUOT} { return handleQuotes(yytext(), false); }
0x7f { if (invertible) {
prevWordAfter.append(yytext());
} }
{OPBRAC} { if (normalizeOtherBrackets) {
return getNext(openparen, yytext()); }
else {
return getNext();
}
}
{CLBRAC} { if (normalizeOtherBrackets) {
return getNext(closeparen, yytext()); }
else {
return getNext();
}
}
\{ { if (normalizeOtherBrackets) {
return getNext(openbrace, yytext()); }
else {
return getNext();
}
}
\} { if (normalizeOtherBrackets) {
return getNext(closebrace, yytext()); }
else {
return getNext();
}
}
\( { if (normalizeParentheses) {
return getNext(openparen, yytext()); }
else {
return getNext();
}
}
\) { if (normalizeParentheses) {
return getNext(closeparen, yytext()); }
else {
return getNext();
}
}
{HYPHENS} { if (yylength() >= 3 && yylength() <= 4 && ptb3Dashes) {
return getNext(ptbmdash, yytext());
} else {
return getNext();
}
}
{LDOTS} { return handleEllipsis(yytext()); }
{FNMARKS} { return getNext(); }
{ASTS} { if (escapeForwardSlashAsterisk) {
return getNext(delimit(yytext(), '*'), yytext()); }
else {
return getNext();
}
}
{INSENTP} { return getNext(); }
[?!]+ { return getNext(); }
[.¡¿\u037E\u0589\u061F\u06D4\u0700-\u0702\u07FA\u3002] { return getNext(); }
= { return getNext(); }
\/ { if (escapeForwardSlashAsterisk) {
return getNext(delimit(yytext(), '/'), yytext()); }
else {
return getNext();
}
}
/* {HTHING}/[^a-zA-Z0-9.+] { return getNext(removeSoftHyphens(yytext()),
yytext()); } */
{HTHING}\./{INSENTP} { return getNext(removeSoftHyphens(yytext()),
yytext()); }
{HTHING} { return getNext(removeSoftHyphens(yytext()), yytext()); }
{THING}\./{INSENTP} { return getNext(); }
{THING} { return getNext(); }
{THINGA}\./{INSENTP} { return getNormalizedAmpNext(); }
{THINGA} { return getNormalizedAmpNext(); }
'/[A-Za-z][^ \t\n\r\u00A0] { /* invert quote - often but not always right */
return handleQuotes(yytext(), true);
}
{REDAUX} { return handleQuotes(yytext(), false); }
{QUOTES} { return handleQuotes(yytext(), false); }
{FAKEDUCKFEET} { return getNext(); }
{MISCSYMBOL} { return getNext(); }
\0|{SPACES}|[\u200B\u200E-\u200F\uFEFF] { if (invertible) {
prevWordAfter.append(yytext());
}
}
{NEWLINE} { if (tokenizeNLs) {
return getNext(NEWLINE_TOKEN, yytext()); // js: for tokenizing carriage returns
} else if (invertible) {
prevWordAfter.append(yytext());
}
}
{ if (invertible) {
prevWordAfter.append(yytext());
}
}
. { String str = yytext();
int first = str.charAt(0);
String msg = String.format("Untokenizable: %s (U+%s, decimal: %s)", yytext(), Integer.toHexString(first).toUpperCase(), Integer.toString(first));
switch (untokenizable) {
case NONE_DELETE:
if (invertible) {
prevWordAfter.append(str);
}
break;
case FIRST_DELETE:
if (invertible) {
prevWordAfter.append(str);
}
if ( ! this.seenUntokenizableCharacter) {
LOGGER.warning(msg);
this.seenUntokenizableCharacter = true;
}
break;
case ALL_DELETE:
if (invertible) {
prevWordAfter.append(str);
}
LOGGER.warning(msg);
this.seenUntokenizableCharacter = true;
break;
case NONE_KEEP:
return getNext();
case FIRST_KEEP:
if ( ! this.seenUntokenizableCharacter) {
LOGGER.warning(msg);
this.seenUntokenizableCharacter = true;
}
return getNext();
case ALL_KEEP:
LOGGER.warning(msg);
this.seenUntokenizableCharacter = true;
return getNext();
}
}
<> { if (invertible) {
prevWordAfter.append(yytext());
String str = prevWordAfter.toString();
prevWordAfter.setLength(0);
prevWord.set(AfterAnnotation.class, str);
}
return null;
}