Welcome to mirror list, hosted at ThFree Co, Russian Federation.

ExtractGHKM.cpp « extract-ghkm « phrase-extract « training « scripts - github.com/moses-smt/mosesdecoder.git - Unnamed repository; edit this file 'description' to name the repository.
summaryrefslogtreecommitdiff
blob: dad326131158c714c137130d254bba2a31b21ac2 (plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
/***********************************************************************
 Moses - statistical machine translation system
 Copyright (C) 2006-2011 University of Edinburgh
 
 This library is free software; you can redistribute it and/or
 modify it under the terms of the GNU Lesser General Public
 License as published by the Free Software Foundation; either
 version 2.1 of the License, or (at your option) any later version.
 
 This library 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
 Lesser General Public License for more details.
 
 You should have received a copy of the GNU Lesser General Public
 License along with this library; if not, write to the Free Software
 Foundation, Inc., 51 Franklin Street, Fifth Floor, Boston, MA  02110-1301  USA
***********************************************************************/

#include "ExtractGHKM.h"

#include "Alignment.h"
#include "AlignmentGraph.h"
#include "Exception.h"
#include "Node.h"
#include "Options.h"
#include "ParseTree.h"
#include "ScfgRule.h"
#include "ScfgRuleWriter.h"
#include "Span.h"
#include "XmlTreeParser.h"

#include <boost/program_options.hpp>

#include <cassert>
#include <cstdlib>
#include <fstream>
#include <iostream>
#include <iterator>
#include <string>
#include <sstream>
#include <vector>

namespace Moses {
namespace GHKM {

int ExtractGHKM::Main(int argc, char *argv[])
{
  // Process command-line options.
  Options options;
  ProcessOptions(argc, argv, options);

  // Open input files.
  std::ifstream targetStream;
  std::ifstream sourceStream;
  std::ifstream alignmentStream;
  OpenInputFileOrDie(options.targetFile, targetStream);
  OpenInputFileOrDie(options.sourceFile, sourceStream);
  OpenInputFileOrDie(options.alignmentFile, alignmentStream);

  // Open output files.
  std::ofstream extractStream;
  std::ofstream invExtractStream;
  std::ofstream glueGrammarStream;
  std::ofstream unknownWordStream;
  std::string invExtractFileName = options.extractFile + std::string(".inv");
  OpenOutputFileOrDie(options.extractFile, extractStream);
  OpenOutputFileOrDie(invExtractFileName, invExtractStream);
  if (!options.glueGrammarFile.empty()) {
    OpenOutputFileOrDie(options.glueGrammarFile, glueGrammarStream);
  }
  if (!options.unknownWordFile.empty()) {
    OpenOutputFileOrDie(options.unknownWordFile, unknownWordStream);
  }

  // Target label sets for producing glue grammar.
  std::set<std::string> labelSet;
  std::set<std::string> topLabelSet;

  // Word count statistics for producing unknown word labels.
  std::map<std::string, int> wordCount;
  std::map<std::string, std::string> wordLabel;

  std::string targetLine;
  std::string sourceLine;
  std::string alignmentLine;
  ScfgRuleWriter writer(extractStream, invExtractStream, options);
  size_t lineNum = 0;
  while (true) {
    std::getline(targetStream, targetLine);
    std::getline(sourceStream, sourceLine);
    std::getline(alignmentStream, alignmentLine);

    if (targetStream.eof() && sourceStream.eof() && alignmentStream.eof()) {
      break;
    }

    if (targetStream.eof() || sourceStream.eof() || alignmentStream.eof()) {
      Error("Files must contain same number of lines");
    }

    ++lineNum;

    // Parse target tree.
    std::auto_ptr<ParseTree> t(ParseXmlTree(targetLine));
    if (!t.get()) {
      std::ostringstream s;
      s << "Failed to parse XML tree at line " << lineNum;
      Error(s.str());
    }

    // Read source tokens.
    std::vector<std::string> sourceTokens(ReadTokens(sourceLine));

    // Read word alignments.
    Alignment alignment;
    try {
      alignment = ReadAlignment(alignmentLine);
    } catch (const Exception &e) {
      std::ostringstream s;
      s << "Failed to read alignment at line " << lineNum << ": ";
      s << e.GetMsg();
      Error(s.str());
    }

    // Record tree labels for use in glue grammar.
    if (!options.glueGrammarFile.empty()) {
      // Record labels that cover the full sentence to topLabelSet.
      ParseTree *p = t.get();
      topLabelSet.insert(p->GetLabel());
      while (p->GetChildren().size() == 1) {
        p = p->GetChildren()[0];
        if (p->IsLeaf()) {
          break;
        }
        topLabelSet.insert(p->GetLabel());
      }
      // Record all labels to labelSet.
      RecordTreeLabels(*t, labelSet);
    }

    // Record word counts.
    if (!options.unknownWordFile.empty()) {
      CollectWordLabelCounts(*t, wordCount, wordLabel);
    }

    // Form an alignment graph from the target tree, source words, and
    // alignment.
    AlignmentGraph graph(t.get(), sourceTokens, alignment);

    // Extract minimal rules, adding each rule to its root node's rule set.
    graph.ExtractMinimalRules(options);

    // Extract composed rules.
    if (!options.minimal) {
      graph.ExtractComposedRules(options);
    }

    // Write the rules, subject to scope pruning.
    const std::vector<Node *> &targetNodes = graph.GetTargetNodes();
    for (std::vector<Node *>::const_iterator p = targetNodes.begin();
         p != targetNodes.end(); ++p) {
      const std::vector<const Subgraph *> &rules = (*p)->GetRules();
      for (std::vector<const Subgraph *>::const_iterator q = rules.begin();
           q != rules.end(); ++q) {
        ScfgRule r(**q);
        // TODO Can scope pruning be done earlier?
        if (r.Scope() <= options.maxScope) {
          writer.Write(r);
        }
      }
    }
  }

  if (!options.glueGrammarFile.empty()) {
    WriteGlueGrammar(labelSet, topLabelSet, glueGrammarStream);
  }

  if (!options.unknownWordFile.empty()) {
    WriteUnknownWordLabel(wordCount, wordLabel, unknownWordStream);
  }

  return 0;
}

void ExtractGHKM::OpenInputFileOrDie(const std::string &filename,
                                     std::ifstream &stream)
{
  stream.open(filename.c_str());
  if (!stream) {
    std::ostringstream msg;
    msg << "failed to open input file: " << filename;
    Error(msg.str());
  }
}

void ExtractGHKM::OpenOutputFileOrDie(const std::string &filename,
                                      std::ofstream &stream)
{
  stream.open(filename.c_str());
  if (!stream) {
    std::ostringstream msg;
    msg << "failed to open output file: " << filename;
    Error(msg.str());
  }
}

void ExtractGHKM::ProcessOptions(int argc, char *argv[],
                                 Options &options) const
{
  namespace po = boost::program_options;
  namespace cls = boost::program_options::command_line_style;

  // Construct the 'top' of the usage message: the bit that comes before the
  // options list.
  std::ostringstream usageTop;
  usageTop << "Usage: " << GetName()
           << " [OPTION]... TARGET SOURCE ALIGNMENT EXTRACT\n\n"
           << "SCFG rule extractor based on the GHKM algorithm described in\n"
           << "Galley et al. (2004).\n\n"
           << "Options";

  // Construct the 'bottom' of the usage message.
  std::ostringstream usageBottom;
  usageBottom << "\nImplementation Notes:\n"
              << "\nThe parse tree is assumed to contain part-of-speech preterminal nodes.\n"
              << "\n"
              << "For the composed rule constraints: rule depth is the maximum distance from the\nrule's root node to a sink node, not counting preterminal expansions or word\nalignments.  Rule size is the measure defined in DeNeefe et al (2007): the\nnumber of non-part-of-speech, non-leaf constituent labels in the target tree.\nNode count is the number of target tree nodes (excluding target words).\n"
              << "\n"
              << "Scope pruning (Hopkins and Langmead, 2010) is applied to both minimal and\ncomposed rules.\n"
              << "\n"
              << "Unaligned source words are attached to the tree using the following heuristic:\nif there are aligned source words to both the left and the right of an unaligned\nsource word then it is attached to the lowest common ancestor of its nearest\nsuch left and right neighbours.  Otherwise, it is attached to the root of the\nparse tree.\n"
              << "\n"
              << "Unless the --AllowUnary option is given, unary rules containing no lexical\nsource items are eliminated using the method described in Chung et al. (2011).\nThe parsing algorithm used in Moses is unable to handle such rules.\n"
              << "\n"
              << "References:\n"
              << "Galley, M., Hopkins, M., Knight, K., and Marcu, D. (2004)\n"
              << "\"What's in a Translation Rule?\", In Proceedings of HLT/NAACL 2004.\n"
              << "\n"
              << "DeNeefe, S., Knight, K., Wang, W., and Marcu, D. (2007)\n"
              << "\"What Can Syntax-Based MT Learn from Phrase-Based MT?\", In Proceedings of\nEMNLP-CoNLL 2007.\n"
              << "\n"
              << "Hopkins, M. and Langmead, G. (2010)\n"
              << "\"SCFG Decoding Without Binarization\", In Proceedings of EMNLP 2010.\n"
              << "\n"
              << "Chung, T. and Fang, L. and Gildea, D. (2011)\n"
              << "\"Issues Concerning Decoding with Synchronous Context-free Grammar\", In\nProceedings of ACL/HLT 2011.";

  // Declare the command line options that are visible to the user.
  po::options_description visible(usageTop.str());
  visible.add_options()
    //("help", "print this help message and exit")
    ("AllowUnary",
        "allow fully non-lexical unary rules")
    ("GlueGrammar",
        po::value(&options.glueGrammarFile),
        "write glue grammar to named file")
    ("MaxNodes",
        po::value(&options.maxNodes)->default_value(options.maxNodes),
        "set maximum number of tree nodes for composed rules")
    ("MaxRuleDepth",
        po::value(&options.maxRuleDepth)->default_value(options.maxRuleDepth),
        "set maximum depth for composed rules")
    ("MaxRuleSize",
        po::value(&options.maxRuleSize)->default_value(options.maxRuleSize),
        "set maximum size for composed rules")
    ("MaxScope",
        po::value(&options.maxScope)->default_value(options.maxScope),
        "set maximum allowed scope")
    ("Minimal",
        "extract minimal rules only")
    ("UnknownWordLabel",
        po::value(&options.unknownWordFile),
        "write unknown word labels to named file")
    ("UnpairedExtractFormat",
        "do not pair non-terminals in extract files")
  ;

  // Declare the command line options that are hidden from the user
  // (these are used as positional options).
  po::options_description hidden("Hidden options");
  hidden.add_options()
    ("TargetFile",
        po::value(&options.targetFile),
        "target file")
    ("SourceFile",
        po::value(&options.sourceFile),
        "source file")
    ("AlignmentFile",
        po::value(&options.alignmentFile),
        "alignment file")
    ("ExtractFile",
        po::value(&options.extractFile),
        "extract file")
  ;

  // Compose the full set of command-line options.
  po::options_description cmdLineOptions;
  cmdLineOptions.add(visible).add(hidden);

  // Register the positional options.
  po::positional_options_description p;
  p.add("TargetFile", 1);
  p.add("SourceFile", 1);
  p.add("AlignmentFile", 1);
  p.add("ExtractFile", 1);

  // Process the command-line.
  po::variables_map vm;
  const int optionStyle = cls::allow_long
                        | cls::long_allow_adjacent
                        | cls::long_allow_next;
  try {
    po::store(po::command_line_parser(argc, argv).style(optionStyle).
              options(cmdLineOptions).positional(p).run(), vm);
    po::notify(vm);
  } catch (const std::exception &e) {
    std::ostringstream msg;
    msg << e.what() << "\n\n" << visible << usageBottom.str();
    Error(msg.str());
  }

  if (vm.count("help")) {
    std::cout << visible << usageBottom.str() << std::endl;
    std::exit(0);
  }

  // Check all positional options were given.
  if (!vm.count("TargetFile") ||
      !vm.count("SourceFile") ||
      !vm.count("AlignmentFile") ||
      !vm.count("ExtractFile")) {
    std::ostringstream msg;
    std::cerr << visible << usageBottom.str() << std::endl;
    std::exit(1);
  }

  // Process Boolean options.
  if (vm.count("AllowUnary")) {
    options.allowUnary = true;
  }
  if (vm.count("Minimal")) {
    options.minimal = true;
  }
  if (vm.count("UnpairedExtractFormat")) {
    options.unpairedExtractFormat = true;
  }
}

void ExtractGHKM::Error(const std::string &msg) const
{
  std::cerr << GetName() << ": " << msg << std::endl;
  std::exit(1);
}

std::vector<std::string> ExtractGHKM::ReadTokens(const std::string &s)
{
  std::vector<std::string> tokens;

  std::string whitespace = " \t";

  std::string::size_type begin = s.find_first_not_of(whitespace);
  assert(begin != std::string::npos);
  while (true) {
    std::string::size_type end = s.find_first_of(whitespace, begin);
    std::string token;
    if (end == std::string::npos) {
      token = s.substr(begin);
    } else {
      token = s.substr(begin, end-begin);
    }
    tokens.push_back(token);
    if (end == std::string::npos) {
      break;
    }
    begin = s.find_first_not_of(whitespace, end);
    if (begin == std::string::npos) {
      break;
    }
  }

  return tokens;
}

void ExtractGHKM::WriteGlueGrammar(const std::set<std::string> &labelSet,
                                   const std::set<std::string> &topLabelSet,
                                   std::ostream &out)
{
  // chose a top label that is not already a label
  std::string topLabel = "QQQQQQ";
  for(int i = 1; i <= topLabel.length(); i++) {
    if (labelSet.find(topLabel.substr(0,i)) == labelSet.end() ) {
      topLabel = topLabel.substr(0,i);
      break;
    }
  }

  // basic rules
  out << "<s> [X] ||| <s> [" << topLabel << "] ||| 1  ||| " << std::endl;
  out << "[X][" << topLabel << "] </s> [X] ||| [X][" << topLabel << "] </s> [" << topLabel << "] ||| 1 ||| 0-0 " << std::endl;

  // top rules
  for (std::set<std::string>::const_iterator i = topLabelSet.begin();
       i != topLabelSet.end(); ++i) {
    out << "<s> [X][" << *i << "] </s> [X] ||| <s> [X][" << *i << "] </s> [" << topLabel << "] ||| 1 ||| 1-1" << std::endl;
  }

  // glue rules
  for(std::set<std::string>::const_iterator i = labelSet.begin();
      i != labelSet.end(); i++ ) {
    out << "[X][" << topLabel << "] [X][" << *i << "] [X] ||| [X][" << topLabel << "] [X][" << *i << "] [" << topLabel << "] ||| 2.718 ||| 0-0 1-1" << std::endl;
  }
  // glue rule for unknown word...
  out << "[X][" << topLabel << "] [X][X] [X] ||| [X][" << topLabel << "] [X][X] [" << topLabel << "] ||| 2.718 |||  0-0 1-1 " << std::endl;
}

void ExtractGHKM::RecordTreeLabels(const ParseTree &t,
                                   std::set<std::string> &labelSet)
{
  labelSet.insert(t.GetLabel());
  const std::vector<ParseTree *> &children = t.GetChildren();
  for (std::vector<ParseTree *>::const_iterator p = children.begin();
       p != children.end(); ++p) {
    const ParseTree &child = **p;
    if (!child.IsLeaf()) {
      RecordTreeLabels(child, labelSet);
    }
  }
}

void ExtractGHKM::CollectWordLabelCounts(
    ParseTree &root,
    std::map<std::string, int> &wordCount,
    std::map<std::string, std::string> &wordLabel)
{
  std::vector<const ParseTree*> leaves;
  root.GetLeaves(std::back_inserter(leaves));
  for (std::vector<const ParseTree *>::const_iterator p = leaves.begin();
       p != leaves.end(); ++p) {
    const ParseTree &leaf = **p;
    const std::string &word = leaf.GetLabel();
    const std::string &label = leaf.GetParent()->GetLabel();
    ++wordCount[word];
    wordLabel[word] = label;
  }
}

void ExtractGHKM::WriteUnknownWordLabel(
    const std::map<std::string, int> &wordCount,
    const std::map<std::string, std::string> &wordLabel,
    std::ostream &out)
{
  std::map<std::string, int> labelCount;
  int total = 0;
  for (std::map<std::string, int>::const_iterator p = wordCount.begin();
       p != wordCount.end(); ++p) {
    // Only consider singletons.
    if (p->second == 1) {
      std::map<std::string, std::string>::const_iterator q =
        wordLabel.find(p->first);
      assert(q != wordLabel.end());
      ++labelCount[q->second];
      ++total;
    }
  }
  for (std::map<std::string, int>::const_iterator p = labelCount.begin();
       p != labelCount.end(); ++p) {
    double ratio = static_cast<double>(p->second) / static_cast<double>(total);
    if (ratio > 0.03) {
      out << p->first << " " << ratio << std::endl;
    }
  }
}

}  // namespace GHKM
}  // namespace Moses