Support additional proximity characters
Change-Id: Ifbe0d7e4eafea1926bbce968eae4724dd5769689
diff --git a/native/src/correction.cpp b/native/src/correction.cpp
index 7323747..dafc0fd 100644
--- a/native/src/correction.cpp
+++ b/native/src/correction.cpp
@@ -383,7 +383,10 @@
incrementInputIndex();
} else {
--mTransposedCount;
- if (DEBUG_CORRECTION) {
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0
+ || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
DUMP_WORD(mWord, mOutputIndex);
AKLOGI("UNRELATED(0): %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
mTransposedCount, mExcessiveCount, c);
@@ -404,13 +407,17 @@
: mProximityInfo->getMatchedProximityId(
mInputIndex, c, checkProximityChars, &proximityIndex);
- if (ProximityInfo::UNRELATED_CHAR == matchedProximityCharId) {
+ if (ProximityInfo::UNRELATED_CHAR == matchedProximityCharId
+ || ProximityInfo::ADDITIONAL_PROXIMITY_CHAR == matchedProximityCharId) {
if (canTryCorrection && mOutputIndex > 0
&& mCorrectionStates[mOutputIndex].mProximityMatching
&& mCorrectionStates[mOutputIndex].mExceeding
&& isEquivalentChar(mProximityInfo->getMatchedProximityId(
mInputIndex, mWord[mOutputIndex - 1], false))) {
- if (DEBUG_CORRECTION) {
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0
+ || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
AKLOGI("CONVERSION p->e %c", mWord[mOutputIndex - 1]);
}
// Conversion p->e
@@ -429,7 +436,8 @@
}
}
- if (ProximityInfo::UNRELATED_CHAR == matchedProximityCharId) {
+ if (ProximityInfo::UNRELATED_CHAR == matchedProximityCharId
+ || ProximityInfo::ADDITIONAL_PROXIMITY_CHAR == matchedProximityCharId) {
// TODO: Optimize
// As the current char turned out to be an unrelated char,
// we will try other correction-types. Please note that mCorrectionStates[mOutputIndex]
@@ -481,12 +489,47 @@
++mExcessiveCount;
incrementInputIndex();
}
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0
+ || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
+ DUMP_WORD(mWord, mOutputIndex);
+ if (mTransposing) {
+ AKLOGI("TRANSPOSE: %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
+ mTransposedCount, mExcessiveCount, c);
+ } else {
+ AKLOGI("EXCEED: %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
+ mTransposedCount, mExcessiveCount, c);
+ }
+ }
} else if (mSkipping) {
// 3. Skip correction
++mSkippedCount;
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0
+ || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
+ AKLOGI("SKIP: %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
+ mTransposedCount, mExcessiveCount, c);
+ }
return processSkipChar(c, isTerminal, false);
+ } else if (ProximityInfo::ADDITIONAL_PROXIMITY_CHAR == matchedProximityCharId) {
+ // As a last resort, use additional proximity characters
+ mProximityMatching = true;
+ ++mProximityCount;
+ mDistances[mOutputIndex] = ADDITIONAL_PROXIMITY_CHAR_DISTANCE_INFO;
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0
+ || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
+ AKLOGI("ADDITIONALPROX: %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
+ mTransposedCount, mExcessiveCount, c);
+ }
} else {
- if (DEBUG_CORRECTION) {
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0
+ || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
DUMP_WORD(mWord, mOutputIndex);
AKLOGI("UNRELATED(1): %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
mTransposedCount, mExcessiveCount, c);
@@ -506,6 +549,13 @@
++mProximityCount;
mDistances[mOutputIndex] =
mProximityInfo->getNormalizedSquaredDistance(mInputIndex, proximityIndex);
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0
+ || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
+ AKLOGI("PROX: %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
+ mTransposedCount, mExcessiveCount, c);
+ }
}
addCharToCurrentWord(c);
@@ -539,7 +589,9 @@
|| isSameAsUserTypedLength) && isTerminal) {
mTerminalInputIndex = mInputIndex - 1;
mTerminalOutputIndex = mOutputIndex - 1;
- if (DEBUG_CORRECTION) {
+ if (DEBUG_CORRECTION
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == mInputLength)
+ && (MIN_OUTPUT_INDEX_FOR_DEBUG <= 0 || MIN_OUTPUT_INDEX_FOR_DEBUG < mOutputIndex)) {
DUMP_WORD(mWord, mOutputIndex);
AKLOGI("ONTERMINAL(1): %d, %d, %d, %d, %c", mProximityCount, mSkippedCount,
mTransposedCount, mExcessiveCount, c);
@@ -678,7 +730,7 @@
if (excessiveCount > 0) {
multiplyRate(WORDS_WITH_EXCESSIVE_CHARACTER_DEMOTION_RATE, &finalFreq);
if (!lastCharExceeded && !proximityInfo->existsAdjacentProximityChars(excessivePos)) {
- if (DEBUG_CORRECTION_FREQ) {
+ if (DEBUG_DICT_FULL) {
AKLOGI("Double excessive demotion");
}
// If an excessive character is not adjacent to the left char or the right char,
@@ -687,51 +739,46 @@
}
}
+ const bool performTouchPositionCorrection =
+ CALIBRATE_SCORE_BY_TOUCH_COORDINATES && proximityInfo->touchPositionCorrectionEnabled()
+ && skippedCount == 0 && excessiveCount == 0 && transposedCount == 0;
// Score calibration by touch coordinates is being done only for pure-fat finger typing error
// cases.
// TODO: Remove this constraint.
- if (CALIBRATE_SCORE_BY_TOUCH_COORDINATES && proximityInfo->touchPositionCorrectionEnabled()
- && skippedCount == 0 && excessiveCount == 0 && transposedCount == 0) {
- for (int i = 0; i < outputLength; ++i) {
- const int squaredDistance = correction->mDistances[i];
- if (i < adjustedProximityMatchedCount) {
- multiplyIntCapped(typedLetterMultiplier, &finalFreq);
- }
- if (squaredDistance >= 0) {
- // Promote or demote the score according to the distance from the sweet spot
- static const float A = ZERO_DISTANCE_PROMOTION_RATE / 100.0f;
- static const float B = 1.0f;
- static const float C = 0.5f;
- static const float R1 = NEUTRAL_SCORE_SQUARED_RADIUS;
- static const float R2 = HALF_SCORE_SQUARED_RADIUS;
- const float x = (float)squaredDistance
- / ProximityInfo::NORMALIZED_SQUARED_DISTANCE_SCALING_FACTOR;
- const float factor = (x < R1)
- ? (A * (R1 - x) + B * x) / R1
- : (B * (R2 - x) + C * (x - R1)) / (R2 - R1);
- // factor is piecewise linear function like:
- // A -_ .
- // ^-_ .
- // B \ .
- // \ .
- // C \ .
- // 0 R1 R2
- if (factor <= 0) {
- return -1;
- }
- multiplyRate((int)(factor * 100), &finalFreq);
- } else if (squaredDistance == PROXIMITY_CHAR_WITHOUT_DISTANCE_INFO) {
- multiplyRate(WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE, &finalFreq);
- }
- }
- } else {
- // Promotion for a word with proximity characters
- for (int i = 0; i < adjustedProximityMatchedCount; ++i) {
- // A word with proximity corrections
- if (DEBUG_DICT_FULL) {
- AKLOGI("Found a proximity correction.");
- }
+ for (int i = 0; i < outputLength; ++i) {
+ const int squaredDistance = correction->mDistances[i];
+ if (i < adjustedProximityMatchedCount) {
multiplyIntCapped(typedLetterMultiplier, &finalFreq);
+ }
+
+ if (performTouchPositionCorrection && squaredDistance >= 0) {
+ // Promote or demote the score according to the distance from the sweet spot
+ static const float A = ZERO_DISTANCE_PROMOTION_RATE / 100.0f;
+ static const float B = 1.0f;
+ static const float C = 0.5f;
+ static const float MIN = 0.3f;
+ static const float R1 = NEUTRAL_SCORE_SQUARED_RADIUS;
+ static const float R2 = HALF_SCORE_SQUARED_RADIUS;
+ const float x = (float)squaredDistance
+ / ProximityInfo::NORMALIZED_SQUARED_DISTANCE_SCALING_FACTOR;
+ const float factor = max((x < R1)
+ ? (A * (R1 - x) + B * x) / R1
+ : (B * (R2 - x) + C * (x - R1)) / (R2 - R1), MIN);
+ // factor is piecewise linear function like:
+ // A -_ .
+ // ^-_ .
+ // B \ .
+ // \_ .
+ // C ------------.
+ // .
+ // 0 R1 R2 .
+ multiplyRate((int)(factor * 100), &finalFreq);
+ } else if (performTouchPositionCorrection
+ && squaredDistance == PROXIMITY_CHAR_WITHOUT_DISTANCE_INFO) {
+ multiplyRate(WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE, &finalFreq);
+ } else if (squaredDistance == ADDITIONAL_PROXIMITY_CHAR_DISTANCE_INFO) {
+ multiplyRate(WORDS_WITH_ADDITIONAL_PROXIMITY_CHARACTER_DEMOTION_RATE, &finalFreq);
+ } else if (i < adjustedProximityMatchedCount) {
multiplyRate(WORDS_WITH_PROXIMITY_CHARACTER_DEMOTION_RATE, &finalFreq);
}
}
@@ -794,7 +841,8 @@
AKLOGI("calc: %d, %d", outputLength, sameLength);
}
- if (DEBUG_CORRECTION_FREQ) {
+ if (DEBUG_CORRECTION_FREQ
+ && (INPUTLENGTH_FOR_DEBUG <= 0 || INPUTLENGTH_FOR_DEBUG == 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,
diff --git a/native/src/correction.h b/native/src/correction.h
index b246070..a711c99 100644
--- a/native/src/correction.h
+++ b/native/src/correction.h
@@ -85,7 +85,7 @@
}
}
- 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);
diff --git a/native/src/defines.h b/native/src/defines.h
index 3f3f5ba..02c1fe0 100644
--- a/native/src/defines.h
+++ b/native/src/defines.h
@@ -172,6 +172,7 @@
#define NOT_A_COORDINATE -1
#define EQUIVALENT_CHAR_WITHOUT_DISTANCE_INFO -2
#define PROXIMITY_CHAR_WITHOUT_DISTANCE_INFO -3
+#define ADDITIONAL_PROXIMITY_CHAR_DISTANCE_INFO -4
#define NOT_A_INDEX -1
#define NOT_A_FREQUENCY -1
@@ -194,6 +195,7 @@
#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_MATCH_SKIP_PROMOTION_RATE 105
#define WORDS_WITH_JUST_ONE_CORRECTION_PROMOTION_RATE 160
#define CORRECTION_COUNT_RATE_DEMOTION_RATE_BASE 45
@@ -210,6 +212,9 @@
// This is only used for the size of array. Not to be used in c functions.
#define MAX_WORD_LENGTH_INTERNAL 48
+// This must be equal to ADDITIONAL_PROXIMITY_CHAR_DELIMITER_CODE in KeyDetector.java
+#define ADDITIONAL_PROXIMITY_CHAR_DELIMITER_CODE 2
+
// Word limit for sub queues used in WordsPriorityQueuePool. Sub queues are temporary queues used
// for better performance.
// Holds up to 1 candidate for each word
@@ -241,4 +246,8 @@
// The ratio of neutral area radius to sweet spot radius.
#define NEUTRAL_AREA_RADIUS_RATIO 1.3f
+// DEBUG
+#define INPUTLENGTH_FOR_DEBUG -1
+#define MIN_OUTPUT_INDEX_FOR_DEBUG -1
+
#endif // LATINIME_DEFINES_H
diff --git a/native/src/proximity_info.cpp b/native/src/proximity_info.cpp
index e0e9380..b6bab22 100644
--- a/native/src/proximity_info.cpp
+++ b/native/src/proximity_info.cpp
@@ -261,7 +261,8 @@
// Not an exact nor an accent-alike match: search the list of close keys
int j = 1;
- while (j < MAX_PROXIMITY_CHARS_SIZE && currentChars[j] > 0) {
+ while (j < MAX_PROXIMITY_CHARS_SIZE
+ && currentChars[j] > ADDITIONAL_PROXIMITY_CHAR_DELIMITER_CODE) {
const bool matched = (currentChars[j] == baseLowerC || currentChars[j] == c);
if (matched) {
if (proximityIndex) {
@@ -271,6 +272,21 @@
}
++j;
}
+ if (j < MAX_PROXIMITY_CHARS_SIZE
+ && currentChars[j] == ADDITIONAL_PROXIMITY_CHAR_DELIMITER_CODE) {
+ ++j;
+ while (j < MAX_PROXIMITY_CHARS_SIZE
+ && currentChars[j] > ADDITIONAL_PROXIMITY_CHAR_DELIMITER_CODE) {
+ const bool matched = (currentChars[j] == baseLowerC || currentChars[j] == c);
+ if (matched) {
+ if (proximityIndex) {
+ *proximityIndex = j;
+ }
+ return ADDITIONAL_PROXIMITY_CHAR;
+ }
+ ++j;
+ }
+ }
// Was not included, signal this as an unrelated character.
return UNRELATED_CHAR;
diff --git a/native/src/proximity_info.h b/native/src/proximity_info.h
index 9ca5505..b77c1bb 100644
--- a/native/src/proximity_info.h
+++ b/native/src/proximity_info.h
@@ -38,7 +38,9 @@
// It is a char located nearby on the keyboard
NEAR_PROXIMITY_CHAR,
// It is an unrelated char
- UNRELATED_CHAR
+ UNRELATED_CHAR,
+ // Additional proximity char which can differ by language.
+ ADDITIONAL_PROXIMITY_CHAR
} ProximityType;
ProximityInfo(const int maxProximityCharsSize, const int keyboardWidth,