am 702e153f: Merge "Support migration/dump of Beginning-of-Sentence entries."

* commit '702e153fbc28d54aeb2ded40b9f3f31c1fd154e9':
  Support migration/dump of Beginning-of-Sentence entries.
This commit is contained in:
Keisuke Kuroyanagi 2014-06-24 06:18:11 +00:00 committed by Android Git Automerger
commit c1ded3dd79
9 changed files with 111 additions and 46 deletions

View file

@ -64,11 +64,12 @@ public final class BinaryDictionary extends Dictionary {
public static final int NOT_A_VALID_TIMESTAMP = -1;
// Format to get unigram flags from native side via getWordPropertyNative().
private static final int FORMAT_WORD_PROPERTY_OUTPUT_FLAG_COUNT = 4;
private static final int FORMAT_WORD_PROPERTY_OUTPUT_FLAG_COUNT = 5;
private static final int FORMAT_WORD_PROPERTY_IS_NOT_A_WORD_INDEX = 0;
private static final int FORMAT_WORD_PROPERTY_IS_BLACKLISTED_INDEX = 1;
private static final int FORMAT_WORD_PROPERTY_HAS_BIGRAMS_INDEX = 2;
private static final int FORMAT_WORD_PROPERTY_HAS_SHORTCUTS_INDEX = 3;
private static final int FORMAT_WORD_PROPERTY_IS_BEGINNING_OF_SENTENCE_INDEX = 4;
// Format to get probability and historical info from native side via getWordPropertyNative().
public static final int FORMAT_WORD_PROPERTY_OUTPUT_PROBABILITY_INFO_COUNT = 4;
@ -176,10 +177,12 @@ public final class BinaryDictionary extends Dictionary {
private static native int getBigramProbabilityNative(long dict, int[] word0,
boolean isBeginningOfSentence, int[] word1);
private static native void getWordPropertyNative(long dict, int[] word,
int[] outCodePoints, boolean[] outFlags, int[] outProbabilityInfo,
ArrayList<int[]> outBigramTargets, ArrayList<int[]> outBigramProbabilityInfo,
ArrayList<int[]> outShortcutTargets, ArrayList<Integer> outShortcutProbabilities);
private static native int getNextWordNative(long dict, int token, int[] outCodePoints);
boolean isBeginningOfSentence, int[] outCodePoints, boolean[] outFlags,
int[] outProbabilityInfo, ArrayList<int[]> outBigramTargets,
ArrayList<int[]> outBigramProbabilityInfo, ArrayList<int[]> outShortcutTargets,
ArrayList<Integer> outShortcutProbabilities);
private static native int getNextWordNative(long dict, int token, int[] outCodePoints,
boolean[] outIsBeginningOfSentence);
private static native void getSuggestionsNative(long dict, long proximityInfo,
long traverseSession, int[] xCoordinates, int[] yCoordinates, int[] times,
int[] pointerIds, int[] inputCodePoints, int inputSize, int[] suggestOptions,
@ -358,8 +361,8 @@ public final class BinaryDictionary extends Dictionary {
prevWordsInfo.mIsBeginningOfSentence, codePoints1);
}
public WordProperty getWordProperty(final String word) {
if (TextUtils.isEmpty(word)) {
public WordProperty getWordProperty(final String word, final boolean isBeginningOfSentence) {
if (word == null) {
return null;
}
final int[] codePoints = StringUtils.toCodePointArray(word);
@ -371,14 +374,15 @@ public final class BinaryDictionary extends Dictionary {
final ArrayList<int[]> outBigramProbabilityInfo = new ArrayList<>();
final ArrayList<int[]> outShortcutTargets = new ArrayList<>();
final ArrayList<Integer> outShortcutProbabilities = new ArrayList<>();
getWordPropertyNative(mNativeDict, codePoints, outCodePoints, outFlags, outProbabilityInfo,
outBigramTargets, outBigramProbabilityInfo, outShortcutTargets,
outShortcutProbabilities);
getWordPropertyNative(mNativeDict, codePoints, isBeginningOfSentence, outCodePoints,
outFlags, outProbabilityInfo, outBigramTargets, outBigramProbabilityInfo,
outShortcutTargets, outShortcutProbabilities);
return new WordProperty(codePoints,
outFlags[FORMAT_WORD_PROPERTY_IS_NOT_A_WORD_INDEX],
outFlags[FORMAT_WORD_PROPERTY_IS_BLACKLISTED_INDEX],
outFlags[FORMAT_WORD_PROPERTY_HAS_BIGRAMS_INDEX],
outFlags[FORMAT_WORD_PROPERTY_HAS_SHORTCUTS_INDEX], outProbabilityInfo,
outFlags[FORMAT_WORD_PROPERTY_HAS_SHORTCUTS_INDEX],
outFlags[FORMAT_WORD_PROPERTY_IS_BEGINNING_OF_SENTENCE_INDEX], outProbabilityInfo,
outBigramTargets, outBigramProbabilityInfo, outShortcutTargets,
outShortcutProbabilities);
}
@ -399,9 +403,12 @@ public final class BinaryDictionary extends Dictionary {
*/
public GetNextWordPropertyResult getNextWordProperty(final int token) {
final int[] codePoints = new int[Constants.DICTIONARY_MAX_WORD_LENGTH];
final int nextToken = getNextWordNative(mNativeDict, token, codePoints);
final boolean[] isBeginningOfSentence = new boolean[1];
final int nextToken = getNextWordNative(mNativeDict, token, codePoints,
isBeginningOfSentence);
final String word = StringUtils.getStringFromNullTerminatedCodePointArray(codePoints);
return new GetNextWordPropertyResult(getWordProperty(word), nextToken);
return new GetNextWordPropertyResult(
getWordProperty(word, isBeginningOfSentence[0]), nextToken);
}
// Add a unigram entry to binary dictionary with unigram attributes in native code.

View file

@ -70,8 +70,8 @@ public final class WordProperty implements Comparable<WordProperty> {
// Construct word property using information from native code.
// This represents invalid word when the probability is BinaryDictionary.NOT_A_PROBABILITY.
public WordProperty(final int[] codePoints, final boolean isNotAWord,
final boolean isBlacklisted, final boolean hasBigram,
final boolean hasShortcuts, final int[] probabilityInfo,
final boolean isBlacklisted, final boolean hasBigram, final boolean hasShortcuts,
final boolean isBeginningOfSentence, final int[] probabilityInfo,
final ArrayList<int[]> bigramTargets, final ArrayList<int[]> bigramProbabilityInfo,
final ArrayList<int[]> shortcutTargets,
final ArrayList<Integer> shortcutProbabilities) {
@ -79,7 +79,7 @@ public final class WordProperty implements Comparable<WordProperty> {
mProbabilityInfo = createProbabilityInfoFromArray(probabilityInfo);
mShortcutTargets = new ArrayList<>();
mBigrams = new ArrayList<>();
mIsBeginningOfSentence = false;
mIsBeginningOfSentence = isBeginningOfSentence;
mIsNotAWord = isNotAWord;
mIsBlacklistEntry = isBlacklisted;
mHasShortcuts = hasShortcuts;

View file

@ -31,6 +31,7 @@ public class CombinedFormatUtils {
public static final String HISTORICAL_INFO_TAG = "historicalInfo";
public static final String HISTORICAL_INFO_SEPARATOR = ":";
public static final String WORD_TAG = "word";
public static final String BEGINNING_OF_SENTENCE_TAG = "beginning_of_sentence";
public static final String NOT_A_WORD_TAG = "not_a_word";
public static final String BLACKLISTED_TAG = "blacklisted";
@ -56,6 +57,9 @@ public class CombinedFormatUtils {
builder.append(" " + WORD_TAG + "=" + wordProperty.mWord);
builder.append(",");
builder.append(formatProbabilityInfo(wordProperty.mProbabilityInfo));
if (wordProperty.mIsBeginningOfSentence) {
builder.append("," + BEGINNING_OF_SENTENCE_TAG + "=true");
}
if (wordProperty.mIsNotAWord) {
builder.append("," + NOT_A_WORD_TAG + "=true");
}

View file

@ -301,7 +301,7 @@ static jint latinime_BinaryDictionary_getBigramProbability(JNIEnv *env, jclass c
// If token is 0, this method newly starts iterating the dictionary. This method returns 0 when
// the dictionary does not have a next word.
static jint latinime_BinaryDictionary_getNextWord(JNIEnv *env, jclass clazz,
jlong dict, jint token, jintArray outCodePoints) {
jlong dict, jint token, jintArray outCodePoints, jbooleanArray outIsBeginningOfSentence) {
Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict);
if (!dictionary) return 0;
const jsize codePointBufSize = env->GetArrayLength(outCodePoints);
@ -317,19 +317,39 @@ static jint latinime_BinaryDictionary_getNextWord(JNIEnv *env, jclass clazz,
JniDataUtils::outputCodePoints(env, outCodePoints, 0 /* start */,
MAX_WORD_LENGTH /* maxLength */, wordCodePoints, wordCodePointCount,
false /* needsNullTermination */);
bool isBeginningOfSentence = false;
if (wordCodePointCount > 0 && wordCodePoints[0] == CODE_POINT_BEGINNING_OF_SENTENCE) {
isBeginningOfSentence = true;
}
JniDataUtils::putBooleanToArray(env, outIsBeginningOfSentence, 0 /* index */,
isBeginningOfSentence);
return nextToken;
}
static void latinime_BinaryDictionary_getWordProperty(JNIEnv *env, jclass clazz,
jlong dict, jintArray word, jintArray outCodePoints, jbooleanArray outFlags,
jintArray outProbabilityInfo, jobject outBigramTargets, jobject outBigramProbabilityInfo,
jobject outShortcutTargets, jobject outShortcutProbabilities) {
jlong dict, jintArray word, jboolean isBeginningOfSentence, jintArray outCodePoints,
jbooleanArray outFlags, jintArray outProbabilityInfo, jobject outBigramTargets,
jobject outBigramProbabilityInfo, jobject outShortcutTargets,
jobject outShortcutProbabilities) {
Dictionary *dictionary = reinterpret_cast<Dictionary *>(dict);
if (!dictionary) return;
const jsize wordLength = env->GetArrayLength(word);
int wordCodePoints[wordLength];
if (wordLength > MAX_WORD_LENGTH) {
AKLOGE("Invalid wordLength: %d", wordLength);
return;
}
int wordCodePoints[MAX_WORD_LENGTH];
env->GetIntArrayRegion(word, 0, wordLength, wordCodePoints);
const WordProperty wordProperty = dictionary->getWordProperty(wordCodePoints, wordLength);
int codePointCount = wordLength;
if (isBeginningOfSentence) {
codePointCount = CharUtils::attachBeginningOfSentenceMarker(
wordCodePoints, wordLength, MAX_WORD_LENGTH);
if (codePointCount < 0) {
AKLOGE("Cannot attach Beginning-of-Sentence marker.");
return;
}
}
const WordProperty wordProperty = dictionary->getWordProperty(wordCodePoints, codePointCount);
wordProperty.outputProperties(env, outCodePoints, outFlags, outProbabilityInfo,
outBigramTargets, outBigramProbabilityInfo, outShortcutTargets,
outShortcutProbabilities);
@ -554,7 +574,6 @@ static bool latinime_BinaryDictionary_migrateNative(JNIEnv *env, jclass clazz, j
return false;
}
// TODO: Migrate historical information.
int wordCodePoints[MAX_WORD_LENGTH];
int wordCodePointCount = 0;
int token = 0;
@ -563,6 +582,10 @@ static bool latinime_BinaryDictionary_migrateNative(JNIEnv *env, jclass clazz, j
token = dictionary->getNextWordAndNextToken(token, wordCodePoints, &wordCodePointCount);
const WordProperty wordProperty = dictionary->getWordProperty(wordCodePoints,
wordCodePointCount);
if (wordCodePoints[0] == CODE_POINT_BEGINNING_OF_SENTENCE) {
// Skip beginning-of-sentence unigram.
continue;
}
if (dictionaryStructureWithBufferPolicy->needsToRunGC(true /* mindsBlockByGC */)) {
dictionaryStructureWithBufferPolicy = runGCAndGetNewStructurePolicy(
std::move(dictionaryStructureWithBufferPolicy), dictFilePathChars);
@ -592,7 +615,7 @@ static bool latinime_BinaryDictionary_migrateNative(JNIEnv *env, jclass clazz, j
}
}
const PrevWordsInfo prevWordsInfo(wordCodePoints, wordCodePointCount,
false /* isBeginningOfSentence */);
wordProperty.getUnigramProperty()->representsBeginningOfSentence());
for (const BigramProperty &bigramProperty : *wordProperty.getBigramProperties()) {
if (!dictionaryStructureWithBufferPolicy->addNgramEntry(&prevWordsInfo,
&bigramProperty)) {
@ -669,13 +692,13 @@ static const JNINativeMethod sMethods[] = {
},
{
const_cast<char *>("getWordPropertyNative"),
const_cast<char *>("(J[I[I[Z[ILjava/util/ArrayList;Ljava/util/ArrayList;"
const_cast<char *>("(J[IZ[I[Z[ILjava/util/ArrayList;Ljava/util/ArrayList;"
"Ljava/util/ArrayList;Ljava/util/ArrayList;)V"),
reinterpret_cast<void *>(latinime_BinaryDictionary_getWordProperty)
},
{
const_cast<char *>("getNextWordNative"),
const_cast<char *>("(JI[I)I"),
const_cast<char *>("(JI[I[Z)I"),
reinterpret_cast<void *>(latinime_BinaryDictionary_getNextWord)
},
{

View file

@ -28,7 +28,8 @@ void WordProperty::outputProperties(JNIEnv *const env, jintArray outCodePoints,
MAX_WORD_LENGTH /* maxLength */, mCodePoints.data(), mCodePoints.size(),
false /* needsNullTermination */);
jboolean flags[] = {mUnigramProperty.isNotAWord(), mUnigramProperty.isBlacklisted(),
!mBigrams.empty(), mUnigramProperty.hasShortcuts()};
!mBigrams.empty(), mUnigramProperty.hasShortcuts(),
mUnigramProperty.representsBeginningOfSentence()};
env->SetBooleanArrayRegion(outFlags, 0 /* start */, NELEMS(flags), flags);
int probabilityInfo[] = {mUnigramProperty.getProbability(), mUnigramProperty.getTimestamp(),
mUnigramProperty.getLevel(), mUnigramProperty.getCount()};

View file

@ -98,6 +98,10 @@ class CharUtils {
// Beginning-of-Sentence.
static AK_FORCE_INLINE int attachBeginningOfSentenceMarker(int *const codePoints,
const int codePointCount, const int maxCodePoint) {
if (codePointCount > 0 && codePoints[0] == CODE_POINT_BEGINNING_OF_SENTENCE) {
// Marker has already been attached.
return codePointCount;
}
if (codePointCount >= maxCodePoint) {
// the code points cannot be marked as a Beginning-of-Sentence.
return 0;

View file

@ -69,18 +69,23 @@ class JniDataUtils {
static void outputCodePoints(JNIEnv *env, jintArray intArrayToOutputCodePoints, const int start,
const int maxLength, const int *const codePoints, const int codePointCount,
const bool needsNullTermination) {
const int outputCodePointCount = std::min(maxLength, codePointCount);
int outputCodePonts[outputCodePointCount];
for (int i = 0; i < outputCodePointCount; ++i) {
const int codePointBufSize = std::min(maxLength, codePointCount);
int outputCodePonts[codePointBufSize];
int outputCodePointCount = 0;
for (int i = 0; i < codePointBufSize; ++i) {
const int codePoint = codePoints[i];
int codePointToOutput = codePoint;
if (!CharUtils::isInUnicodeSpace(codePoint)) {
outputCodePonts[i] = CODE_POINT_REPLACEMENT_CHARACTER;
if (codePoint == CODE_POINT_BEGINNING_OF_SENTENCE) {
// Just skip Beginning-of-Sentence marker.
continue;
}
codePointToOutput = CODE_POINT_REPLACEMENT_CHARACTER;
} else if (codePoint >= 0x01 && codePoint <= 0x1F) {
// Control code.
outputCodePonts[i] = CODE_POINT_REPLACEMENT_CHARACTER;
} else {
outputCodePonts[i] = codePoint;
codePointToOutput = CODE_POINT_REPLACEMENT_CHARACTER;
}
outputCodePonts[outputCodePointCount++] = codePointToOutput;
}
env->SetIntArrayRegion(intArrayToOutputCodePoints, start, outputCodePointCount,
outputCodePonts);
@ -90,6 +95,11 @@ class JniDataUtils {
}
}
static void putBooleanToArray(JNIEnv *env, jbooleanArray array, const int index,
const jboolean value) {
env->SetBooleanArrayRegion(array, index, 1 /* len */, &value);
}
static void putIntToArray(JNIEnv *env, jintArray array, const int index, const int value) {
env->SetIntArrayRegion(array, index, 1 /* len */, &value);
}

View file

@ -994,7 +994,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
0 /* offset */, dictFile.length(), true /* useFullEditDistance */,
Locale.getDefault(), TEST_LOCALE, true /* isUpdatable */);
final WordProperty invalidWordProperty = binaryDictionary.getWordProperty("dummyWord");
final WordProperty invalidWordProperty = binaryDictionary.getWordProperty("dummyWord",
false /* isBeginningOfSentence */);
assertFalse(invalidWordProperty.isValid());
final ArrayList<String> words = new ArrayList<>();
@ -1017,7 +1018,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
}
words.add(word);
wordProbabilities.put(word, unigramProbability);
final WordProperty wordProperty = binaryDictionary.getWordProperty(word);
final WordProperty wordProperty = binaryDictionary.getWordProperty(word,
false /* isBeginningOfSentence */);
assertEquals(word, wordProperty.mWord);
assertTrue(wordProperty.isValid());
assertEquals(isNotAWord, wordProperty.mIsNotAWord);
@ -1057,7 +1059,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
continue;
}
final HashSet<String> bigramWord1s = bigrams.get(word0);
final WordProperty wordProperty = binaryDictionary.getWordProperty(word0);
final WordProperty wordProperty = binaryDictionary.getWordProperty(word0,
false /* isBeginningOfSentence */);
assertEquals(bigramWord1s.size(), wordProperty.mBigrams.size());
for (int j = 0; j < wordProperty.mBigrams.size(); j++) {
final String word1 = wordProperty.mBigrams.get(j).mWord;
@ -1094,7 +1097,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
0 /* offset */, dictFile.length(), true /* useFullEditDistance */,
Locale.getDefault(), TEST_LOCALE, true /* isUpdatable */);
final WordProperty invalidWordProperty = binaryDictionary.getWordProperty("dummyWord");
final WordProperty invalidWordProperty = binaryDictionary.getWordProperty("dummyWord",
false /* isBeginningOfSentence */);
assertFalse(invalidWordProperty.isValid());
final ArrayList<String> words = new ArrayList<>();
@ -1188,7 +1192,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
binaryDictionary.addUnigramEntry("aaa", unigramProbability, "zzz",
shortcutProbability, false /* isBeginningOfSentence */,
false /* isNotAWord */, false /* isBlacklisted */, 0 /* timestamp */);
WordProperty wordProperty = binaryDictionary.getWordProperty("aaa");
WordProperty wordProperty = binaryDictionary.getWordProperty("aaa",
false /* isBeginningOfSentence */);
assertEquals(1, wordProperty.mShortcutTargets.size());
assertEquals("zzz", wordProperty.mShortcutTargets.get(0).mWord);
assertEquals(shortcutProbability, wordProperty.mShortcutTargets.get(0).getProbability());
@ -1196,7 +1201,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
binaryDictionary.addUnigramEntry("aaa", unigramProbability, "zzz",
updatedShortcutProbability, false /* isBeginningOfSentence */,
false /* isNotAWord */, false /* isBlacklisted */, 0 /* timestamp */);
wordProperty = binaryDictionary.getWordProperty("aaa");
wordProperty = binaryDictionary.getWordProperty("aaa",
false /* isBeginningOfSentence */);
assertEquals(1, wordProperty.mShortcutTargets.size());
assertEquals("zzz", wordProperty.mShortcutTargets.get(0).mWord);
assertEquals(updatedShortcutProbability,
@ -1207,7 +1213,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
final HashMap<String, Integer> shortcutTargets = new HashMap<>();
shortcutTargets.put("zzz", updatedShortcutProbability);
shortcutTargets.put("yyy", shortcutProbability);
wordProperty = binaryDictionary.getWordProperty("aaa");
wordProperty = binaryDictionary.getWordProperty("aaa",
false /* isBeginningOfSentence */);
assertEquals(2, wordProperty.mShortcutTargets.size());
for (WeightedString shortcutTarget : wordProperty.mShortcutTargets) {
assertTrue(shortcutTargets.containsKey(shortcutTarget.mWord));
@ -1218,7 +1225,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
shortcutTargets.put("zzz", updatedShortcutProbability);
shortcutTargets.put("yyy", shortcutProbability);
binaryDictionary.flushWithGC();
wordProperty = binaryDictionary.getWordProperty("aaa");
wordProperty = binaryDictionary.getWordProperty("aaa",
false /* isBeginningOfSentence */);
assertEquals(2, wordProperty.mShortcutTargets.size());
for (WeightedString shortcutTarget : wordProperty.mShortcutTargets) {
assertTrue(shortcutTargets.containsKey(shortcutTarget.mWord));
@ -1288,7 +1296,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
}
for (final String word : words) {
final WordProperty wordProperty = binaryDictionary.getWordProperty(word);
final WordProperty wordProperty = binaryDictionary.getWordProperty(word,
false /* isBeginningOfSentence */);
assertEquals((int)unigramProbabilities.get(word),
wordProperty.mProbabilityInfo.mProbability);
if (!shortcutTargets.containsKey(word)) {
@ -1332,6 +1341,8 @@ public class BinaryDictionaryTests extends AndroidTestCase {
binaryDictionary.addUnigramEntry("ddd", unigramProbability, null /* shortcutTarget */,
Dictionary.NOT_A_PROBABILITY, false /* isBeginningOfSentence */,
true /* isNotAWord */, true /* isBlacklisted */, 0 /* timestamp */);
binaryDictionary.addNgramEntry(PrevWordsInfo.BEGINNING_OF_SENTENCE,
"aaa", bigramProbability, 0 /* timestamp */);
assertEquals(unigramProbability, binaryDictionary.getFrequency("aaa"));
assertEquals(unigramProbability, binaryDictionary.getFrequency("bbb"));
assertTrue(isValidBigram(binaryDictionary, "aaa", "bbb"));
@ -1343,12 +1354,16 @@ public class BinaryDictionaryTests extends AndroidTestCase {
assertEquals(unigramProbability, binaryDictionary.getFrequency("bbb"));
if (canCheckBigramProbability(toFormatVersion)) {
assertEquals(bigramProbability, getBigramProbability(binaryDictionary, "aaa", "bbb"));
assertEquals(bigramProbability, binaryDictionary.getNgramProbability(
PrevWordsInfo.BEGINNING_OF_SENTENCE, "aaa"));
}
assertTrue(isValidBigram(binaryDictionary, "aaa", "bbb"));
WordProperty wordProperty = binaryDictionary.getWordProperty("ccc");
WordProperty wordProperty = binaryDictionary.getWordProperty("ccc",
false /* isBeginningOfSentence */);
assertEquals(1, wordProperty.mShortcutTargets.size());
assertEquals("xxx", wordProperty.mShortcutTargets.get(0).mWord);
wordProperty = binaryDictionary.getWordProperty("ddd");
wordProperty = binaryDictionary.getWordProperty("ddd",
false /* isBeginningOfSentence */);
assertTrue(wordProperty.mIsBlacklistEntry);
assertTrue(wordProperty.mIsNotAWord);
}

View file

@ -614,7 +614,8 @@ public class BinaryDictDecoderEncoderTests extends AndroidTestCase {
0 /* offset */, file.length(), true /* useFullEditDistance */,
Locale.ENGLISH, dictName, false /* isUpdatable */);
for (final String word : words) {
final WordProperty wordProperty = binaryDictionary.getWordProperty(word);
final WordProperty wordProperty = binaryDictionary.getWordProperty(word,
false /* isBeginningOfSentence */);
assertEquals(word, wordProperty.mWord);
assertEquals(UNIGRAM_FREQ, wordProperty.getProbability());
if (shortcuts.containsKey(word)) {