| 1 | // Copyright (C) 2016 The Qt Company Ltd. |
| 2 | // SPDX-License-Identifier: LicenseRef-Qt-Commercial OR GPL-3.0-only WITH Qt-GPL-exception-1.0 |
| 3 | |
| 4 | #include "simtexth.h" |
| 5 | #include "translator.h" |
| 6 | |
| 7 | #include <QtCore/QByteArray> |
| 8 | #include <QtCore/QString> |
| 9 | #include <QtCore/QList> |
| 10 | |
| 11 | |
| 12 | QT_BEGIN_NAMESPACE |
| 13 | |
| 14 | typedef QList<TranslatorMessage> TML; |
| 15 | |
| 16 | /* |
| 17 | How similar are two texts? The approach used here relies on co-occurrence |
| 18 | matrices and is very efficient. |
| 19 | |
| 20 | Let's see with an example: how similar are "here" and "hither"? The |
| 21 | co-occurrence matrix M for "here" is M[h,e] = 1, M[e,r] = 1, M[r,e] = 1, and 0 |
| 22 | elsewhere; the matrix N for "hither" is N[h,i] = 1, N[i,t] = 1, ..., |
| 23 | N[h,e] = 1, N[e,r] = 1, and 0 elsewhere. The union U of both matrices is the |
| 24 | matrix U[i,j] = max { M[i,j], N[i,j] }, and the intersection V is |
| 25 | V[i,j] = min { M[i,j], N[i,j] }. The score for a pair of texts is |
| 26 | |
| 27 | score = (sum of V[i,j] over all i, j) / (sum of U[i,j] over all i, j), |
| 28 | |
| 29 | a formula suggested by Arnt Gulbrandsen. Here we have |
| 30 | |
| 31 | score = 2 / 6, |
| 32 | |
| 33 | or one third. |
| 34 | |
| 35 | The implementation differs from this in a few details. Most importantly, |
| 36 | repetitions are ignored; for input "xxx", M[x,x] equals 1, not 2. |
| 37 | */ |
| 38 | |
| 39 | /* |
| 40 | Every character is assigned to one of 20 buckets so that the co-occurrence |
| 41 | matrix requires only 20 * 20 = 400 bits, not 256 * 256 = 65536 bits or even |
| 42 | more if we want the whole Unicode. Which character falls in which bucket is |
| 43 | arbitrary. |
| 44 | |
| 45 | The second half of the table is a replica of the first half, because of |
| 46 | laziness. |
| 47 | */ |
| 48 | static const int indexOf[256] = { |
| 49 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 50 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 51 | // ! " # $ % & ' ( ) * + , - . / |
| 52 | 0, 2, 6, 7, 10, 12, 15, 19, 2, 6, 7, 10, 12, 15, 19, 0, |
| 53 | // 0 1 2 3 4 5 6 7 8 9 : ; < = > ? |
| 54 | 1, 3, 4, 5, 8, 9, 11, 13, 14, 16, 2, 6, 7, 10, 12, 15, |
| 55 | // @ A B C D E F G H I J K L M N O |
| 56 | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 6, 10, 11, 12, 13, 14, |
| 57 | // P Q R S T U V W X Y Z [ \ ] ^ _ |
| 58 | 15, 12, 16, 17, 18, 19, 2, 10, 15, 7, 19, 2, 6, 7, 10, 0, |
| 59 | // ` a b c d e f g h i j k l m n o |
| 60 | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 6, 10, 11, 12, 13, 14, |
| 61 | // p q r s t u v w x y z { | } ~ |
| 62 | 15, 12, 16, 17, 18, 19, 2, 10, 15, 7, 19, 2, 6, 7, 10, 0, |
| 63 | |
| 64 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 65 | 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, |
| 66 | 0, 2, 6, 7, 10, 12, 15, 19, 2, 6, 7, 10, 12, 15, 19, 0, |
| 67 | 1, 3, 4, 5, 8, 9, 11, 13, 14, 16, 2, 6, 7, 10, 12, 15, |
| 68 | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 6, 10, 11, 12, 13, 14, |
| 69 | 15, 12, 16, 17, 18, 19, 2, 10, 15, 7, 19, 2, 6, 7, 10, 0, |
| 70 | 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, 6, 10, 11, 12, 13, 14, |
| 71 | 15, 12, 16, 17, 18, 19, 2, 10, 15, 7, 19, 2, 6, 7, 10, 0 |
| 72 | }; |
| 73 | |
| 74 | /* |
| 75 | The entry bitCount[i] (for i between 0 and 255) is the number of bits used to |
| 76 | represent i in binary. |
| 77 | */ |
| 78 | static const int bitCount[256] = { |
| 79 | 0, 1, 1, 2, 1, 2, 2, 3, 1, 2, 2, 3, 2, 3, 3, 4, |
| 80 | 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, |
| 81 | 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, |
| 82 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, |
| 83 | 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, |
| 84 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, |
| 85 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, |
| 86 | 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, |
| 87 | 1, 2, 2, 3, 2, 3, 3, 4, 2, 3, 3, 4, 3, 4, 4, 5, |
| 88 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, |
| 89 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, |
| 90 | 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, |
| 91 | 2, 3, 3, 4, 3, 4, 4, 5, 3, 4, 4, 5, 4, 5, 5, 6, |
| 92 | 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, |
| 93 | 3, 4, 4, 5, 4, 5, 5, 6, 4, 5, 5, 6, 5, 6, 6, 7, |
| 94 | 4, 5, 5, 6, 5, 6, 6, 7, 5, 6, 6, 7, 6, 7, 7, 8 |
| 95 | }; |
| 96 | |
| 97 | static inline void setCoOccurence(CoMatrix &m, char c, char d) |
| 98 | { |
| 99 | int k = indexOf[(uchar) c] + 20 * indexOf[(uchar) d]; |
| 100 | m.b[k >> 3] |= (1 << (k & 0x7)); |
| 101 | } |
| 102 | |
| 103 | CoMatrix::CoMatrix(const QString &str) |
| 104 | { |
| 105 | QByteArray ba = str.toUtf8(); |
| 106 | const char *text = ba.constData(); |
| 107 | char c = '\0', d; |
| 108 | memset( s: b, c: 0, n: 52 ); |
| 109 | /* |
| 110 | The Knuth books are not in the office only for show; they help make |
| 111 | loops 30% faster and 20% as readable. |
| 112 | */ |
| 113 | while ( (d = *text) != '\0' ) { |
| 114 | setCoOccurence(m&: *this, c, d); |
| 115 | if ( (c = *++text) != '\0' ) { |
| 116 | setCoOccurence(m&: *this, c: d, d: c); |
| 117 | text++; |
| 118 | } |
| 119 | } |
| 120 | } |
| 121 | |
| 122 | static inline int worth(const CoMatrix &m) |
| 123 | { |
| 124 | int w = 0; |
| 125 | for (int i = 0; i < 50; i++) |
| 126 | w += bitCount[m.b[i]]; |
| 127 | return w; |
| 128 | } |
| 129 | |
| 130 | static inline CoMatrix reunion(const CoMatrix &m, const CoMatrix &n) |
| 131 | { |
| 132 | CoMatrix p; |
| 133 | for (int i = 0; i < 13; ++i) |
| 134 | p.w[i] = m.w[i] | n.w[i]; |
| 135 | return p; |
| 136 | } |
| 137 | |
| 138 | static inline CoMatrix intersection(const CoMatrix &m, const CoMatrix &n) |
| 139 | { |
| 140 | CoMatrix p; |
| 141 | for (int i = 0; i < 13; ++i) |
| 142 | p.w[i] = m.w[i] & n.w[i]; |
| 143 | return p; |
| 144 | } |
| 145 | |
| 146 | StringSimilarityMatcher::StringSimilarityMatcher(const QString &stringToMatch) |
| 147 | : m_cm(stringToMatch) |
| 148 | { |
| 149 | m_length = stringToMatch.size(); |
| 150 | } |
| 151 | |
| 152 | int StringSimilarityMatcher::getSimilarityScore(const QString &strCandidate) |
| 153 | { |
| 154 | CoMatrix cmTarget(strCandidate); |
| 155 | int delta = qAbs(t: m_length - strCandidate.size()); |
| 156 | int score = ( (worth(m: intersection(m: m_cm, n: cmTarget)) + 1) << 10 ) / |
| 157 | ( worth(m: reunion(m: m_cm, n: cmTarget)) + (delta << 1) + 1 ); |
| 158 | return score; |
| 159 | } |
| 160 | |
| 161 | CandidateList similarTextHeuristicCandidates(const Translator *tor, |
| 162 | const QString &text, int maxCandidates) |
| 163 | { |
| 164 | QList<int> scores; |
| 165 | CandidateList candidates; |
| 166 | StringSimilarityMatcher matcher(text); |
| 167 | |
| 168 | for (const TranslatorMessage &mtm : tor->messages()) { |
| 169 | if (mtm.type() == TranslatorMessage::Unfinished |
| 170 | || mtm.translation().isEmpty()) |
| 171 | continue; |
| 172 | |
| 173 | QString s = mtm.sourceText(); |
| 174 | int score = matcher.getSimilarityScore(strCandidate: s); |
| 175 | |
| 176 | if (candidates.size() == maxCandidates && score > scores[maxCandidates - 1] ) |
| 177 | candidates.removeLast(); |
| 178 | |
| 179 | if (candidates.size() < maxCandidates && score >= textSimilarityThreshold) { |
| 180 | Candidate cand(mtm.context(), s, mtm.comment(), mtm.translation()); |
| 181 | |
| 182 | int i; |
| 183 | for (i = 0; i < candidates.size(); i++) { |
| 184 | if (score >= scores.at(i)) { |
| 185 | if (score == scores.at(i)) { |
| 186 | if (candidates.at(i) == cand) |
| 187 | goto continue_outer_loop; |
| 188 | } else { |
| 189 | break; |
| 190 | } |
| 191 | } |
| 192 | } |
| 193 | scores.insert(i, t: score); |
| 194 | candidates.insert(i, t: cand); |
| 195 | } |
| 196 | continue_outer_loop: |
| 197 | ; |
| 198 | } |
| 199 | return candidates; |
| 200 | } |
| 201 | |
| 202 | QT_END_NAMESPACE |
| 203 | |