#!/usr/bin/env python #Copyright 2007,2008 Sebastian Hagen # This file is part of wordnet_tools. # wordnet_tools is free software; you can redistribute it and/or modify # it under the terms of the GNU General Public License version 2 # as published by the Free Software Foundation # wordnet_tools 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 wordnet_tools; if not, write to the Free Software # Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA 02110-1301 USA # This program requires python >= 2.4. # This program converts wordnet index/data file pairs into dict index/data # files usable by dictd. # This is basically a reimplementation of the wnfilter program by Rik Faith, # which unfortunately doesn't work correctly for wordnet files in the newer # formats. This version of wordnet_structures should parse wordnet 3.0 files # correctly, and create output very similar to what wnfilter would have # written. import datetime from textwrap import TextWrapper try: import fcntl except ImportError: fcntl = None CAT_ADJECTIVE = 0 CAT_ADVERB = 1 CAT_NOUN = 2 CAT_VERB = 3 category_map = { 'n': CAT_NOUN, 'v': CAT_VERB, 'a': CAT_ADJECTIVE, 's': CAT_ADJECTIVE, 'r': CAT_ADVERB } class SynsetWrapperDirect: __slots__ = ('s',) def __init__(self, index, synset_map): self.s = synset_map[index] def synset_get(self): return self.s class SynsetWrapperPointer: __slots__ = ('index', 'map') def __init__(self, index, synset_map): self.index = index self.map = synset_map def synset_get(self): return self.map[self.index] def __repr__(self): return '%s(%s)' % (self.__class__.__name__, (self.index, self.synset_map)) class WordIndex: __slots__ = ('lemma', 'category', 'ptrs', 'synsets', 'tagsense_count') def __init__(self, lemma, category, ptrs, synsets, tagsense_count): self.lemma = lemma self.category = category self.ptrs = ptrs self.synsets = synsets self.tagsense_count = tagsense_count @classmethod def build_from_line(cls, line_data, synset_map, synset_wrap=SynsetWrapperPointer): line_split = line_data.split() lemma = line_split[0].replace('_',' ') category = category_map[line_split[1]] synset_count = int(line_split[2],10) ptr_count = int(line_split[3],10) ptrs = [line_split[i] for i in range(3, 3+ptr_count)] tagsense_count = int(line_split[5 + ptr_count],10) synsets = [synset_wrap(int(line_split[i],10), synset_map) for i in range(6 + ptr_count, 6 + ptr_count + synset_count)] return cls(lemma, category, ptrs, synsets, tagsense_count) @classmethod def build_from_file(cls, f, synset_map, synset_wrap=SynsetWrapperPointer, rv_base=None): if (rv_base is None): rv = {} else: rv = rv_base for line in f: if (line.startswith(' ')): continue wi = cls.build_from_line(line, synset_map, synset_wrap) word = wi.lemma.lower() if not (word in rv): rv[word] = [] rv[word].append(wi) return rv def __repr__(self): return '%s%s' % (self.__class__.__name__, (self.lemma, self.category, self.ptrs, self.synsets, self.tagsense_count)) class WordIndexDictFormatter(WordIndex): category_map_rev = { CAT_NOUN: 'n', CAT_VERB: 'v', CAT_ADJECTIVE: 'adj', CAT_ADVERB: 'adv' } linesep = '\n' LINE_WIDTH_MAX = 68 prefix_fmtf_line_first = ' %s 1: ' prefix_fmtn_line_first = ' %*s' prefix_fmtf_line_nonfirst = ' %d: ' prefix_fmtn_line_nonfirst = ' %*s ' def dict_str(self): """Build a human-readable definition for this word, including data for each synset""" tw = TextWrapper(width=self.LINE_WIDTH_MAX, initial_indent=(self.prefix_fmtf_line_first % self.category_map_rev[self.category]), subsequent_indent=(self.prefix_fmtn_line_first % (len(self.category_map_rev[self.category]), ''))) lines = (tw.wrap(self.synsets[0].synset_get().dict_str())) i = 2 prefix_fmtn_line_nonfirst = self.prefix_fmtn_line_nonfirst pfln_len = 0 for ss_wrap in self.synsets[1:]: # adjust indenting based on index-number with pfln_len_new = len('%d' % (i,)) if (pfln_len_new > pfln_len): pfln_len = pfln_len_new pfln_str = (self.prefix_fmtn_line_nonfirst % (pfln_len, '')) # format data for this synset synset = ss_wrap.synset_get() tw = TextWrapper(width=self.LINE_WIDTH_MAX, initial_indent=(self.prefix_fmtf_line_nonfirst % i), subsequent_indent=pfln_str) lines.extend(tw.wrap(synset.dict_str())) i += 1 return self.linesep.join(lines) class Synset: __slots__ = ('offset', 'type', 'words', 'ptrs', 'gloss', 'frames', 'synset_map') def __init__(self, offset, ss_type, words, ptrs, gloss, frames=(), synset_map=None): self.offset = offset self.type = ss_type self.words = words self.ptrs = ptrs self.gloss = gloss self.frames = frames self.synset_map = synset_map @classmethod def build_from_line(cls, line_data, synset_map=None): """Build instance from provided line from a wordnet data file. The optional synset_map argument will be passed to the instance constructor.""" line_split = line_data.split() synset_offset = int(line_split[0],10) ss_type = category_map[line_split[2]] word_count = int(line_split[3],16) words = [line_split[i].replace('_',' ') for i in range(4, 4 + word_count*2,2)] ptr_count = int(line_split[4 + word_count*2],10) ptrs = [(line_split[i], # pointer symbol int(line_split[i+1],10), # synset offset category_map[line_split[i+2]], # synset type (int(line_split[i+3][:2],16), int(line_split[i+3][2:],16))) # offsets of specifically related words for i in range(5 + word_count*2,4 + word_count*2 + ptr_count*4,4)] tok = line_split[5 + word_count*2 + ptr_count*4] base = 6 + word_count*2 + ptr_count*4 if (tok != '|'): frame_count = int(tok, 10) frames = [(int(line_split[i+1],10), int(line_split[i+2],16)) for i in range(base, base + frame_count*3, 3)] base += frame_count*3 + 1 else: frames = [] line_split2 = line_data.split(None, base) if (len(line_split2) < base): gloss = None else: gloss = line_split2[-1] return cls(synset_offset, ss_type, words, ptrs, gloss, frames, synset_map) @classmethod def build_from_file(cls, f): rv = {} comments = [] for line in f: if (line.startswith(' ')): line_s = line.lstrip().rstrip('\n') line_elements = line_s.split(None,1) try: int(line_elements[0]) except ValueError: continue if (len(line_elements) == 1): line_elements.append('') comments.append(line_elements[1]) continue synset = cls.build_from_line(line.rstrip(), rv) rv[synset.offset] = synset return (rv, comments) def dict_str(self): """Format this synset into a human-readable line-wrapped dict block.""" rv = self.gloss.rstrip() if (len(self.words) > 1): rv += ' [syn: %s]' % (', '.join([('{%s}' % word) for word in self.words])) if not (self.synset_map is None): antonyms = [] for (ptr_symbol, synset_offset, synset_type, sourcetarget) in self.ptrs: if ((synset_type == self.type) and (ptr_symbol == '!')): ss = self.synset_map[synset_offset] antonyms.extend(ss.words) antonyms = list(set(antonyms)) # remove any duplicate entries antonyms.sort() if (antonyms): rv += ' [ant: %s]' % (', '.join([('{%s}' % word) for word in antonyms])) return rv def __repr__(self): return '%s%s' % (self.__class__.__name__, (self.offset, self.type, self.words, self.ptrs, self.gloss, self.frames, self.synset_map)) class SynsetFileBackedMapping: """Access a wordnet synset file as a mapping""" def __init__(self, f, file_lock=True, lock_block=False): self.file = f if (file_lock): if (fcntl is None): raise EnvironmentError('Unable to perform file locking without fcntl module') lock_op = fcntl.LOCK_SH if not (lock_block): lock_op |= fcntl.LOCK_NB fcntl.lockf(f.fileno(), lock_op) def __getitem__(self, key): self.file.seek(key) return Synset.build_from_line(self.file.readline().rstrip('\n'), self) def initial_comments_get(self): self.file.seek(0) comments = [] for line in self.file: if not (line.startswith(' ')): break line_s = line.lstrip().rstrip('\n') line_elements = line_s.split(None,1) try: int(line_elements[0]) except ValueError: continue if (len(line_elements) == 1): line_elements.append('') comments.append(line_elements[1]) continue return comments class WordnetDict: db_info_fmt = '''This file was converted from the original database on: %(conversion_datetime)s The original data is available from: %(wn_url)s The original data was distributed with the notice shown below. No additional restrictions are claimed. Please redistribute this changed version under the same conditions and restriction that apply to the original version.\n\n %(wn_license)s''' datetime_fmt = '%Y-%m-%dT%H:%M:%S' base64_map = 'ABCDEFGHIJKLMNOPQRSTUVWXYZabcdefghijklmnopqrstuvwxyz0123456789+/' def __init__(self, wn_url, desc_short, desc_long): self.word_data = {} self.wn_url = wn_url self.desc_short = desc_short self.desc_long = desc_long self.wn_license = None def wn_dict_add(self, file_index, file_data, synset_wrap_cls=SynsetWrapperDirect): file_data.seek(0) file_index.seek(0) synsets = SynsetFileBackedMapping(file_data) license_lines = synsets.initial_comments_get() WordIndexDictFormatter.build_from_file(file_index, synsets, synset_wrap=synset_wrap_cls, rv_base=self.word_data) if (license_lines): self.wn_license = '\n'.join(license_lines) + '\n' @classmethod def base64_encode(cls, i): """Encode a non-negative integer into a dictd compatible base64 string""" if (i < 0): raise ValueError('Value %r for i is negative' % (i,)) r = 63 e = 1 while (r < i): e += 1 r = 64**e - 1 rv = '' while (e > 0): e -= 1 d = (i / 64**e) rv += cls.base64_map[d] i = i % (64**e) return rv @classmethod def dict_entry_write(cls, file_index, file_data, key, entry, linesep='\n'): """Write a single dict entry for <key> to index and data files""" entry_start = file_data.tell() file_data.write(entry) entry_len = len(entry) file_index.write('%s\t%s\t%s%s' % (key, cls.base64_encode(entry_start), cls.base64_encode(entry_len), linesep)) def dict_generate(self, file_index, file_data): file_index.seek(0) file_data.seek(0) # The dictd file format is fairly iffy on the subject of special # headwords: either dictd is buggy, or the manpage doesn't tell the whole # story about the format. # The upshot is that order of these entries in the index *matters*. # Putting them at the beginning and in alphabetic order is afaict ok. # Some other orders completely and quietly break the ability to look # those headwords up. # -- problem encountered with 1.10.2, at 2007-08-05. file_data.write('\n') wn_url = self.wn_url conversion_datetime = datetime.datetime.now().strftime(self.datetime_fmt) wn_license = self.wn_license word_data = self.word_data.copy() for (key, value) in ( ('00-database-allchars', '\n'), ('00-database-info', '00-database-info\n%s\n' % (self.db_info_fmt % vars())), ('00-database-long', '00-database-long\n%s\n' % self.desc_long), ('00-database-short', '00-database-short\n%s\n' % self.desc_short), ('00-database-url', '00-database-url\n%s\n' % self.wn_url)): word_data[key] = value words = word_data.keys() words.sort() for word in words: wval = word_data[word] if (isinstance(wval, basestring)): # special case: preformatted entries self.dict_entry_write(file_index, file_data, word, wval) continue for wi in wval: word_cs = word # Use case-sensitivity information of first entry of first synset that # matches this word case-insensitively for ss_wrapper in wi.synsets: synset = ss_wrapper.synset_get() for ss_word in synset.words: if (ss_word.lower() == word_cs.lower()): word_cs = ss_word break else: continue break else: continue break outstr = '' for wi in wval: outstr += wi.dict_str() + '\n' outstr = '%s%s%s' % (word_cs, wi.linesep, outstr) self.dict_entry_write(file_index, file_data, word, outstr, wi.linesep) file_index.truncate() file_data.truncate() if (__name__ == '__main__'): import optparse op = optparse.OptionParser(usage='usage: %prog [options] (<wn_index_file> <wn_data_file>)+') op.add_option('-i', '--outindex', dest='oi', default='wn.index', help='filename of index file to write to') op.add_option('-d', '--outdata', dest='od', default='wn.dict', help='filename of data file to write to') op.add_option('--wn_url', dest='wn_url', default='ftp://ftp.cogsci.princeton.edu/pub/wordnet/2.0', help='URL for wordnet sources') op.add_option('--db_desc_short', dest='desc_short', default=' WordNet (r) 3.0 (2006)', help='short dict DB description') op.add_option('--db_desc_long', dest='desc_long', default=' WordNet (r): A Lexical Database for English from the\n Cognitive Science Laboratory at Princeton University', help='long dict DB description') (options, args) = op.parse_args() wnd = WordnetDict(wn_url=options.wn_url, desc_short=options.desc_short, desc_long=options.desc_long) for i in range(0,len(args),2): print 'Opening index file %r...' % args[i] file_index = file(args[i]) print 'Opening data file %r...' % args[i+1] file_data = file(args[i+1]) print 'Parsing index file...' wnd.wn_dict_add(file_index, file_data, synset_wrap_cls=SynsetWrapperPointer) print 'All index files parsed. Parsing data input files and writing output to index file %r and data file %r.' % (options.oi, options.od) wnd.dict_generate(file(options.oi, 'w'),file(options.od, 'w')) print 'All done.'