aboutsummaryrefslogtreecommitdiffstats
path: root/native/src
diff options
context:
space:
mode:
Diffstat (limited to 'native/src')
-rw-r--r--native/src/correction.cpp60
-rw-r--r--native/src/correction.h3
-rw-r--r--native/src/correction_state.h4
-rw-r--r--native/src/defines.h11
-rw-r--r--native/src/words_priority_queue.h4
5 files changed, 62 insertions, 20 deletions
diff --git a/native/src/correction.cpp b/native/src/correction.cpp
index 8275c5d7e..5f11452ae 100644
--- a/native/src/correction.cpp
+++ b/native/src/correction.cpp
@@ -24,6 +24,7 @@
#include "char_utils.h"
#include "correction.h"
+#include "defines.h"
#include "dictionary.h"
#include "proximity_info.h"
@@ -210,6 +211,7 @@ bool Correction::initProcessState(const int outputIndex) {
mMatching = false;
mProximityMatching = false;
+ mAdditionalProximityMatching = false;
mTransposing = false;
mExceeding = false;
mSkipping = false;
@@ -256,6 +258,7 @@ void Correction::incrementOutputIndex() {
mCorrectionStates[mOutputIndex].mMatching = mMatching;
mCorrectionStates[mOutputIndex].mProximityMatching = mProximityMatching;
+ mCorrectionStates[mOutputIndex].mAdditionalProximityMatching = mAdditionalProximityMatching;
mCorrectionStates[mOutputIndex].mTransposing = mTransposing;
mCorrectionStates[mOutputIndex].mExceeding = mExceeding;
mCorrectionStates[mOutputIndex].mSkipping = mSkipping;
@@ -304,6 +307,11 @@ inline bool isEquivalentChar(ProximityInfo::ProximityType type) {
return type == ProximityInfo::EQUIVALENT_CHAR;
}
+inline bool isProximityCharOrEquivalentChar(ProximityInfo::ProximityType type) {
+ return type == ProximityInfo::EQUIVALENT_CHAR
+ || type == ProximityInfo::NEAR_PROXIMITY_CHAR;
+}
+
Correction::CorrectionType Correction::processCharAndCalcState(
const int32_t c, const bool isTerminal) {
const int correctionCount = (mSkippedCount + mExcessiveCount + mTransposedCount);
@@ -438,6 +446,9 @@ Correction::CorrectionType Correction::processCharAndCalcState(
if (ProximityInfo::UNRELATED_CHAR == matchedProximityCharId
|| ProximityInfo::ADDITIONAL_PROXIMITY_CHAR == matchedProximityCharId) {
+ if (ProximityInfo::ADDITIONAL_PROXIMITY_CHAR == matchedProximityCharId) {
+ mAdditionalProximityMatching = true;
+ }
// TODO: Optimize
// As the current char turned out to be an unrelated char,
// we will try other correction-types. Please note that mCorrectionStates[mOutputIndex]
@@ -479,6 +490,18 @@ Correction::CorrectionType Correction::processCharAndCalcState(
++mSkippedCount;
--mProximityCount;
return processSkipChar(c, isTerminal, false);
+ } else if (mInputIndex - 1 < mInputLength
+ && mSkippedCount > 0
+ && mCorrectionStates[mOutputIndex].mSkipping
+ && mCorrectionStates[mOutputIndex].mAdditionalProximityMatching
+ && isProximityCharOrEquivalentChar(
+ mProximityInfo->getMatchedProximityId(mInputIndex + 1, c, false))) {
+ // Conversion s->a
+ incrementInputIndex();
+ --mSkippedCount;
+ mProximityMatching = true;
+ ++mProximityCount;
+ mDistances[mOutputIndex] = ADDITIONAL_PROXIMITY_CHAR_DISTANCE_INFO;
} else if ((mExceeding || mTransposing) && mInputIndex - 1 < mInputLength
&& isEquivalentChar(
mProximityInfo->getMatchedProximityId(mInputIndex + 1, c, false))) {
@@ -618,7 +641,7 @@ inline static int getQuoteCount(const unsigned short* word, const int length) {
}
inline static bool isUpperCase(unsigned short c) {
- return isAsciiUpper(toBaseChar(c));
+ return isAsciiUpper(toBaseChar(c));
}
//////////////////////
@@ -666,6 +689,10 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
int finalFreq = freq;
+ if (DEBUG_CORRECTION_FREQ
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == inputLength)) {
+ AKLOGI("FinalFreq0: %d", finalFreq);
+ }
// TODO: Optimize this.
if (transposedCount > 0 || proximityMatchedCount > 0 || skipped || excessiveCount > 0) {
ed = getCurrentEditDistance(editDistanceTable, correction->mInputLength, outputLength,
@@ -681,12 +708,15 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
}
ed = max(0, ed - quoteDiffCount);
-
+ adjustedProximityMatchedCount = min(max(0, ed - (outputLength - inputLength)),
+ proximityMatchedCount);
if (transposedCount < 1) {
if (ed == 1 && (inputLength == outputLength - 1 || inputLength == outputLength + 1)) {
// Promote a word with just one skipped or excessive char
if (sameLength) {
- multiplyRate(WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_RATE, &finalFreq);
+ multiplyRate(WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_RATE
+ + WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_MULTIPLIER * outputLength,
+ &finalFreq);
} else {
multiplyIntCapped(typedLetterMultiplier, &finalFreq);
}
@@ -695,8 +725,6 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
sameLength = true;
}
}
- adjustedProximityMatchedCount = min(max(0, ed - (outputLength - inputLength)),
- proximityMatchedCount);
} else {
const int matchWeight = powerIntCapped(typedLetterMultiplier, matchCount);
multiplyIntCapped(matchWeight, &finalFreq);
@@ -744,6 +772,7 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
&& skippedCount == 0 && excessiveCount == 0 && transposedCount == 0;
// Score calibration by touch coordinates is being done only for pure-fat finger typing error
// cases.
+ int additionalProximityCount = 0;
// TODO: Remove this constraint.
if (performTouchPositionCorrection) {
for (int i = 0; i < outputLength; ++i) {
@@ -776,12 +805,12 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
} else if (squaredDistance == PROXIMITY_CHAR_WITHOUT_DISTANCE_INFO) {
multiplyRate(WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE, &finalFreq);
} else if (squaredDistance == ADDITIONAL_PROXIMITY_CHAR_DISTANCE_INFO) {
+ ++additionalProximityCount;
multiplyRate(WORDS_WITH_ADDITIONAL_PROXIMITY_CHARACTER_DEMOTION_RATE, &finalFreq);
}
}
} else {
// Demote additional proximity characters
- int additionalProximityCount = 0;
for (int i = 0; i < outputLength; ++i) {
const int squaredDistance = correction->mDistances[i];
if (squaredDistance == ADDITIONAL_PROXIMITY_CHAR_DISTANCE_INFO) {
@@ -803,6 +832,13 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
}
}
+ // If the user types too many(three or more) proximity characters with additional proximity
+ // character,do not treat as the same length word.
+ if (sameLength && additionalProximityCount > 0 && (adjustedProximityMatchedCount >= 3
+ || transposedCount > 0 || skipped || excessiveCount > 0)) {
+ sameLength = false;
+ }
+
const int errorCount = adjustedProximityMatchedCount > 0
? adjustedProximityMatchedCount
: (proximityMatchedCount + transposedCount);
@@ -813,13 +849,14 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
if (ed == 0) {
// Full exact match
if (sameLength && transposedCount == 0 && !skipped && excessiveCount == 0
- && quoteDiffCount == 0) {
+ && quoteDiffCount == 0 && additionalProximityCount == 0) {
finalFreq = capped255MultForFullMatchAccentsOrCapitalizationDifference(finalFreq);
}
}
// Promote a word with no correction
- if (proximityMatchedCount == 0 && transposedCount == 0 && !skipped && excessiveCount == 0) {
+ if (proximityMatchedCount == 0 && transposedCount == 0 && !skipped && excessiveCount == 0
+ && additionalProximityCount == 0) {
multiplyRate(FULL_MATCHED_WORDS_PROMOTION_RATE, &finalFreq);
}
@@ -863,10 +900,11 @@ int Correction::RankingAlgorithm::calculateFinalFreq(const int inputIndex, const
if (DEBUG_CORRECTION_FREQ
&& (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == inputLength)) {
+ DUMP_WORD(proximityInfo->getPrimaryInputWord(), inputLength);
DUMP_WORD(correction->mWord, outputLength);
- AKLOGI("FinalFreq: [P%d, S%d, T%d, E%d] %d, %d, %d, %d, %d, %d", proximityMatchedCount,
- skippedCount, transposedCount, excessiveCount, outputLength, lastCharExceeded,
- sameLength, quoteDiffCount, ed, finalFreq);
+ AKLOGI("FinalFreq: [P%d, S%d, T%d, E%d, A%d] %d, %d, %d, %d, %d, %d", proximityMatchedCount,
+ skippedCount, transposedCount, excessiveCount, additionalProximityCount,
+ outputLength, lastCharExceeded, sameLength, quoteDiffCount, ed, finalFreq);
}
return finalFreq;
diff --git a/native/src/correction.h b/native/src/correction.h
index a711c994d..9559bbf12 100644
--- a/native/src/correction.h
+++ b/native/src/correction.h
@@ -85,7 +85,7 @@ class Correction {
}
}
- Correction(const int typedLetterMultiplier, const int fullWordMultiplier);
+ Correction(const int typedLetterMultiplier, const int fullWordMultiplier);
void initCorrection(
const ProximityInfo *pi, const int inputLength, const int maxWordLength);
void initCorrectionState(const int rootPos, const int childCount, const bool traverseAll);
@@ -221,6 +221,7 @@ class Correction {
bool mMatching;
bool mProximityMatching;
+ bool mAdditionalProximityMatching;
bool mExceeding;
bool mTransposing;
bool mSkipping;
diff --git a/native/src/correction_state.h b/native/src/correction_state.h
index c04146e54..5b2cbd3a2 100644
--- a/native/src/correction_state.h
+++ b/native/src/correction_state.h
@@ -47,9 +47,9 @@ struct CorrectionState {
bool mExceeding;
bool mSkipping;
bool mProximityMatching;
+ bool mAdditionalProximityMatching;
bool mNeedsToTraverseAllNodes;
-
};
inline static void initCorrectionState(CorrectionState *state, const int rootPos,
@@ -77,7 +77,7 @@ inline static void initCorrectionState(CorrectionState *state, const int rootPos
state->mTransposing = false;
state->mExceeding = false;
state->mSkipping = false;
-
+ state->mAdditionalProximityMatching = false;
}
} // namespace latinime
diff --git a/native/src/defines.h b/native/src/defines.h
index 02c1fe0a2..5b5c54850 100644
--- a/native/src/defines.h
+++ b/native/src/defines.h
@@ -195,9 +195,10 @@ static void prof_out(void) {
#define WORDS_WITH_TRANSPOSED_CHARACTERS_DEMOTION_RATE 70
#define FULL_MATCHED_WORDS_PROMOTION_RATE 120
#define WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE 90
-#define WORDS_WITH_ADDITIONAL_PROXIMITY_CHARACTER_DEMOTION_RATE 30
+#define WORDS_WITH_ADDITIONAL_PROXIMITY_CHARACTER_DEMOTION_RATE 70
#define WORDS_WITH_MATCH_SKIP_PROMOTION_RATE 105
-#define WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_RATE 160
+#define WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_RATE 148
+#define WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_MULTIPLIER 3
#define CORRECTION_COUNT_RATE_DEMOTION_RATE_BASE 45
#define INPUT_EXCEEDS_OUTPUT_DEMOTION_RATE 70
#define FIRST_CHAR_DIFFERENT_DEMOTION_RATE 96
@@ -240,14 +241,14 @@ static void prof_out(void) {
#define MIN_USER_TYPED_LENGTH_FOR_MULTIPLE_WORD_SUGGESTION 3
#define MIN_USER_TYPED_LENGTH_FOR_EXCESSIVE_CHARACTER_SUGGESTION 3
-#define min(a,b) ((a)<(b)?(a):(b))
-#define max(a,b) ((a)>(b)?(a):(b))
+template<typename T> inline T min(T a, T b) { return a < b ? a : b; }
+template<typename T> inline T max(T a, T b) { return a > b ? a : b; }
// The ratio of neutral area radius to sweet spot radius.
#define NEUTRAL_AREA_RADIUS_RATIO 1.3f
// DEBUG
-#define INPUTLENGTH_FOR_DEBUG -1
+#define INPUTLENGTH_FOR_DEBUG 10
#define MIN_OUTPUT_INDEX_FOR_DEBUG -1
#endif // LATINIME_DEFINES_H
diff --git a/native/src/words_priority_queue.h b/native/src/words_priority_queue.h
index e8cd983b1..249962eec 100644
--- a/native/src/words_priority_queue.h
+++ b/native/src/words_priority_queue.h
@@ -17,6 +17,7 @@
#ifndef LATINIME_WORDS_PRIORITY_QUEUE_H
#define LATINIME_WORDS_PRIORITY_QUEUE_H
+#include <cstring> // for memcpy()
#include <iostream>
#include <queue>
#include "defines.h"
@@ -93,7 +94,8 @@ class WordsPriorityQueue {
int outputSuggestions(int *frequencies, unsigned short *outputChars) {
mHighestSuggestedWord = 0;
- const unsigned int size = min(MAX_WORDS, mSuggestions.size());
+ const unsigned int size = min(
+ MAX_WORDS, static_cast<unsigned int>(mSuggestions.size()));
int index = size - 1;
while (!mSuggestions.empty() && index >= 0) {
SuggestedWord* sw = mSuggestions.top();
generated by cgit v1.2.3-83-g751a (git 2.47.1) at 2025-05-28 14:40:03 +0000