• Home
  • Line#
  • Scopes#
  • Navigate#
  • Raw
  • Download
1 /*
2  * Copyright (C) 2018 The Android Open Source Project
3  *
4  * Licensed under the Apache License, Version 2.0 (the "License");
5  * you may not use this file except in compliance with the License.
6  * You may obtain a copy of the License at
7  *
8  *      http://www.apache.org/licenses/LICENSE-2.0
9  *
10  * Unless required by applicable law or agreed to in writing, software
11  * distributed under the License is distributed on an "AS IS" BASIS,
12  * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
13  * See the License for the specific language governing permissions and
14  * limitations under the License.
15  */
16 
17 #include "utils/tokenizer.h"
18 
19 #include <algorithm>
20 
21 #include "utils/base/logging.h"
22 #include "utils/base/macros.h"
23 #include "utils/strings/utf8.h"
24 
25 namespace libtextclassifier3 {
26 
Tokenizer(const TokenizationType type,const UniLib * unilib,const std::vector<const TokenizationCodepointRange * > & codepoint_ranges,const std::vector<const CodepointRange * > & internal_tokenizer_codepoint_ranges,const bool split_on_script_change,const bool icu_preserve_whitespace_tokens)27 Tokenizer::Tokenizer(
28     const TokenizationType type, const UniLib* unilib,
29     const std::vector<const TokenizationCodepointRange*>& codepoint_ranges,
30     const std::vector<const CodepointRange*>&
31         internal_tokenizer_codepoint_ranges,
32     const bool split_on_script_change,
33     const bool icu_preserve_whitespace_tokens)
34     : type_(type),
35       unilib_(unilib),
36       split_on_script_change_(split_on_script_change),
37       icu_preserve_whitespace_tokens_(icu_preserve_whitespace_tokens) {
38   for (const TokenizationCodepointRange* range : codepoint_ranges) {
39     codepoint_ranges_.emplace_back(range->UnPack());
40   }
41 
42   std::sort(codepoint_ranges_.begin(), codepoint_ranges_.end(),
43             [](const std::unique_ptr<const TokenizationCodepointRangeT>& a,
44                const std::unique_ptr<const TokenizationCodepointRangeT>& b) {
45               return a->start < b->start;
46             });
47 
48   SortCodepointRanges(internal_tokenizer_codepoint_ranges,
49                       &internal_tokenizer_codepoint_ranges_);
50 }
51 
FindTokenizationRange(int codepoint) const52 const TokenizationCodepointRangeT* Tokenizer::FindTokenizationRange(
53     int codepoint) const {
54   auto it = std::lower_bound(
55       codepoint_ranges_.begin(), codepoint_ranges_.end(), codepoint,
56       [](const std::unique_ptr<const TokenizationCodepointRangeT>& range,
57          int codepoint) {
58         // This function compares range with the codepoint for the purpose of
59         // finding the first greater or equal range. Because of the use of
60         // std::lower_bound it needs to return true when range < codepoint;
61         // the first time it will return false the lower bound is found and
62         // returned.
63         //
64         // It might seem weird that the condition is range.end <= codepoint
65         // here but when codepoint == range.end it means it's actually just
66         // outside of the range, thus the range is less than the codepoint.
67         return range->end <= codepoint;
68       });
69   if (it != codepoint_ranges_.end() && (*it)->start <= codepoint &&
70       (*it)->end > codepoint) {
71     return it->get();
72   } else {
73     return nullptr;
74   }
75 }
76 
GetScriptAndRole(char32 codepoint,TokenizationCodepointRange_::Role * role,int * script) const77 void Tokenizer::GetScriptAndRole(char32 codepoint,
78                                  TokenizationCodepointRange_::Role* role,
79                                  int* script) const {
80   const TokenizationCodepointRangeT* range = FindTokenizationRange(codepoint);
81   if (range) {
82     *role = range->role;
83     *script = range->script_id;
84   } else {
85     *role = TokenizationCodepointRange_::Role_DEFAULT_ROLE;
86     *script = kUnknownScript;
87   }
88 }
89 
Tokenize(const std::string & text) const90 std::vector<Token> Tokenizer::Tokenize(const std::string& text) const {
91   UnicodeText text_unicode = UTF8ToUnicodeText(text, /*do_copy=*/false);
92   return Tokenize(text_unicode);
93 }
94 
Tokenize(const UnicodeText & text_unicode) const95 std::vector<Token> Tokenizer::Tokenize(const UnicodeText& text_unicode) const {
96   switch (type_) {
97     case TokenizationType_INTERNAL_TOKENIZER:
98       return InternalTokenize(text_unicode);
99     case TokenizationType_ICU:
100       TC3_FALLTHROUGH_INTENDED;
101     case TokenizationType_MIXED: {
102       std::vector<Token> result;
103       if (!ICUTokenize(text_unicode, &result)) {
104         return {};
105       }
106       if (type_ == TokenizationType_MIXED) {
107         InternalRetokenize(text_unicode, &result);
108       }
109       return result;
110     }
111     default:
112       TC3_LOG(ERROR) << "Unknown tokenization type specified. Using internal.";
113       return InternalTokenize(text_unicode);
114   }
115 }
116 
InternalTokenize(const UnicodeText & text_unicode) const117 std::vector<Token> Tokenizer::InternalTokenize(
118     const UnicodeText& text_unicode) const {
119   std::vector<Token> result;
120   Token new_token("", 0, 0);
121   int codepoint_index = 0;
122 
123   int last_script = kInvalidScript;
124   for (auto it = text_unicode.begin(); it != text_unicode.end();
125        ++it, ++codepoint_index) {
126     TokenizationCodepointRange_::Role role;
127     int script;
128     GetScriptAndRole(*it, &role, &script);
129 
130     if (role & TokenizationCodepointRange_::Role_SPLIT_BEFORE ||
131         (split_on_script_change_ && last_script != kInvalidScript &&
132          last_script != script)) {
133       if (!new_token.value.empty()) {
134         result.push_back(new_token);
135       }
136       new_token = Token("", codepoint_index, codepoint_index);
137     }
138     if (!(role & TokenizationCodepointRange_::Role_DISCARD_CODEPOINT)) {
139       new_token.value += std::string(
140           it.utf8_data(),
141           it.utf8_data() + GetNumBytesForNonZeroUTF8Char(it.utf8_data()));
142       ++new_token.end;
143     }
144     if (role & TokenizationCodepointRange_::Role_SPLIT_AFTER) {
145       if (!new_token.value.empty()) {
146         result.push_back(new_token);
147       }
148       new_token = Token("", codepoint_index + 1, codepoint_index + 1);
149     }
150 
151     last_script = script;
152   }
153   if (!new_token.value.empty()) {
154     result.push_back(new_token);
155   }
156 
157   return result;
158 }
159 
TokenizeSubstring(const UnicodeText & unicode_text,CodepointSpan span,std::vector<Token> * result) const160 void Tokenizer::TokenizeSubstring(const UnicodeText& unicode_text,
161                                   CodepointSpan span,
162                                   std::vector<Token>* result) const {
163   if (span.first < 0) {
164     // There is no span to tokenize.
165     return;
166   }
167 
168   // Extract the substring.
169   UnicodeText text = UnicodeText::Substring(unicode_text, span.first,
170                                             span.second, /*do_copy=*/false);
171 
172   // Run the tokenizer and update the token bounds to reflect the offset of the
173   // substring.
174   std::vector<Token> tokens = InternalTokenize(text);
175 
176   // Avoids progressive capacity increases in the for loop.
177   result->reserve(result->size() + tokens.size());
178   for (Token& token : tokens) {
179     token.start += span.first;
180     token.end += span.first;
181     result->emplace_back(std::move(token));
182   }
183 }
184 
InternalRetokenize(const UnicodeText & unicode_text,std::vector<Token> * tokens) const185 void Tokenizer::InternalRetokenize(const UnicodeText& unicode_text,
186                                    std::vector<Token>* tokens) const {
187   std::vector<Token> result;
188   CodepointSpan span(-1, -1);
189   for (Token& token : *tokens) {
190     const UnicodeText unicode_token_value =
191         UTF8ToUnicodeText(token.value, /*do_copy=*/false);
192     bool should_retokenize = true;
193     for (const int codepoint : unicode_token_value) {
194       if (!IsCodepointInRanges(codepoint,
195                                internal_tokenizer_codepoint_ranges_)) {
196         should_retokenize = false;
197         break;
198       }
199     }
200 
201     if (should_retokenize) {
202       if (span.first < 0) {
203         span.first = token.start;
204       }
205       span.second = token.end;
206     } else {
207       TokenizeSubstring(unicode_text, span, &result);
208       span.first = -1;
209       result.emplace_back(std::move(token));
210     }
211   }
212   TokenizeSubstring(unicode_text, span, &result);
213 
214   *tokens = std::move(result);
215 }
216 
ICUTokenize(const UnicodeText & context_unicode,std::vector<Token> * result) const217 bool Tokenizer::ICUTokenize(const UnicodeText& context_unicode,
218                             std::vector<Token>* result) const {
219   std::unique_ptr<UniLib::BreakIterator> break_iterator =
220       unilib_->CreateBreakIterator(context_unicode);
221   if (!break_iterator) {
222     return false;
223   }
224   int last_break_index = 0;
225   int break_index = 0;
226   int last_unicode_index = 0;
227   int unicode_index = 0;
228   auto token_begin_it = context_unicode.begin();
229   while ((break_index = break_iterator->Next()) !=
230          UniLib::BreakIterator::kDone) {
231     const int token_length = break_index - last_break_index;
232     unicode_index = last_unicode_index + token_length;
233 
234     auto token_end_it = token_begin_it;
235     std::advance(token_end_it, token_length);
236 
237     // Determine if the whole token is whitespace.
238     bool is_whitespace = true;
239     for (auto char_it = token_begin_it; char_it < token_end_it; ++char_it) {
240       if (!unilib_->IsWhitespace(*char_it)) {
241         is_whitespace = false;
242         break;
243       }
244     }
245 
246     const std::string token =
247         context_unicode.UTF8Substring(token_begin_it, token_end_it);
248 
249     if (!is_whitespace || icu_preserve_whitespace_tokens_) {
250       result->push_back(Token(token, last_unicode_index, unicode_index));
251     }
252 
253     last_break_index = break_index;
254     last_unicode_index = unicode_index;
255     token_begin_it = token_end_it;
256   }
257 
258   return true;
259 }
260 
261 }  // namespace libtextclassifier3
262