Codebase list mozc / debian/2.26.4220.100+dfsg-3 src / rewriter / collocation_rewriter.cc
debian/2.26.4220.100+dfsg-3

Tree @debian/2.26.4220.100+dfsg-3 (Download .tar.gz)

collocation_rewriter.cc @debian/2.26.4220.100+dfsg-3raw · history · blame

  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
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
// Copyright 2010-2020, Google Inc.
// All rights reserved.
//
// Redistribution and use in source and binary forms, with or without
// modification, are permitted provided that the following conditions are
// met:
//
//     * Redistributions of source code must retain the above copyright
// notice, this list of conditions and the following disclaimer.
//     * Redistributions in binary form must reproduce the above
// copyright notice, this list of conditions and the following disclaimer
// in the documentation and/or other materials provided with the
// distribution.
//     * Neither the name of Google Inc. nor the names of its
// contributors may be used to endorse or promote products derived from
// this software without specific prior written permission.
//
// THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
// "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
// LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
// A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT
// OWNER OR CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL,
// SPECIAL, EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT
// LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE,
// DATA, OR PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY
// THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT
// (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE
// OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.

#include "rewriter/collocation_rewriter.h"

#include <algorithm>
#include <memory>
#include <string>
#include <vector>

#include "base/flags.h"
#include "base/hash.h"
#include "base/logging.h"
#include "base/util.h"
#include "converter/segments.h"
#include "data_manager/data_manager_interface.h"
#include "dictionary/pos_matcher.h"
#include "request/conversion_request.h"
#include "rewriter/collocation_util.h"
#include "storage/existence_filter.h"
#include "absl/strings/str_cat.h"
#include "absl/strings/string_view.h"

DEFINE_bool(use_collocation, true, "use collocation rewrite");

namespace mozc {

using mozc::storage::ExistenceFilter;

namespace {
const size_t kCandidateSize = 12;
const int kMaxCostDiff = 3453;  // -500*log(1/1000)

// For collocation, we use two segments.
enum SegmentLookupType {
  LEFT,
  RIGHT,
};

// returns true if the given string contains number including Kanji.
bool ContainsNumber(const std::string &str) {
  for (ConstChar32Iterator iter(str); !iter.Done(); iter.Next()) {
    if (CollocationUtil::IsNumber(iter.Get())) {
      return true;
    }
  }
  return false;
}

// Returns true if value matches the pattern XXXPPPYYY, where XXX is a Kanji
// sequence, PPP is the given pattern, and YYY is a sequence containing at least
// one Kanji character. In the value matches the pattern, XXX and YYY are
// substituted to |first_content| and |second|, respectively. Returns false if
// the value isn't of the form XXXPPPYYY.
bool ParseCompound(const absl::string_view value,
                   const absl::string_view pattern,
                   absl::string_view *first_content,
                   absl::string_view *second) {
  DCHECK(!value.empty());
  DCHECK(!pattern.empty());

  // Find the |first_content| candidate and check if it consists of Kanji only.
  absl::string_view::const_iterator pattern_begin =
      std::find(value.begin(), value.end(), pattern[0]);
  if (pattern_begin == value.end()) {
    return false;
  }
  *first_content = absl::string_view(
      value.data(), std::distance(value.begin(), pattern_begin));
  if (!Util::IsScriptType(*first_content, Util::KANJI)) {
    return false;
  }

  // Check if the middle part matches |pattern|.
  const absl::string_view remaining_value =
      absl::ClippedSubstr(value, first_content->size());
  if (!Util::StartsWith(remaining_value, pattern)) {
    return false;
  }

  // Check if the last substring is eligible for |second|.
  *second = absl::ClippedSubstr(remaining_value, pattern.size());
  if (second->empty() || !Util::ContainsScriptType(*second, Util::KANJI)) {
    return false;
  }

  // Just verify that |value| = |first_content| + |pattern| + |second|.
  DCHECK_EQ(value, std::string(*first_content) + std::string(pattern) +
                       std::string(*second));
  return true;
}

// A helper function to push back a string view to a vector.
inline void PushBackStringView(const absl::string_view s,
                               std::vector<std::string> *v) {
  v->emplace_back(s.data(), s.size());
}

// Handles compound such as "本を読む"(one segment)
// we want to rewrite using it as if it was "<本|を><読む>"
// so that we can use collocation data like "厚い本"
void ResolveCompoundSegment(const std::string &top_value,
                            const std::string &value, SegmentLookupType type,
                            std::vector<std::string> *output) {
  // see "http://ja.wikipedia.org/wiki/助詞"
  static const char kPat1[] = "が";
  // "の" was not good...
  // static const char kPat2[] = "の";
  static const char kPat3[] = "を";
  static const char kPat4[] = "に";
  static const char kPat5[] = "へ";
  static const char kPat6[] = "と";
  static const char kPat7[] = "から";
  static const char kPat8[] = "より";
  static const char kPat9[] = "で";

  static const struct {
    const char *pat;
    size_t len;
  } kParticles[] = {{kPat1, arraysize(kPat1) - 1},
                    //    {kPat2, arraysize(kPat2) - 1},
                    {kPat3, arraysize(kPat3) - 1},
                    {kPat4, arraysize(kPat4) - 1},
                    {kPat5, arraysize(kPat5) - 1},
                    {kPat6, arraysize(kPat6) - 1},
                    {kPat7, arraysize(kPat7) - 1},
                    {kPat8, arraysize(kPat8) - 1},
                    {kPat9, arraysize(kPat9) - 1},
                    {nullptr, 0}};

  for (size_t i = 0; kParticles[i].pat != nullptr; ++i) {
    const absl::string_view particle(kParticles[i].pat, kParticles[i].len);
    absl::string_view first_content, second;
    if (!ParseCompound(top_value, particle, &first_content, &second)) {
      continue;
    }
    if (ParseCompound(value, particle, &first_content, &second)) {
      if (type == LEFT) {
        output->emplace_back(second.data(), second.size());
        output->push_back(absl::StrCat(first_content, particle));
      } else {
        output->emplace_back(first_content.data(), first_content.size());
      }
      return;
    }
  }
}

bool IsNaturalContent(const Segment::Candidate &cand,
                      const Segment::Candidate &top_cand,
                      SegmentLookupType type,
                      std::vector<std::string> *output) {
  const std::string &content = cand.content_value;
  const std::string &value = cand.value;
  const std::string &top_content = top_cand.content_value;
  const std::string &top_value = top_cand.value;

  const size_t top_content_len = Util::CharsLen(top_content);
  const size_t content_len = Util::CharsLen(content);

  if (type == RIGHT && value != top_value && top_content_len >= 2 &&
      content_len == 1) {
    return false;
  }

  if (type == LEFT) {
    output->push_back(value);
  } else {
    output->push_back(content);
    // "舞って" workaround
    // V+"て" is often treated as one compound.
    static const char kPat[] = "て";
    if (Util::EndsWith(content, absl::string_view(kPat, arraysize(kPat) - 1))) {
      PushBackStringView(Util::Utf8SubString(content, 0, content_len - 1),
                         output);
    }
  }

  // we don't rewrite NUMBER to others and vice versa
  if (ContainsNumber(value) != ContainsNumber(top_value)) {
    return false;
  }

  const absl::string_view top_aux_value =
      Util::Utf8SubString(top_value, top_content_len, std::string::npos);
  const size_t top_aux_value_len = Util::CharsLen(top_aux_value);
  const Util::ScriptType top_value_script_type = Util::GetScriptType(top_value);

  // we don't rewrite KATAKANA segment
  // for example, we don't rewrite "コーヒー飲みます" to "珈琲飲みます"
  if (type == LEFT && top_aux_value_len == 0 && top_value != value &&
      top_value_script_type == Util::KATAKANA) {
    return false;
  }

  // special cases
  if (top_content_len == 1) {
    const char *begin = top_content.data();
    const char *end = top_content.data() + top_content.size();
    size_t mblen = 0;
    const char32 wchar = Util::UTF8ToUCS4(begin, end, &mblen);

    switch (wchar) {
      case 0x304a:  // "お"
      case 0x5fa1:  // "御"
      case 0x3054:  // "ご"
        return true;
      default:
        break;
    }
  }

  const absl::string_view aux_value =
      Util::Utf8SubString(value, content_len, std::string::npos);

  // Remove number in normalization for the left segment.
  std::string aux_normalized, top_aux_normalized;
  CollocationUtil::GetNormalizedScript(aux_value, (type == LEFT),
                                       &aux_normalized);
  CollocationUtil::GetNormalizedScript(top_aux_value, (type == LEFT),
                                       &top_aux_normalized);
  if (!aux_normalized.empty() &&
      !Util::IsScriptType(aux_normalized, Util::HIRAGANA)) {
    if (type == RIGHT) {
      return false;
    }
    if (aux_normalized != top_aux_normalized) {
      return false;
    }
  }

  ResolveCompoundSegment(top_value, value, type, output);

  const size_t aux_value_len = Util::CharsLen(aux_value);
  const size_t value_len = Util::CharsLen(value);

  // "<XXいる|>" can be rewrited to "<YY|いる>" and vice versa
  {
    static const char kPat[] = "いる";  // "いる"
    const absl::string_view kSuffix(kPat, arraysize(kPat) - 1);
    if (top_aux_value_len == 0 && aux_value_len == 2 &&
        Util::EndsWith(top_value, kSuffix) &&
        Util::EndsWith(aux_value, kSuffix)) {
      if (type == RIGHT) {
        // "YYいる" in addition to "YY"
        output->push_back(value);
      }
      return true;
    }
    if (aux_value_len == 0 && top_aux_value_len == 2 &&
        Util::EndsWith(value, kSuffix) &&
        Util::EndsWith(top_aux_value, kSuffix)) {
      if (type == RIGHT) {
        // "YY" in addition to "YYいる"
        PushBackStringView(Util::Utf8SubString(value, 0, value_len - 2),
                           output);
      }
      return true;
    }
  }

  // "<XXせる|>" can be rewrited to "<YY|せる>" and vice versa
  {
    const char kPat[] = "せる";
    const absl::string_view kSuffix(kPat, arraysize(kPat) - 1);
    if (top_aux_value_len == 0 && aux_value_len == 2 &&
        Util::EndsWith(top_value, kSuffix) &&
        Util::EndsWith(aux_value, kSuffix)) {
      if (type == RIGHT) {
        // "YYせる" in addition to "YY"
        output->push_back(value);
      }
      return true;
    }
    if (aux_value_len == 0 && top_aux_value_len == 2 &&
        Util::EndsWith(value, kSuffix) &&
        Util::EndsWith(top_aux_value, kSuffix)) {
      if (type == RIGHT) {
        // "YY" in addition to "YYせる"
        PushBackStringView(Util::Utf8SubString(value, 0, value_len - 2),
                           output);
      }
      return true;
    }
  }

  const Util::ScriptType content_script_type = Util::GetScriptType(content);

  // "<XX|する>" can be rewrited using "<XXす|る>" and "<XX|する>"
  // in "<XX|する>", XX must be single script type
  {
    static const char kPat[] = "する";
    const absl::string_view kSuffix(kPat, arraysize(kPat) - 1);
    if (aux_value_len == 2 && Util::EndsWith(aux_value, kSuffix)) {
      if (content_script_type != Util::KATAKANA &&
          content_script_type != Util::HIRAGANA &&
          content_script_type != Util::KANJI &&
          content_script_type != Util::ALPHABET) {
        return false;
      }
      if (type == RIGHT) {
        // "YYす" in addition to "YY"
        PushBackStringView(Util::Utf8SubString(value, 0, value_len - 1),
                           output);
      }
      return true;
    }
  }

  // "<XXる>" can be rewrited using "<XX|る>"
  // "まとめる", "衰える"
  {
    static const char kPat[] = "る";
    const absl::string_view kSuffix(kPat, arraysize(kPat) - 1);
    if (aux_value_len == 0 && Util::EndsWith(value, kSuffix)) {
      if (type == RIGHT) {
        // "YY" in addition to "YYる"
        PushBackStringView(Util::Utf8SubString(value, 0, value_len - 1),
                           output);
      }
      return true;
    }
  }

  // "<XXす>" can be rewrited using "XXする"
  {
    static const char kPat[] = "す";
    const absl::string_view kSuffix(kPat, arraysize(kPat) - 1);
    if (Util::EndsWith(value, kSuffix) &&
        Util::IsScriptType(Util::Utf8SubString(value, 0, value_len - 1),
                           Util::KANJI)) {
      if (type == RIGHT) {
        const char kRu[] = "る";
        // "YYする" in addition to "YY"
        output->push_back(
            absl::StrCat(value, absl::string_view(kRu, arraysize(kRu) - 1)));
      }
      return true;
    }
  }

  // "<XXし|た>" can be rewrited using "<XX|した>"
  {
    static const char kPat[] = "した";
    const absl::string_view kShi(kPat, 3), kTa(kPat + 3, 3);
    if (Util::EndsWith(content, kShi) && aux_value == kTa &&
        Util::EndsWith(top_content, kShi) && top_aux_value == kTa) {
      if (type == RIGHT) {
        const absl::string_view val =
            Util::Utf8SubString(content, 0, content_len - 1);
        // XX must be KANJI
        if (Util::IsScriptType(val, Util::KANJI)) {
          PushBackStringView(val, output);
        }
      }
      return true;
    }
  }

  const int aux_len = value_len - content_len;
  const int top_aux_len = Util::CharsLen(top_value) - top_content_len;
  if (aux_len != top_aux_len) {
    return false;
  }

  const Util::ScriptType top_content_script_type =
      Util::GetScriptType(top_content);

  // we don't rewrite HIRAGANA to KATAKANA
  if (top_content_script_type == Util::HIRAGANA &&
      content_script_type == Util::KATAKANA) {
    return false;
  }

  // we don't rewrite second KATAKANA
  // for example, we don't rewrite "このコーヒー" to "この珈琲"
  if (type == RIGHT && top_content_script_type == Util::KATAKANA &&
      value != top_value) {
    return false;
  }

  if (top_content_len == 1 && top_content_script_type == Util::HIRAGANA) {
    return false;
  }

  // suppress "<身|ています>" etc.
  if (top_content_len == 1 && content_len == 1 && top_aux_value_len >= 2 &&
      aux_value_len >= 2 && top_content_script_type == Util::KANJI &&
      content_script_type == Util::KANJI && top_content != content) {
    return false;
  }

  return true;
}

// Just a wrapper of IsNaturalContent for debug.
bool VerifyNaturalContent(const Segment::Candidate &cand,
                          const Segment::Candidate &top_cand,
                          SegmentLookupType type) {
  std::vector<std::string> nexts;
  return IsNaturalContent(cand, top_cand, RIGHT, &nexts);
}

inline bool IsKeyUnknown(const Segment &seg) {
  return Util::IsScriptType(seg.key(), Util::UNKNOWN_SCRIPT);
}

}  // namespace

bool CollocationRewriter::RewriteCollocation(Segments *segments) const {
  // return false if at least one segment is fixed.
  for (size_t i = segments->history_segments_size();
       i < segments->segments_size(); ++i) {
    if (segments->segment(i).segment_type() == Segment::FIXED_VALUE) {
      return false;
    }
  }

  std::vector<bool> segs_changed(segments->segments_size(), false);
  bool changed = false;

  for (size_t i = segments->history_segments_size();
       i < segments->segments_size(); ++i) {
    bool rewrited_next = false;

    if (IsKeyUnknown(segments->segment(i))) {
      continue;
    }

    if (i + 1 < segments->segments_size() &&
        RewriteUsingNextSegment(segments->mutable_segment(i + 1),
                                segments->mutable_segment(i))) {
      changed = true;
      rewrited_next = true;
      segs_changed[i] = true;
      segs_changed[i + 1] = true;
    }

    if (!segs_changed[i] && !rewrited_next && i > 0 &&
        RewriteFromPrevSegment(segments->segment(i - 1).candidate(0),
                               segments->mutable_segment(i))) {
      changed = true;
      segs_changed[i - 1] = true;
      segs_changed[i] = true;
    }

    const Segment::Candidate &cand = segments->segment(i).candidate(0);
    if (i >= 2 &&
        // Cross over only adverbs
        // Segment is adverb if;
        //  1) lid and rid is adverb.
        //  2) or rid is adverb suffix.
        ((pos_matcher_.IsAdverb(segments->segment(i - 1).candidate(0).lid) &&
          pos_matcher_.IsAdverb(segments->segment(i - 1).candidate(0).rid)) ||
         pos_matcher_.IsAdverbSegmentSuffix(
             segments->segment(i - 1).candidate(0).rid)) &&
        (cand.content_value != cand.value ||
         cand.value != "・")) {  // "・" workaround
      if (!segs_changed[i - 2] && !segs_changed[i] &&
          RewriteUsingNextSegment(segments->mutable_segment(i),
                                  segments->mutable_segment(i - 2))) {
        changed = true;
        segs_changed[i] = true;
        segs_changed[i - 2] = true;
      } else if (!segs_changed[i] &&
                 RewriteFromPrevSegment(segments->segment(i - 2).candidate(0),
                                        segments->mutable_segment(i))) {
        changed = true;
        segs_changed[i] = true;
        segs_changed[i - 2] = true;
      }
    }
  }

  return changed;
}

class CollocationRewriter::CollocationFilter {
 public:
  CollocationFilter(const char *existence_data, size_t size)
      : filter_(ExistenceFilter::Read(existence_data, size)) {}
  ~CollocationFilter() {}

  bool Exists(const std::string &left, const std::string &right) const {
    if (left.empty() || right.empty()) {
      return false;
    }
    std::string key;
    key.reserve(left.size() + right.size());
    key.assign(left).append(right);
    const uint64 id = Hash::Fingerprint(key);
    return filter_->Exists(id);
  }

 private:
  std::unique_ptr<ExistenceFilter> filter_;

  DISALLOW_COPY_AND_ASSIGN(CollocationFilter);
};

class CollocationRewriter::SuppressionFilter {
 public:
  SuppressionFilter(const char *suppression_data, size_t size)
      : filter_(ExistenceFilter::Read(suppression_data, size)) {}
  ~SuppressionFilter() {}

  bool Exists(const Segment::Candidate &cand) const {
    // TODO(noriyukit): We should share key generation rule with
    // gen_collocation_suppression_data_main.cc.
    std::string key;
    key.reserve(cand.content_value.size() + 1 + cand.content_key.size());
    key.assign(cand.content_value).append("\t").append(cand.content_key);
    const uint64 id = Hash::Fingerprint(key);
    return filter_->Exists(id);
  }

 private:
  std::unique_ptr<ExistenceFilter> filter_;

  DISALLOW_COPY_AND_ASSIGN(SuppressionFilter);
};

CollocationRewriter::CollocationRewriter(
    const DataManagerInterface *data_manager)
    : pos_matcher_(data_manager->GetPOSMatcherData()),
      first_name_id_(pos_matcher_.GetFirstNameId()),
      last_name_id_(pos_matcher_.GetLastNameId()) {
  const char *data = nullptr;
  size_t size = 0;

  data_manager->GetCollocationData(&data, &size);
  collocation_filter_.reset(new CollocationFilter(data, size));

  data_manager->GetCollocationSuppressionData(&data, &size);
  suppression_filter_.reset(new SuppressionFilter(data, size));
}

CollocationRewriter::~CollocationRewriter() {}

bool CollocationRewriter::Rewrite(const ConversionRequest &request,
                                  Segments *segments) const {
  if (!FLAGS_use_collocation) {
    return false;
  }
  return RewriteCollocation(segments);
}

bool CollocationRewriter::IsName(const Segment::Candidate &cand) const {
  const bool ret = (cand.lid == last_name_id_ || cand.lid == first_name_id_);
  if (ret) {
    VLOG(3) << cand.value << " is name sagment";
  }
  return ret;
}

bool CollocationRewriter::RewriteFromPrevSegment(
    const Segment::Candidate &prev_cand, Segment *seg) const {
  std::string prev;
  CollocationUtil::GetNormalizedScript(prev_cand.value, true, &prev);

  const size_t i_max = std::min(seg->candidates_size(), kCandidateSize);

  // Reuse |curs| and |cur| in the loop as this method is performance critical.
  std::vector<std::string> curs;
  std::string cur;
  for (size_t i = 0; i < i_max; ++i) {
    if (seg->candidate(i).cost > seg->candidate(0).cost + kMaxCostDiff) {
      continue;
    }
    if (IsName(seg->candidate(i))) {
      continue;
    }
    if (suppression_filter_->Exists(seg->candidate(i))) {
      continue;
    }
    curs.clear();
    if (!IsNaturalContent(seg->candidate(i), seg->candidate(0), RIGHT, &curs)) {
      continue;
    }

    for (int j = 0; j < curs.size(); ++j) {
      cur.clear();
      CollocationUtil::GetNormalizedScript(curs[j], false, &cur);
      if (collocation_filter_->Exists(prev, cur)) {
        if (i != 0) {
          VLOG(3) << prev << cur << " " << seg->candidate(0).value << "->"
                  << seg->candidate(i).value;
        }
        seg->move_candidate(i, 0);
        seg->mutable_candidate(0)->attributes |=
            Segment::Candidate::CONTEXT_SENSITIVE;
        return true;
      }
    }
  }
  return false;
}

bool CollocationRewriter::RewriteUsingNextSegment(Segment *next_seg,
                                                  Segment *seg) const {
  const size_t i_max = std::min(seg->candidates_size(), kCandidateSize);
  const size_t j_max = std::min(next_seg->candidates_size(), kCandidateSize);

  // Cache the results for the next segment
  std::vector<int> next_seg_ok(j_max);  // Avoiding std::vector<bool>
  std::vector<std::vector<std::string> > normalized_string(j_max);

  // Reuse |nexts| in the loop as this method is performance critical.
  std::vector<std::string> nexts;
  for (size_t j = 0; j < j_max; ++j) {
    next_seg_ok[j] = 0;

    if (IsName(next_seg->candidate(j))) {
      continue;
    }
    if (suppression_filter_->Exists(next_seg->candidate(j))) {
      continue;
    }
    nexts.clear();
    if (!IsNaturalContent(next_seg->candidate(j), next_seg->candidate(0), RIGHT,
                          &nexts)) {
      continue;
    }

    next_seg_ok[j] = 1;
    for (std::vector<std::string>::const_iterator it = nexts.begin();
         it != nexts.end(); ++it) {
      normalized_string[j].push_back(std::string());
      CollocationUtil::GetNormalizedScript(*it, false,
                                           &normalized_string[j].back());
    }
  }

  // Reuse |curs| and |cur| in the loop as this method is performance critical.
  std::vector<std::string> curs;
  std::string cur;
  for (size_t i = 0; i < i_max; ++i) {
    if (seg->candidate(i).cost > seg->candidate(0).cost + kMaxCostDiff) {
      continue;
    }
    if (IsName(seg->candidate(i))) {
      continue;
    }
    if (suppression_filter_->Exists(seg->candidate(i))) {
      continue;
    }
    curs.clear();
    if (!IsNaturalContent(seg->candidate(i), seg->candidate(0), LEFT, &curs)) {
      continue;
    }

    for (int k = 0; k < curs.size(); ++k) {
      cur.clear();
      CollocationUtil::GetNormalizedScript(curs[k], true, &cur);
      for (size_t j = 0; j < j_max; ++j) {
        if (next_seg->candidate(j).cost >
            next_seg->candidate(0).cost + kMaxCostDiff) {
          continue;
        }
        if (!next_seg_ok[j]) {
          continue;
        }

        for (int l = 0; l < normalized_string[j].size(); ++l) {
          const std::string &next = normalized_string[j][l];
          if (collocation_filter_->Exists(cur, next)) {
            DCHECK(VerifyNaturalContent(next_seg->candidate(j),
                                        next_seg->candidate(0), RIGHT))
                << "IsNaturalContent() should not fail here.";
            seg->move_candidate(i, 0);
            seg->mutable_candidate(0)->attributes |=
                Segment::Candidate::CONTEXT_SENSITIVE;
            next_seg->move_candidate(j, 0);
            next_seg->mutable_candidate(0)->attributes |=
                Segment::Candidate::CONTEXT_SENSITIVE;
            return true;
          }
        }
      }
    }
  }
  return false;
}

}  // namespace mozc