diff --git a/java/proguard.flags b/java/proguard.flags
index 35b3ac3..0c5ad9f 100644
--- a/java/proguard.flags
+++ b/java/proguard.flags
@@ -19,6 +19,6 @@
 # to preserve changing those methods' signature.
 -keep class com.android.inputmethod.latin.AssetFileAddress
 -keep class com.android.inputmethod.latin.Dictionary
--keep class com.android.inputmethod.latin.PrevWordsInfo
+-keep class com.android.inputmethod.latin.NgramContext
 -keep class com.android.inputmethod.latin.makedict.ProbabilityInfo
 -keep class com.android.inputmethod.latin.utils.LanguageModelParam
diff --git a/java/res/values/strings-talkback-descriptions.xml b/java/res/values/strings-talkback-descriptions.xml
index 80c7bdb..36fa7b3 100644
--- a/java/res/values/strings-talkback-descriptions.xml
+++ b/java/res/values/strings-talkback-descriptions.xml
@@ -139,6 +139,18 @@
     <string name="spoken_symbol_unknown">Unknown symbol</string>
     <!-- Spoken description for unknown emoji code point. -->
     <string name="spoken_emoji_unknown">Unknown emoji</string>
+    <!-- Spoken description for emoticons ":-!". -->
+    <string name="spoken_emoticon_3A_2D_21_20">Bored face</string>
+    <!-- Spoken description for emoticons ":-$". -->
+    <string name="spoken_emoticon_3A_2D_24_20">Embarrassed face</string>
+    <!-- Spoken description for emoticons "B-)". -->
+    <string name="spoken_emoticon_42_2D_29_20">Face wearing sunglasses</string>
+    <!-- Spoken description for emoticons ":O". -->
+    <string name="spoken_emoticon_3A_4F_20">Surprised face</string>
+    <!-- Spoken description for emoticons ":-*". -->
+    <string name="spoken_emoticon_3A_2D_2A_20">Kissing face</string>
+    <!-- Spoken description for emoticons ":-[". -->
+    <string name="spoken_emoticon_3A_2D_5B_20">Frowning face</string>
 
     <!-- Spoken descriptions when opening a more keys keyboard that has alternative characters. -->
     <string name="spoken_open_more_keys_keyboard">Alternative characters are available</string>
diff --git a/java/src/com/android/inputmethod/accessibility/KeyCodeDescriptionMapper.java b/java/src/com/android/inputmethod/accessibility/KeyCodeDescriptionMapper.java
index 7a3510e..edcdd4c 100644
--- a/java/src/com/android/inputmethod/accessibility/KeyCodeDescriptionMapper.java
+++ b/java/src/com/android/inputmethod/accessibility/KeyCodeDescriptionMapper.java
@@ -37,6 +37,8 @@
     private static final String SPOKEN_LETTER_RESOURCE_NAME_FORMAT = "spoken_accented_letter_%04X";
     private static final String SPOKEN_SYMBOL_RESOURCE_NAME_FORMAT = "spoken_symbol_%04X";
     private static final String SPOKEN_EMOJI_RESOURCE_NAME_FORMAT = "spoken_emoji_%04X";
+    private static final String SPOKEN_EMOTICON_RESOURCE_NAME_PREFIX = "spoken_emoticon";
+    private static final String SPOKEN_EMOTICON_CODE_POINT_FORMAT = "_%02X";
 
     // The resource ID of the string spoken for obscured keys
     private static final int OBSCURED_KEY_RES_ID = R.string.spoken_description_dot;
@@ -109,7 +111,9 @@
         }
 
         if (code == Constants.CODE_OUTPUT_TEXT) {
-            return key.getOutputText();
+            final String outputText = key.getOutputText();
+            final String description = getSpokenEmoticonDescription(context, outputText);
+            return TextUtils.isEmpty(description) ? outputText : description;
         }
 
         // Just attempt to speak the description.
@@ -340,4 +344,21 @@
         }
         return resId;
     }
+
+    // TODO: Remove this method once TTS supports emoticon verbalization.
+    private String getSpokenEmoticonDescription(final Context context, final String outputText) {
+        final StringBuilder sb = new StringBuilder(SPOKEN_EMOTICON_RESOURCE_NAME_PREFIX);
+        final int textLength = outputText.length();
+        for (int index = 0; index < textLength; index = outputText.offsetByCodePoints(index, 1)) {
+            final int codePoint = outputText.codePointAt(index);
+            sb.append(String.format(Locale.ROOT, SPOKEN_EMOTICON_CODE_POINT_FORMAT, codePoint));
+        }
+        final String resourceName = sb.toString();
+        final Resources resources = context.getResources();
+        // Note that the resource package name may differ from the context package name.
+        final String resourcePackageName = resources.getResourcePackageName(
+                R.string.spoken_description_unknown);
+        final int resId = resources.getIdentifier(resourceName, "string", resourcePackageName);
+        return (resId == 0) ? null : resources.getString(resId);
+    }
 }
diff --git a/java/src/com/android/inputmethod/latin/BinaryDictionary.java b/java/src/com/android/inputmethod/latin/BinaryDictionary.java
index b164c17..1da33ed 100644
--- a/java/src/com/android/inputmethod/latin/BinaryDictionary.java
+++ b/java/src/com/android/inputmethod/latin/BinaryDictionary.java
@@ -255,7 +255,7 @@
 
     @Override
     public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion,
             final int sessionId, final float weightForLocale,
             final float[] inOutWeightOfLangModelVsSpatialModel) {
@@ -264,7 +264,7 @@
         }
         final DicTraverseSession session = getTraverseSession(sessionId);
         Arrays.fill(session.mInputCodePoints, Constants.NOT_A_CODE);
-        prevWordsInfo.outputToArray(session.mPrevWordCodePointArrays,
+        ngramContext.outputToArray(session.mPrevWordCodePointArrays,
                 session.mIsBeginningOfSentenceArray);
         final InputPointers inputPointers = composer.getInputPointers();
         final boolean isGesture = composer.isBatchMode();
@@ -299,7 +299,7 @@
                 inputPointers.getYCoordinates(), inputPointers.getTimes(),
                 inputPointers.getPointerIds(), session.mInputCodePoints, inputSize,
                 session.mNativeSuggestOptions.getOptions(), session.mPrevWordCodePointArrays,
-                session.mIsBeginningOfSentenceArray, prevWordsInfo.getPrevWordCount(),
+                session.mIsBeginningOfSentenceArray, ngramContext.getPrevWordCount(),
                 session.mOutputSuggestionCount, session.mOutputCodePoints, session.mOutputScores,
                 session.mSpaceIndices, session.mOutputTypes,
                 session.mOutputAutoCommitFirstWordConfidence,
@@ -357,17 +357,17 @@
     }
 
     @UsedForTesting
-    public boolean isValidNgram(final PrevWordsInfo prevWordsInfo, final String word) {
-        return getNgramProbability(prevWordsInfo, word) != NOT_A_PROBABILITY;
+    public boolean isValidNgram(final NgramContext ngramContext, final String word) {
+        return getNgramProbability(ngramContext, word) != NOT_A_PROBABILITY;
     }
 
-    public int getNgramProbability(final PrevWordsInfo prevWordsInfo, final String word) {
-        if (!prevWordsInfo.isValid() || TextUtils.isEmpty(word)) {
+    public int getNgramProbability(final NgramContext ngramContext, final String word) {
+        if (!ngramContext.isValid() || TextUtils.isEmpty(word)) {
             return NOT_A_PROBABILITY;
         }
-        final int[][] prevWordCodePointArrays = new int[prevWordsInfo.getPrevWordCount()][];
-        final boolean[] isBeginningOfSentenceArray = new boolean[prevWordsInfo.getPrevWordCount()];
-        prevWordsInfo.outputToArray(prevWordCodePointArrays, isBeginningOfSentenceArray);
+        final int[][] prevWordCodePointArrays = new int[ngramContext.getPrevWordCount()][];
+        final boolean[] isBeginningOfSentenceArray = new boolean[ngramContext.getPrevWordCount()];
+        ngramContext.outputToArray(prevWordCodePointArrays, isBeginningOfSentenceArray);
         final int[] wordCodePoints = StringUtils.toCodePointArray(word);
         return getNgramProbabilityNative(mNativeDict, prevWordCodePointArrays,
                 isBeginningOfSentenceArray, wordCodePoints);
@@ -456,14 +456,14 @@
     }
 
     // Add an n-gram entry to the binary dictionary with timestamp in native code.
-    public boolean addNgramEntry(final PrevWordsInfo prevWordsInfo, final String word,
+    public boolean addNgramEntry(final NgramContext ngramContext, final String word,
             final int probability, final int timestamp) {
-        if (!prevWordsInfo.isValid() || TextUtils.isEmpty(word)) {
+        if (!ngramContext.isValid() || TextUtils.isEmpty(word)) {
             return false;
         }
-        final int[][] prevWordCodePointArrays = new int[prevWordsInfo.getPrevWordCount()][];
-        final boolean[] isBeginningOfSentenceArray = new boolean[prevWordsInfo.getPrevWordCount()];
-        prevWordsInfo.outputToArray(prevWordCodePointArrays, isBeginningOfSentenceArray);
+        final int[][] prevWordCodePointArrays = new int[ngramContext.getPrevWordCount()][];
+        final boolean[] isBeginningOfSentenceArray = new boolean[ngramContext.getPrevWordCount()];
+        ngramContext.outputToArray(prevWordCodePointArrays, isBeginningOfSentenceArray);
         final int[] wordCodePoints = StringUtils.toCodePointArray(word);
         if (!addNgramEntryNative(mNativeDict, prevWordCodePointArrays,
                 isBeginningOfSentenceArray, wordCodePoints, probability, timestamp)) {
@@ -474,13 +474,13 @@
     }
 
     // Remove an n-gram entry from the binary dictionary in native code.
-    public boolean removeNgramEntry(final PrevWordsInfo prevWordsInfo, final String word) {
-        if (!prevWordsInfo.isValid() || TextUtils.isEmpty(word)) {
+    public boolean removeNgramEntry(final NgramContext ngramContext, final String word) {
+        if (!ngramContext.isValid() || TextUtils.isEmpty(word)) {
             return false;
         }
-        final int[][] prevWordCodePointArrays = new int[prevWordsInfo.getPrevWordCount()][];
-        final boolean[] isBeginningOfSentenceArray = new boolean[prevWordsInfo.getPrevWordCount()];
-        prevWordsInfo.outputToArray(prevWordCodePointArrays, isBeginningOfSentenceArray);
+        final int[][] prevWordCodePointArrays = new int[ngramContext.getPrevWordCount()][];
+        final boolean[] isBeginningOfSentenceArray = new boolean[ngramContext.getPrevWordCount()];
+        ngramContext.outputToArray(prevWordCodePointArrays, isBeginningOfSentenceArray);
         final int[] wordCodePoints = StringUtils.toCodePointArray(word);
         if (!removeNgramEntryNative(mNativeDict, prevWordCodePointArrays,
                 isBeginningOfSentenceArray, wordCodePoints)) {
diff --git a/java/src/com/android/inputmethod/latin/ContactsBinaryDictionary.java b/java/src/com/android/inputmethod/latin/ContactsBinaryDictionary.java
index 162a209..78c6cbd 100644
--- a/java/src/com/android/inputmethod/latin/ContactsBinaryDictionary.java
+++ b/java/src/com/android/inputmethod/latin/ContactsBinaryDictionary.java
@@ -218,7 +218,7 @@
      */
     private void addNameLocked(final String name) {
         int len = StringUtils.codePointCount(name);
-        PrevWordsInfo prevWordsInfo = PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+        NgramContext ngramContext = NgramContext.EMPTY_PREV_WORDS_INFO;
         // TODO: Better tokenization for non-Latin writing systems
         for (int i = 0; i < len; i++) {
             if (Character.isLetter(name.codePointAt(i))) {
@@ -233,19 +233,19 @@
                 final int wordLen = StringUtils.codePointCount(word);
                 if (wordLen <= MAX_WORD_LENGTH && wordLen > 1) {
                     if (DEBUG) {
-                        Log.d(TAG, "addName " + name + ", " + word + ", "  + prevWordsInfo);
+                        Log.d(TAG, "addName " + name + ", " + word + ", "  + ngramContext);
                     }
                     runGCIfRequiredLocked(true /* mindsBlockByGC */);
                     addUnigramLocked(word, FREQUENCY_FOR_CONTACTS,
                             null /* shortcut */, 0 /* shortcutFreq */, false /* isNotAWord */,
                             false /* isBlacklisted */, BinaryDictionary.NOT_A_VALID_TIMESTAMP);
-                    if (!prevWordsInfo.isValid() && mUseFirstLastBigrams) {
+                    if (!ngramContext.isValid() && mUseFirstLastBigrams) {
                         runGCIfRequiredLocked(true /* mindsBlockByGC */);
-                        addNgramEntryLocked(prevWordsInfo, word, FREQUENCY_FOR_CONTACTS_BIGRAM,
+                        addNgramEntryLocked(ngramContext, word, FREQUENCY_FOR_CONTACTS_BIGRAM,
                                 BinaryDictionary.NOT_A_VALID_TIMESTAMP);
                     }
-                    prevWordsInfo = prevWordsInfo.getNextPrevWordsInfo(
-                            new PrevWordsInfo.WordInfo(word));
+                    ngramContext = ngramContext.getNextNgramContext(
+                            new NgramContext.WordInfo(word));
                 }
             }
         }
diff --git a/java/src/com/android/inputmethod/latin/Dictionary.java b/java/src/com/android/inputmethod/latin/Dictionary.java
index b58a52b..43561ba 100644
--- a/java/src/com/android/inputmethod/latin/Dictionary.java
+++ b/java/src/com/android/inputmethod/latin/Dictionary.java
@@ -81,10 +81,9 @@
     }
 
     /**
-     * Searches for suggestions for a given context. For the moment the context is only the
-     * previous word.
+     * Searches for suggestions for a given context.
      * @param composer the key sequence to match with coordinate info, as a WordComposer
-     * @param prevWordsInfo the information of previous words.
+     * @param ngramContext the context for n-gram.
      * @param proximityInfo the object for key proximity. May be ignored by some implementations.
      * @param settingsValuesForSuggestion the settings values used for the suggestion.
      * @param sessionId the session id.
@@ -96,7 +95,7 @@
      * @return the list of suggestions (possibly null if none)
      */
     abstract public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion,
             final int sessionId, final float weightForLocale,
             final float[] inOutWeightOfLangModelVsSpatialModel);
@@ -191,7 +190,7 @@
 
         @Override
         public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer,
-                final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+                final NgramContext ngramContext, final ProximityInfo proximityInfo,
                 final SettingsValuesForSuggestion settingsValuesForSuggestion,
                 final int sessionId, final float weightForLocale,
                 final float[] inOutWeightOfLangModelVsSpatialModel) {
diff --git a/java/src/com/android/inputmethod/latin/DictionaryCollection.java b/java/src/com/android/inputmethod/latin/DictionaryCollection.java
index b26b378..a6d7205 100644
--- a/java/src/com/android/inputmethod/latin/DictionaryCollection.java
+++ b/java/src/com/android/inputmethod/latin/DictionaryCollection.java
@@ -60,7 +60,7 @@
 
     @Override
     public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion,
             final int sessionId, final float weightForLocale,
             final float[] inOutWeightOfLangModelVsSpatialModel) {
@@ -69,13 +69,13 @@
         // To avoid creating unnecessary objects, we get the list out of the first
         // dictionary and add the rest to it if not null, hence the get(0)
         ArrayList<SuggestedWordInfo> suggestions = dictionaries.get(0).getSuggestions(composer,
-                prevWordsInfo, proximityInfo, settingsValuesForSuggestion, sessionId,
+                ngramContext, proximityInfo, settingsValuesForSuggestion, sessionId,
                 weightForLocale, inOutWeightOfLangModelVsSpatialModel);
         if (null == suggestions) suggestions = new ArrayList<>();
         final int length = dictionaries.size();
         for (int i = 1; i < length; ++ i) {
             final ArrayList<SuggestedWordInfo> sugg = dictionaries.get(i).getSuggestions(composer,
-                    prevWordsInfo, proximityInfo, settingsValuesForSuggestion, sessionId,
+                    ngramContext, proximityInfo, settingsValuesForSuggestion, sessionId,
                     weightForLocale, inOutWeightOfLangModelVsSpatialModel);
             if (null != sugg) suggestions.addAll(sugg);
         }
diff --git a/java/src/com/android/inputmethod/latin/DictionaryFacilitator.java b/java/src/com/android/inputmethod/latin/DictionaryFacilitator.java
index aa15bd6..6dc1e82 100644
--- a/java/src/com/android/inputmethod/latin/DictionaryFacilitator.java
+++ b/java/src/com/android/inputmethod/latin/DictionaryFacilitator.java
@@ -25,7 +25,7 @@
 import com.android.inputmethod.annotations.UsedForTesting;
 import com.android.inputmethod.keyboard.ProximityInfo;
 import com.android.inputmethod.latin.ExpandableBinaryDictionary.AddMultipleDictionaryEntriesCallback;
-import com.android.inputmethod.latin.PrevWordsInfo.WordInfo;
+import com.android.inputmethod.latin.NgramContext.WordInfo;
 import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo;
 import com.android.inputmethod.latin.personalization.ContextualDictionary;
 import com.android.inputmethod.latin.personalization.PersonalizationDataChunk;
@@ -509,23 +509,23 @@
     }
 
     public void addToUserHistory(final String suggestion, final boolean wasAutoCapitalized,
-            final PrevWordsInfo prevWordsInfo, final int timeStampInSeconds,
+            final NgramContext ngramContext, final int timeStampInSeconds,
             final boolean blockPotentiallyOffensive) {
         final DictionaryGroup dictionaryGroup = getDictionaryGroupForActiveLanguage();
         final String[] words = suggestion.split(Constants.WORD_SEPARATOR);
-        PrevWordsInfo prevWordsInfoForCurrentWord = prevWordsInfo;
+        NgramContext ngramContextForCurrentWord = ngramContext;
         for (int i = 0; i < words.length; i++) {
             final String currentWord = words[i];
             final boolean wasCurrentWordAutoCapitalized = (i == 0) ? wasAutoCapitalized : false;
-            addWordToUserHistory(dictionaryGroup, prevWordsInfoForCurrentWord, currentWord,
+            addWordToUserHistory(dictionaryGroup, ngramContextForCurrentWord, currentWord,
                     wasCurrentWordAutoCapitalized, timeStampInSeconds, blockPotentiallyOffensive);
-            prevWordsInfoForCurrentWord =
-                    prevWordsInfoForCurrentWord.getNextPrevWordsInfo(new WordInfo(currentWord));
+            ngramContextForCurrentWord =
+                    ngramContextForCurrentWord.getNextNgramContext(new WordInfo(currentWord));
         }
     }
 
     private void addWordToUserHistory(final DictionaryGroup dictionaryGroup,
-            final PrevWordsInfo prevWordsInfo, final String word, final boolean wasAutoCapitalized,
+            final NgramContext ngramContext, final String word, final boolean wasAutoCapitalized,
             final int timeStampInSeconds, final boolean blockPotentiallyOffensive) {
         final ExpandableBinaryDictionary userHistoryDictionary =
                 dictionaryGroup.getSubDict(Dictionary.TYPE_USER_HISTORY);
@@ -571,7 +571,7 @@
         // We demote unrecognized words (frequency < 0, below) by specifying them as "invalid".
         // We don't add words with 0-frequency (assuming they would be profanity etc.).
         final boolean isValid = maxFreq > 0;
-        UserHistoryDictionary.addToDictionary(userHistoryDictionary, prevWordsInfo, secondWord,
+        UserHistoryDictionary.addToDictionary(userHistoryDictionary, ngramContext, secondWord,
                 isValid, timeStampInSeconds,
                 new DistracterFilterCheckingIsInDictionary(
                         mDistracterFilter, userHistoryDictionary));
@@ -593,12 +593,11 @@
 
     // TODO: Revise the way to fusion suggestion results.
     public SuggestionResults getSuggestionResults(final WordComposer composer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion, final int sessionId) {
         final DictionaryGroup[] dictionaryGroups = mDictionaryGroups;
         final SuggestionResults suggestionResults = new SuggestionResults(
-                SuggestedWords.MAX_SUGGESTIONS,
-                prevWordsInfo.mPrevWordsInfo[0].mIsBeginningOfSentence);
+                SuggestedWords.MAX_SUGGESTIONS, ngramContext.isBeginningOfSentenceContext());
         final float[] weightOfLangModelVsSpatialModel =
                 new float[] { Dictionary.NOT_A_WEIGHT_OF_LANG_MODEL_VS_SPATIAL_MODEL };
         for (final DictionaryGroup dictionaryGroup : dictionaryGroups) {
@@ -606,7 +605,7 @@
                 final Dictionary dictionary = dictionaryGroup.getDict(dictType);
                 if (null == dictionary) continue;
                 final ArrayList<SuggestedWordInfo> dictionarySuggestions =
-                        dictionary.getSuggestions(composer, prevWordsInfo, proximityInfo,
+                        dictionary.getSuggestions(composer, ngramContext, proximityInfo,
                                 settingsValuesForSuggestion, sessionId,
                                 dictionaryGroup.mWeightForLocale, weightOfLangModelVsSpatialModel);
                 if (null == dictionarySuggestions) continue;
@@ -720,7 +719,7 @@
         if (contextualDict == null) {
             return;
         }
-        PrevWordsInfo prevWordsInfo = PrevWordsInfo.BEGINNING_OF_SENTENCE;
+        NgramContext ngramContext = NgramContext.BEGINNING_OF_SENTENCE;
         for (int i = 0; i < phrase.length; i++) {
             final String[] subPhrase = Arrays.copyOfRange(phrase, i /* start */, phrase.length);
             final String subPhraseStr = TextUtils.join(Constants.WORD_SEPARATOR, subPhrase);
@@ -730,7 +729,7 @@
                     false /* isNotAWord */, false /* isBlacklisted */,
                     BinaryDictionary.NOT_A_VALID_TIMESTAMP,
                     DistracterFilter.EMPTY_DISTRACTER_FILTER);
-            contextualDict.addNgramEntry(prevWordsInfo, subPhraseStr,
+            contextualDict.addNgramEntry(ngramContext, subPhraseStr,
                     bigramProbabilityForPhrases, BinaryDictionary.NOT_A_VALID_TIMESTAMP);
 
             if (i < phrase.length - 1) {
@@ -740,11 +739,11 @@
                         false /* isNotAWord */, false /* isBlacklisted */,
                         BinaryDictionary.NOT_A_VALID_TIMESTAMP,
                         DistracterFilter.EMPTY_DISTRACTER_FILTER);
-                contextualDict.addNgramEntry(prevWordsInfo, phrase[i],
+                contextualDict.addNgramEntry(ngramContext, phrase[i],
                         bigramProbabilityForWords, BinaryDictionary.NOT_A_VALID_TIMESTAMP);
             }
-            prevWordsInfo =
-                    prevWordsInfo.getNextPrevWordsInfo(new PrevWordsInfo.WordInfo(phrase[i]));
+            ngramContext =
+                    ngramContext.getNextNgramContext(new NgramContext.WordInfo(phrase[i]));
         }
     }
 
diff --git a/java/src/com/android/inputmethod/latin/ExpandableBinaryDictionary.java b/java/src/com/android/inputmethod/latin/ExpandableBinaryDictionary.java
index 68f2b62..1bdadc3 100644
--- a/java/src/com/android/inputmethod/latin/ExpandableBinaryDictionary.java
+++ b/java/src/com/android/inputmethod/latin/ExpandableBinaryDictionary.java
@@ -305,7 +305,7 @@
                     @Override
                     public Boolean call() throws Exception {
                         return !distracterFilter.isDistracterToWordsInDictionaries(
-                                PrevWordsInfo.EMPTY_PREV_WORDS_INFO, word, mLocale);
+                                NgramContext.EMPTY_PREV_WORDS_INFO, word, mLocale);
                     }
                 },
                 new Runnable() {
@@ -354,7 +354,7 @@
     /**
      * Adds n-gram information of a word to the dictionary. May overwrite an existing entry.
      */
-    public void addNgramEntry(final PrevWordsInfo prevWordsInfo, final String word,
+    public void addNgramEntry(final NgramContext ngramContext, final String word,
             final int frequency, final int timestamp) {
         reloadDictionaryIfRequired();
         asyncExecuteTaskWithWriteLock(new Runnable() {
@@ -364,17 +364,17 @@
                     return;
                 }
                 runGCIfRequiredLocked(true /* mindsBlockByGC */);
-                addNgramEntryLocked(prevWordsInfo, word, frequency, timestamp);
+                addNgramEntryLocked(ngramContext, word, frequency, timestamp);
             }
         });
     }
 
-    protected void addNgramEntryLocked(final PrevWordsInfo prevWordsInfo, final String word,
+    protected void addNgramEntryLocked(final NgramContext ngramContext, final String word,
             final int frequency, final int timestamp) {
-        if (!mBinaryDictionary.addNgramEntry(prevWordsInfo, word, frequency, timestamp)) {
+        if (!mBinaryDictionary.addNgramEntry(ngramContext, word, frequency, timestamp)) {
             if (DEBUG) {
                 Log.i(TAG, "Cannot add n-gram entry.");
-                Log.i(TAG, "  PrevWordsInfo: " + prevWordsInfo + ", word: " + word);
+                Log.i(TAG, "  NgramContext: " + ngramContext + ", word: " + word);
             }
         }
     }
@@ -383,7 +383,7 @@
      * Dynamically remove the n-gram entry in the dictionary.
      */
     @UsedForTesting
-    public void removeNgramDynamically(final PrevWordsInfo prevWordsInfo, final String word) {
+    public void removeNgramDynamically(final NgramContext ngramContext, final String word) {
         reloadDictionaryIfRequired();
         asyncExecuteTaskWithWriteLock(new Runnable() {
             @Override
@@ -392,10 +392,10 @@
                     return;
                 }
                 runGCIfRequiredLocked(true /* mindsBlockByGC */);
-                if (!mBinaryDictionary.removeNgramEntry(prevWordsInfo, word)) {
+                if (!mBinaryDictionary.removeNgramEntry(ngramContext, word)) {
                     if (DEBUG) {
                         Log.i(TAG, "Cannot remove n-gram entry.");
-                        Log.i(TAG, "  PrevWordsInfo: " + prevWordsInfo + ", word: " + word);
+                        Log.i(TAG, "  NgramContext: " + ngramContext + ", word: " + word);
                     }
                 }
             }
@@ -434,7 +434,7 @@
 
     @Override
     public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion, final int sessionId,
             final float weightForLocale, final float[] inOutWeightOfLangModelVsSpatialModel) {
         reloadDictionaryIfRequired();
@@ -447,7 +447,7 @@
                     return null;
                 }
                 final ArrayList<SuggestedWordInfo> suggestions =
-                        mBinaryDictionary.getSuggestions(composer, prevWordsInfo, proximityInfo,
+                        mBinaryDictionary.getSuggestions(composer, ngramContext, proximityInfo,
                                 settingsValuesForSuggestion, sessionId, weightForLocale,
                                 inOutWeightOfLangModelVsSpatialModel);
                 if (mBinaryDictionary.isCorrupted()) {
@@ -519,9 +519,9 @@
     }
 
 
-    protected boolean isValidNgramLocked(final PrevWordsInfo prevWordsInfo, final String word) {
+    protected boolean isValidNgramLocked(final NgramContext ngramContext, final String word) {
         if (mBinaryDictionary == null) return false;
-        return mBinaryDictionary.isValidNgram(prevWordsInfo, word);
+        return mBinaryDictionary.isValidNgram(ngramContext, word);
     }
 
     /**
diff --git a/java/src/com/android/inputmethod/latin/LastComposedWord.java b/java/src/com/android/inputmethod/latin/LastComposedWord.java
index 8cbf837..f3f736f 100644
--- a/java/src/com/android/inputmethod/latin/LastComposedWord.java
+++ b/java/src/com/android/inputmethod/latin/LastComposedWord.java
@@ -48,7 +48,7 @@
     public final String mTypedWord;
     public final CharSequence mCommittedWord;
     public final String mSeparatorString;
-    public final PrevWordsInfo mPrevWordsInfo;
+    public final NgramContext mNgramContext;
     public final int mCapitalizedMode;
     public final InputPointers mInputPointers =
             new InputPointers(Constants.DICTIONARY_MAX_WORD_LENGTH);
@@ -64,7 +64,7 @@
     public LastComposedWord(final ArrayList<Event> events,
             final InputPointers inputPointers, final String typedWord,
             final CharSequence committedWord, final String separatorString,
-            final PrevWordsInfo prevWordsInfo, final int capitalizedMode) {
+            final NgramContext ngramContext, final int capitalizedMode) {
         if (inputPointers != null) {
             mInputPointers.copy(inputPointers);
         }
@@ -73,7 +73,7 @@
         mCommittedWord = committedWord;
         mSeparatorString = separatorString;
         mActive = true;
-        mPrevWordsInfo = prevWordsInfo;
+        mNgramContext = ngramContext;
         mCapitalizedMode = capitalizedMode;
     }
 
diff --git a/java/src/com/android/inputmethod/latin/PrevWordsInfo.java b/java/src/com/android/inputmethod/latin/NgramContext.java
similarity index 65%
rename from java/src/com/android/inputmethod/latin/PrevWordsInfo.java
rename to java/src/com/android/inputmethod/latin/NgramContext.java
index 1b7e8f9..c35c6e2 100644
--- a/java/src/com/android/inputmethod/latin/PrevWordsInfo.java
+++ b/java/src/com/android/inputmethod/latin/NgramContext.java
@@ -18,6 +18,7 @@
 
 import android.text.TextUtils;
 
+import com.android.inputmethod.annotations.UsedForTesting;
 import com.android.inputmethod.latin.utils.StringUtils;
 
 import java.util.Arrays;
@@ -26,11 +27,11 @@
  * Class to represent information of previous words. This class is used to add n-gram entries
  * into binary dictionaries, to get predictions, and to get suggestions.
  */
-public class PrevWordsInfo {
-    public static final PrevWordsInfo EMPTY_PREV_WORDS_INFO =
-            new PrevWordsInfo(WordInfo.EMPTY_WORD_INFO);
-    public static final PrevWordsInfo BEGINNING_OF_SENTENCE =
-            new PrevWordsInfo(WordInfo.BEGINNING_OF_SENTENCE);
+public class NgramContext {
+    public static final NgramContext EMPTY_PREV_WORDS_INFO =
+            new NgramContext(WordInfo.EMPTY_WORD_INFO);
+    public static final NgramContext BEGINNING_OF_SENTENCE =
+            new NgramContext(WordInfo.BEGINNING_OF_SENTENCE);
 
     /**
      * Word information used to represent previous words information.
@@ -86,35 +87,66 @@
     // For simplicity of implementation, elements may also be EMPTY_WORD_INFO transiently after the
     // WordComposer was reset and before starting a new composing word, but we should never be
     // calling getSuggetions* in this situation.
-    public final WordInfo[] mPrevWordsInfo;
+    private final WordInfo[] mPrevWordsInfo;
+    private final int mPrevWordsCount;
 
     // Construct from the previous word information.
-    public PrevWordsInfo(final WordInfo prevWordInfo) {
-        mPrevWordsInfo = new WordInfo[] { prevWordInfo };
+    public NgramContext(final WordInfo... prevWordsInfo) {
+        mPrevWordsInfo = prevWordsInfo;
+        mPrevWordsCount = prevWordsInfo.length;
     }
 
-    // Construct from WordInfo array. n-th element represents (n+1)-th previous word's information.
-    public PrevWordsInfo(final WordInfo[] prevWordsInfo) {
-        mPrevWordsInfo = prevWordsInfo;
+    // Construct from WordInfo array and size. The caller shouldn't change prevWordsInfo after
+    // calling this method.
+    private NgramContext(final NgramContext ngramContext, final int prevWordsCount) {
+        if (ngramContext.mPrevWordsCount < prevWordsCount) {
+            throw new IndexOutOfBoundsException("ngramContext.mPrevWordsCount ("
+                    + ngramContext.mPrevWordsCount + ") is smaller than prevWordsCount ("
+                    + prevWordsCount + ")");
+        }
+        mPrevWordsInfo = ngramContext.mPrevWordsInfo;
+        mPrevWordsCount = prevWordsCount;
     }
 
     // Create next prevWordsInfo using current prevWordsInfo.
-    public PrevWordsInfo getNextPrevWordsInfo(final WordInfo wordInfo) {
+    public NgramContext getNextNgramContext(final WordInfo wordInfo) {
         final int nextPrevWordCount = Math.min(Constants.MAX_PREV_WORD_COUNT_FOR_N_GRAM,
-                mPrevWordsInfo.length + 1);
+                mPrevWordsCount + 1);
         final WordInfo[] prevWordsInfo = new WordInfo[nextPrevWordCount];
         prevWordsInfo[0] = wordInfo;
-        System.arraycopy(mPrevWordsInfo, 0, prevWordsInfo, 1, prevWordsInfo.length - 1);
-        return new PrevWordsInfo(prevWordsInfo);
+        System.arraycopy(mPrevWordsInfo, 0, prevWordsInfo, 1, nextPrevWordCount - 1);
+        return new NgramContext(prevWordsInfo);
     }
 
     public boolean isValid() {
-        return mPrevWordsInfo.length > 0 && mPrevWordsInfo[0].isValid();
+        return mPrevWordsCount > 0 && mPrevWordsInfo[0].isValid();
+    }
+
+    public boolean isBeginningOfSentenceContext() {
+        return mPrevWordsCount > 0 && mPrevWordsInfo[0].mIsBeginningOfSentence;
+    }
+
+    // n is 1-indexed.
+    // TODO: Remove
+    public CharSequence getNthPrevWord(final int n) {
+        if (n <= 0 || n > mPrevWordsCount) {
+            return null;
+        }
+        return mPrevWordsInfo[n - 1].mWord;
+    }
+
+    // n is 1-indexed.
+    @UsedForTesting
+    public boolean isNthPrevWordBeginningOfSontence(final int n) {
+        if (n <= 0 || n > mPrevWordsCount) {
+            return false;
+        }
+        return mPrevWordsInfo[n - 1].mIsBeginningOfSentence;
     }
 
     public void outputToArray(final int[][] codePointArrays,
             final boolean[] isBeginningOfSentenceArray) {
-        for (int i = 0; i < mPrevWordsInfo.length; i++) {
+        for (int i = 0; i < mPrevWordsCount; i++) {
             final WordInfo wordInfo = mPrevWordsInfo[i];
             if (wordInfo == null || !wordInfo.isValid()) {
                 codePointArrays[i] = new int[0];
@@ -126,15 +158,13 @@
         }
     }
 
-    public PrevWordsInfo getTrimmedPrevWordsInfo(final int maxPrevWordCount) {
-        final int newSize = Math.min(maxPrevWordCount, mPrevWordsInfo.length);
-        // TODO: Quit creating a new array.
-        final WordInfo[] prevWordsInfo = Arrays.copyOf(mPrevWordsInfo, newSize);
-        return new PrevWordsInfo(prevWordsInfo);
+    public NgramContext getTrimmedNgramContext(final int maxPrevWordCount) {
+        final int newSize = Math.min(maxPrevWordCount, mPrevWordsCount);
+        return new NgramContext(this /* prevWordsInfo */, newSize);
     }
 
     public int getPrevWordCount() {
-        return mPrevWordsInfo.length;
+        return mPrevWordsCount;
     }
 
     @Override
@@ -146,19 +176,25 @@
     @Override
     public boolean equals(Object o) {
         if (this == o) return true;
-        if (!(o instanceof PrevWordsInfo)) return false;
-        final PrevWordsInfo prevWordsInfo = (PrevWordsInfo)o;
+        if (!(o instanceof NgramContext)) return false;
+        final NgramContext prevWordsInfo = (NgramContext)o;
 
-        final int minLength = Math.min(mPrevWordsInfo.length, prevWordsInfo.mPrevWordsInfo.length);
+        final int minLength = Math.min(mPrevWordsCount, prevWordsInfo.mPrevWordsCount);
         for (int i = 0; i < minLength; i++) {
             if (!mPrevWordsInfo[i].equals(prevWordsInfo.mPrevWordsInfo[i])) {
                 return false;
             }
         }
-        final WordInfo[] longerWordsInfo =
-                (mPrevWordsInfo.length > prevWordsInfo.mPrevWordsInfo.length) ?
-                        mPrevWordsInfo : prevWordsInfo.mPrevWordsInfo;
-        for (int i = minLength; i < longerWordsInfo.length; i++) {
+        final WordInfo[] longerWordsInfo;
+        final int longerWordsInfoCount;
+        if (mPrevWordsCount > prevWordsInfo.mPrevWordsCount) {
+            longerWordsInfo = mPrevWordsInfo;
+            longerWordsInfoCount = mPrevWordsCount;
+        } else {
+            longerWordsInfo = prevWordsInfo.mPrevWordsInfo;
+            longerWordsInfoCount = prevWordsInfo.mPrevWordsCount;
+        }
+        for (int i = minLength; i < longerWordsInfoCount; i++) {
             if (longerWordsInfo[i] != null
                     && !WordInfo.EMPTY_WORD_INFO.equals(longerWordsInfo[i])) {
                 return false;
@@ -170,7 +206,7 @@
     @Override
     public String toString() {
         final StringBuffer builder = new StringBuffer();
-        for (int i = 0; i < mPrevWordsInfo.length; i++) {
+        for (int i = 0; i < mPrevWordsCount; i++) {
             final WordInfo wordInfo = mPrevWordsInfo[i];
             builder.append("PrevWord[");
             builder.append(i);
diff --git a/java/src/com/android/inputmethod/latin/ReadOnlyBinaryDictionary.java b/java/src/com/android/inputmethod/latin/ReadOnlyBinaryDictionary.java
index 827367b..bc8bd83 100644
--- a/java/src/com/android/inputmethod/latin/ReadOnlyBinaryDictionary.java
+++ b/java/src/com/android/inputmethod/latin/ReadOnlyBinaryDictionary.java
@@ -51,13 +51,13 @@
 
     @Override
     public ArrayList<SuggestedWordInfo> getSuggestions(final WordComposer composer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion,
             final int sessionId, final float weightForLocale,
             final float[] inOutWeightOfLangModelVsSpatialModel) {
         if (mLock.readLock().tryLock()) {
             try {
-                return mBinaryDictionary.getSuggestions(composer, prevWordsInfo, proximityInfo,
+                return mBinaryDictionary.getSuggestions(composer, ngramContext, proximityInfo,
                         settingsValuesForSuggestion, sessionId, weightForLocale,
                         inOutWeightOfLangModelVsSpatialModel);
             } finally {
diff --git a/java/src/com/android/inputmethod/latin/RichInputConnection.java b/java/src/com/android/inputmethod/latin/RichInputConnection.java
index d672430..7507061 100644
--- a/java/src/com/android/inputmethod/latin/RichInputConnection.java
+++ b/java/src/com/android/inputmethod/latin/RichInputConnection.java
@@ -36,7 +36,7 @@
 import com.android.inputmethod.latin.settings.SpacingAndPunctuations;
 import com.android.inputmethod.latin.utils.CapsModeUtils;
 import com.android.inputmethod.latin.utils.DebugLogUtils;
-import com.android.inputmethod.latin.utils.PrevWordsInfoUtils;
+import com.android.inputmethod.latin.utils.NgramContextUtils;
 import com.android.inputmethod.latin.utils.ScriptUtils;
 import com.android.inputmethod.latin.utils.SpannableStringUtils;
 import com.android.inputmethod.latin.utils.StringUtils;
@@ -593,11 +593,11 @@
     }
 
     @SuppressWarnings("unused")
-    public PrevWordsInfo getPrevWordsInfoFromNthPreviousWord(
+    public NgramContext getNgramContextFromNthPreviousWord(
             final SpacingAndPunctuations spacingAndPunctuations, final int n) {
         mIC = mParent.getCurrentInputConnection();
         if (null == mIC) {
-            return PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+            return NgramContext.EMPTY_PREV_WORDS_INFO;
         }
         final CharSequence prev = getTextBeforeCursor(LOOKBACK_CHARACTER_NUM, 0);
         if (DEBUG_PREVIOUS_TEXT && null != prev) {
@@ -618,7 +618,7 @@
                 }
             }
         }
-        return PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
+        return NgramContextUtils.getNgramContextFromNthPreviousWord(
                 prev, spacingAndPunctuations, n);
     }
 
diff --git a/java/src/com/android/inputmethod/latin/Suggest.java b/java/src/com/android/inputmethod/latin/Suggest.java
index 1ecc995..d2d9b9b 100644
--- a/java/src/com/android/inputmethod/latin/Suggest.java
+++ b/java/src/com/android/inputmethod/latin/Suggest.java
@@ -68,15 +68,15 @@
     }
 
     public void getSuggestedWords(final WordComposer wordComposer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion,
             final boolean isCorrectionEnabled, final int inputStyle, final int sequenceNumber,
             final OnGetSuggestedWordsCallback callback) {
         if (wordComposer.isBatchMode()) {
-            getSuggestedWordsForBatchInput(wordComposer, prevWordsInfo, proximityInfo,
+            getSuggestedWordsForBatchInput(wordComposer, ngramContext, proximityInfo,
                     settingsValuesForSuggestion, inputStyle, sequenceNumber, callback);
         } else {
-            getSuggestedWordsForNonBatchInput(wordComposer, prevWordsInfo, proximityInfo,
+            getSuggestedWordsForNonBatchInput(wordComposer, ngramContext, proximityInfo,
                     settingsValuesForSuggestion, inputStyle, isCorrectionEnabled,
                     sequenceNumber, callback);
         }
@@ -121,7 +121,7 @@
     // Retrieves suggestions for non-batch input (typing, recorrection, predictions...)
     // and calls the callback function with the suggestions.
     private void getSuggestedWordsForNonBatchInput(final WordComposer wordComposer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion,
             final int inputStyleIfNotPrediction, final boolean isCorrectionEnabled,
             final int sequenceNumber, final OnGetSuggestedWordsCallback callback) {
@@ -132,7 +132,7 @@
                 : typedWord;
 
         final SuggestionResults suggestionResults = mDictionaryFacilitator.getSuggestionResults(
-                wordComposer, prevWordsInfo, proximityInfo, settingsValuesForSuggestion,
+                wordComposer, ngramContext, proximityInfo, settingsValuesForSuggestion,
                 SESSION_ID_TYPING);
         final ArrayList<SuggestedWordInfo> suggestionsContainer =
                 getTransformedSuggestedWordInfoList(wordComposer, suggestionResults,
@@ -209,12 +209,12 @@
     // Retrieves suggestions for the batch input
     // and calls the callback function with the suggestions.
     private void getSuggestedWordsForBatchInput(final WordComposer wordComposer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo,
+            final NgramContext ngramContext, final ProximityInfo proximityInfo,
             final SettingsValuesForSuggestion settingsValuesForSuggestion,
             final int inputStyle, final int sequenceNumber,
             final OnGetSuggestedWordsCallback callback) {
         final SuggestionResults suggestionResults = mDictionaryFacilitator.getSuggestionResults(
-                wordComposer, prevWordsInfo, proximityInfo, settingsValuesForSuggestion,
+                wordComposer, ngramContext, proximityInfo, settingsValuesForSuggestion,
                 SESSION_ID_GESTURE);
         final Locale defaultLocale = mDictionaryFacilitator.getLocale();
         final ArrayList<SuggestedWordInfo> suggestionsContainer =
diff --git a/java/src/com/android/inputmethod/latin/WordComposer.java b/java/src/com/android/inputmethod/latin/WordComposer.java
index 567aa07..f85b34b 100644
--- a/java/src/com/android/inputmethod/latin/WordComposer.java
+++ b/java/src/com/android/inputmethod/latin/WordComposer.java
@@ -448,13 +448,13 @@
     // `type' should be one of the LastComposedWord.COMMIT_TYPE_* constants above.
     // committedWord should contain suggestion spans if applicable.
     public LastComposedWord commitWord(final int type, final CharSequence committedWord,
-            final String separatorString, final PrevWordsInfo prevWordsInfo) {
+            final String separatorString, final NgramContext ngramContext) {
         // Note: currently, we come here whenever we commit a word. If it's a MANUAL_PICK
         // or a DECIDED_WORD we may cancel the commit later; otherwise, we should deactivate
         // the last composed word to ensure this does not happen.
         final LastComposedWord lastComposedWord = new LastComposedWord(mEvents,
                 mInputPointers, mTypedWordCache.toString(), committedWord, separatorString,
-                prevWordsInfo, mCapitalizedMode);
+                ngramContext, mCapitalizedMode);
         mInputPointers.reset();
         if (type != LastComposedWord.COMMIT_TYPE_DECIDED_WORD
                 && type != LastComposedWord.COMMIT_TYPE_MANUAL_PICK) {
diff --git a/java/src/com/android/inputmethod/latin/inputlogic/InputLogic.java b/java/src/com/android/inputmethod/latin/inputlogic/InputLogic.java
index 46427e5..d7e1eba 100644
--- a/java/src/com/android/inputmethod/latin/inputlogic/InputLogic.java
+++ b/java/src/com/android/inputmethod/latin/inputlogic/InputLogic.java
@@ -45,7 +45,7 @@
 import com.android.inputmethod.latin.InputPointers;
 import com.android.inputmethod.latin.LastComposedWord;
 import com.android.inputmethod.latin.LatinIME;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.RichInputConnection;
 import com.android.inputmethod.latin.Suggest;
 import com.android.inputmethod.latin.Suggest.OnGetSuggestedWordsCallback;
@@ -1376,7 +1376,7 @@
     }
 
     private void performAdditionToUserHistoryDictionary(final SettingsValues settingsValues,
-            final String suggestion, final PrevWordsInfo prevWordsInfo) {
+            final String suggestion, final NgramContext ngramContext) {
         // If correction is not enabled, we don't add words to the user history dictionary.
         // That's to avoid unintended additions in some sensitive fields, or fields that
         // expect to receive non-words.
@@ -1388,7 +1388,7 @@
         final int timeStampInSeconds = (int)TimeUnit.MILLISECONDS.toSeconds(
                 System.currentTimeMillis());
         mDictionaryFacilitator.addToUserHistory(suggestion, wasAutoCapitalized,
-                prevWordsInfo, timeStampInSeconds, settingsValues.mBlockPotentiallyOffensive);
+                ngramContext, timeStampInSeconds, settingsValues.mBlockPotentiallyOffensive);
     }
 
     public void performUpdateSuggestionStripSync(final SettingsValues settingsValues,
@@ -1519,10 +1519,10 @@
             }
         }
         final int[] codePoints = StringUtils.toCodePointArray(typedWord);
-        // We want the previous word for suggestion. If we have chars in the word
+        // We want the context of preceding words for suggestion. If we have chars in the word
         // before the cursor, then we want the word before that, hence 2; otherwise,
         // we want the word immediately before the cursor, hence 1.
-        final PrevWordsInfo prevWordsInfo = getPrevWordsInfoFromNthPreviousWordForSuggestion(
+        final NgramContext ngramContext = getNgramContextFromNthPreviousWordForSuggestion(
                 settingsValues.mSpacingAndPunctuations,
                 0 == numberOfCharsInWordBeforeCursor ? 1 : 2);
         mWordComposer.setComposingWord(codePoints,
@@ -1591,6 +1591,10 @@
         final String committedWordString = committedWord.toString();
         final int cancelLength = committedWord.length();
         final String separatorString = mLastComposedWord.mSeparatorString;
+        // If our separator is a space, we won't actually commit it,
+        // but set the space state to PHANTOM so that a space will be inserted
+        // on the next keypress
+        final boolean usePhantomSpace = separatorString.equals(Constants.STRING_SPACE);
         // We want java chars, not codepoints for the following.
         final int separatorLength = separatorString.length();
         // TODO: should we check our saved separator against the actual contents of the text view?
@@ -1611,7 +1615,8 @@
         if (!TextUtils.isEmpty(committedWord)) {
             mDictionaryFacilitator.removeWordFromPersonalizedDicts(committedWordString);
         }
-        final String stringToCommit = originallyTypedWord + separatorString;
+        final String stringToCommit = originallyTypedWord +
+                (usePhantomSpace ? "" : separatorString);
         final SpannableString textToCommit = new SpannableString(stringToCommit);
         if (committedWord instanceof SpannableString) {
             final SpannableString committedWordWithSuggestionSpans = (SpannableString)committedWord;
@@ -1663,6 +1668,9 @@
             } else {
                 mConnection.commitText(textToCommit, 1);
             }
+            if (usePhantomSpace) {
+                mSpaceState = SpaceState.PHANTOM;
+            }
         } else {
             // For languages without spaces, we revert the typed string but the cursor is flush
             // with the typed word, so we need to resume suggestions right away.
@@ -1760,24 +1768,24 @@
     }
 
     /**
-     * Get information fo previous words from the nth previous word before the cursor as context
+     * Get n-gram context from the nth previous word before the cursor as context
      * for the suggestion process.
      * @param spacingAndPunctuations the current spacing and punctuations settings.
      * @param nthPreviousWord reverse index of the word to get (1-indexed)
      * @return the information of previous words
      */
     // TODO: Make this private
-    public PrevWordsInfo getPrevWordsInfoFromNthPreviousWordForSuggestion(
+    public NgramContext getNgramContextFromNthPreviousWordForSuggestion(
             final SpacingAndPunctuations spacingAndPunctuations, final int nthPreviousWord) {
         if (spacingAndPunctuations.mCurrentLanguageHasSpaces) {
             // If we are typing in a language with spaces we can just look up the previous
             // word information from textview.
-            return mConnection.getPrevWordsInfoFromNthPreviousWord(
+            return mConnection.getNgramContextFromNthPreviousWord(
                     spacingAndPunctuations, nthPreviousWord);
         } else {
             return LastComposedWord.NOT_A_COMPOSED_WORD == mLastComposedWord ?
-                    PrevWordsInfo.BEGINNING_OF_SENTENCE :
-                            new PrevWordsInfo(new PrevWordsInfo.WordInfo(
+                    NgramContext.BEGINNING_OF_SENTENCE :
+                            new NgramContext(new NgramContext.WordInfo(
                                     mLastComposedWord.mCommittedWord.toString()));
         }
     }
@@ -2140,20 +2148,20 @@
         final CharSequence chosenWordWithSuggestions =
                 SuggestionSpanUtils.getTextWithSuggestionSpan(mLatinIME, chosenWord,
                         suggestedWords);
-        // When we are composing word, get previous words information from the 2nd previous word
-        // because the 1st previous word is the word to be committed. Otherwise get previous words
-        // information from the 1st previous word.
-        final PrevWordsInfo prevWordsInfo = mConnection.getPrevWordsInfoFromNthPreviousWord(
+        // When we are composing word, get n-gram context from the 2nd previous word because the
+        // 1st previous word is the word to be committed. Otherwise get n-gram context from the 1st
+        // previous word.
+        final NgramContext ngramContext = mConnection.getNgramContextFromNthPreviousWord(
                 settingsValues.mSpacingAndPunctuations, mWordComposer.isComposingWord() ? 2 : 1);
         mConnection.commitText(chosenWordWithSuggestions, 1);
         // Add the word to the user history dictionary
-        performAdditionToUserHistoryDictionary(settingsValues, chosenWord, prevWordsInfo);
+        performAdditionToUserHistoryDictionary(settingsValues, chosenWord, ngramContext);
         // TODO: figure out here if this is an auto-correct or if the best word is actually
         // what user typed. Note: currently this is done much later in
         // LastComposedWord#didCommitTypedWord by string equality of the remembered
         // strings.
         mLastComposedWord = mWordComposer.commitWord(commitType,
-                chosenWordWithSuggestions, separatorString, prevWordsInfo);
+                chosenWordWithSuggestions, separatorString, ngramContext);
     }
 
     /**
@@ -2200,7 +2208,7 @@
         mWordComposer.adviseCapitalizedModeBeforeFetchingSuggestions(
                 getActualCapsMode(settingsValues, keyboardShiftMode));
         mSuggest.getSuggestedWords(mWordComposer,
-                getPrevWordsInfoFromNthPreviousWordForSuggestion(
+                getNgramContextFromNthPreviousWordForSuggestion(
                         settingsValues.mSpacingAndPunctuations,
                         // Get the word on which we should search the bigrams. If we are composing
                         // a word, it's whatever is *before* the half-committed word in the buffer,
diff --git a/java/src/com/android/inputmethod/latin/makedict/DictionaryHeader.java b/java/src/com/android/inputmethod/latin/makedict/DictionaryHeader.java
index df447fd..3f9ffd2 100644
--- a/java/src/com/android/inputmethod/latin/makedict/DictionaryHeader.java
+++ b/java/src/com/android/inputmethod/latin/makedict/DictionaryHeader.java
@@ -47,6 +47,7 @@
     public static final String MAX_UNIGRAM_COUNT_KEY = "MAX_UNIGRAM_COUNT";
     public static final String MAX_BIGRAM_COUNT_KEY = "MAX_BIGRAM_COUNT";
     public static final String ATTRIBUTE_VALUE_TRUE = "1";
+    public static final String CODE_POINT_TABLE_KEY = "codePointTable";
 
     public DictionaryHeader(final int headerSize, final DictionaryOptions dictionaryOptions,
             final FormatOptions formatOptions) throws UnsupportedFormatException {
diff --git a/java/src/com/android/inputmethod/latin/makedict/FormatSpec.java b/java/src/com/android/inputmethod/latin/makedict/FormatSpec.java
index ec3c6e2..2661d5d 100644
--- a/java/src/com/android/inputmethod/latin/makedict/FormatSpec.java
+++ b/java/src/com/android/inputmethod/latin/makedict/FormatSpec.java
@@ -237,6 +237,8 @@
     static final int UINT16_MAX = 0xFFFF;
     static final int UINT24_MAX = 0xFFFFFF;
     static final int MSB8 = 0x80;
+    static final int MINIMAL_ONE_BYTE_CHARACTER_VALUE = 0x20;
+    static final int MAXIMAL_ONE_BYTE_CHARACTER_VALUE = 0xFF;
 
     /**
      * Options about file format.
diff --git a/java/src/com/android/inputmethod/latin/personalization/UserHistoryDictionary.java b/java/src/com/android/inputmethod/latin/personalization/UserHistoryDictionary.java
index d1486f6..d616846 100644
--- a/java/src/com/android/inputmethod/latin/personalization/UserHistoryDictionary.java
+++ b/java/src/com/android/inputmethod/latin/personalization/UserHistoryDictionary.java
@@ -23,7 +23,7 @@
 import com.android.inputmethod.latin.Constants;
 import com.android.inputmethod.latin.Dictionary;
 import com.android.inputmethod.latin.ExpandableBinaryDictionary;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.utils.DistracterFilter;
 
 import java.io.File;
@@ -53,14 +53,14 @@
      * Add a word to the user history dictionary.
      *
      * @param userHistoryDictionary the user history dictionary
-     * @param prevWordsInfo the information of previous words
+     * @param ngramContext the n-gram context
      * @param word the word the user inputted
      * @param isValid whether the word is valid or not
      * @param timestamp the timestamp when the word has been inputted
      * @param distracterFilter the filter to check whether the word is a distracter
      */
     public static void addToDictionary(final ExpandableBinaryDictionary userHistoryDictionary,
-            final PrevWordsInfo prevWordsInfo, final String word, final boolean isValid,
+            final NgramContext ngramContext, final String word, final boolean isValid,
             final int timestamp, final DistracterFilter distracterFilter) {
         if (word.length() > Constants.DICTIONARY_MAX_WORD_LENGTH) {
             return;
@@ -71,12 +71,11 @@
                 null /* shortcutTarget */, 0 /* shortcutFreq */, false /* isNotAWord */,
                 false /* isBlacklisted */, timestamp, distracterFilter);
 
-        final boolean isBeginningOfSentenceContext =
-                prevWordsInfo.mPrevWordsInfo[0].mIsBeginningOfSentence;
-        final PrevWordsInfo prevWordsInfoToBeSaved =
-                prevWordsInfo.getTrimmedPrevWordsInfo(SUPPORTED_NGRAM - 1);
-        for (int i = 0; i < prevWordsInfoToBeSaved.getPrevWordCount(); i++) {
-            final CharSequence prevWord = prevWordsInfoToBeSaved.mPrevWordsInfo[i].mWord;
+        final boolean isBeginningOfSentenceContext = ngramContext.isBeginningOfSentenceContext();
+        final NgramContext ngramContextToBeSaved =
+                ngramContext.getTrimmedNgramContext(SUPPORTED_NGRAM - 1);
+        for (int i = 0; i < ngramContextToBeSaved.getPrevWordCount(); i++) {
+            final CharSequence prevWord = ngramContextToBeSaved.getNthPrevWord(1 /* n */);
             if (prevWord == null || (prevWord.length() > Constants.DICTIONARY_MAX_WORD_LENGTH)) {
                 return;
             }
@@ -87,11 +86,11 @@
             if (isBeginningOfSentenceContext) {
                 // Beginning-of-Sentence n-gram entry is added as an n-gram entry of an OOV word.
                 userHistoryDictionary.addNgramEntry(
-                        prevWordsInfoToBeSaved.getTrimmedPrevWordsInfo(i + 1), word,
+                        ngramContextToBeSaved.getTrimmedNgramContext(i + 1), word,
                         FREQUENCY_FOR_WORDS_NOT_IN_DICTS, timestamp);
             } else {
                 userHistoryDictionary.addNgramEntry(
-                        prevWordsInfoToBeSaved.getTrimmedPrevWordsInfo(i + 1), word, frequency,
+                        ngramContextToBeSaved.getTrimmedNgramContext(i + 1), word, frequency,
                         timestamp);
             }
         }
diff --git a/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerService.java b/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerService.java
index 3523916..2a4e14c 100644
--- a/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerService.java
+++ b/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerService.java
@@ -31,7 +31,7 @@
 import com.android.inputmethod.keyboard.ProximityInfo;
 import com.android.inputmethod.latin.DictionaryFacilitator;
 import com.android.inputmethod.latin.DictionaryFacilitatorLruCache;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.R;
 import com.android.inputmethod.latin.RichInputMethodSubtype;
 import com.android.inputmethod.latin.settings.SettingsValuesForSuggestion;
@@ -163,14 +163,14 @@
     }
 
     public SuggestionResults getSuggestionResults(final Locale locale, final WordComposer composer,
-            final PrevWordsInfo prevWordsInfo, final ProximityInfo proximityInfo) {
+            final NgramContext ngramContext, final ProximityInfo proximityInfo) {
         Integer sessionId = null;
         mSemaphore.acquireUninterruptibly();
         try {
             sessionId = mSessionIdPool.poll();
             DictionaryFacilitator dictionaryFacilitatorForLocale =
                     mDictionaryFacilitatorCache.get(locale);
-            return dictionaryFacilitatorForLocale.getSuggestionResults(composer, prevWordsInfo,
+            return dictionaryFacilitatorForLocale.getSuggestionResults(composer, ngramContext,
                     proximityInfo, mSettingsValuesForSuggestion, sessionId);
         } finally {
             if (sessionId != null) {
diff --git a/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerSession.java b/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerSession.java
index 34e0119..8393b30 100644
--- a/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerSession.java
+++ b/java/src/com/android/inputmethod/latin/spellcheck/AndroidSpellCheckerSession.java
@@ -25,7 +25,7 @@
 import android.view.textservice.TextInfo;
 
 import com.android.inputmethod.compat.TextInfoCompatUtils;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.utils.StringUtils;
 
 import java.util.ArrayList;
@@ -62,8 +62,8 @@
             final int offset = ssi.getOffsetAt(i);
             final int length = ssi.getLengthAt(i);
             final CharSequence subText = typedText.subSequence(offset, offset + length);
-            final PrevWordsInfo prevWordsInfo =
-                    new PrevWordsInfo(new PrevWordsInfo.WordInfo(currentWord));
+            final NgramContext ngramContext =
+                    new NgramContext(new NgramContext.WordInfo(currentWord));
             currentWord = subText;
             if (!subText.toString().contains(AndroidSpellCheckerService.SINGLE_QUOTE)) {
                 continue;
@@ -80,7 +80,7 @@
                 if (TextUtils.isEmpty(splitText)) {
                     continue;
                 }
-                if (mSuggestionsCache.getSuggestionsFromCache(splitText.toString(), prevWordsInfo)
+                if (mSuggestionsCache.getSuggestionsFromCache(splitText.toString(), ngramContext)
                         == null) {
                     continue;
                 }
@@ -208,10 +208,10 @@
                 } else {
                     prevWord = null;
                 }
-                final PrevWordsInfo prevWordsInfo =
-                        new PrevWordsInfo(new PrevWordsInfo.WordInfo(prevWord));
+                final NgramContext ngramContext =
+                        new NgramContext(new NgramContext.WordInfo(prevWord));
                 final TextInfo textInfo = textInfos[i];
-                retval[i] = onGetSuggestionsInternal(textInfo, prevWordsInfo, suggestionsLimit);
+                retval[i] = onGetSuggestionsInternal(textInfo, ngramContext, suggestionsLimit);
                 retval[i].setCookieAndSequence(textInfo.getCookie(), textInfo.getSequence());
             }
             return retval;
diff --git a/java/src/com/android/inputmethod/latin/spellcheck/AndroidWordLevelSpellCheckerSession.java b/java/src/com/android/inputmethod/latin/spellcheck/AndroidWordLevelSpellCheckerSession.java
index d668672..7b6aacd 100644
--- a/java/src/com/android/inputmethod/latin/spellcheck/AndroidWordLevelSpellCheckerSession.java
+++ b/java/src/com/android/inputmethod/latin/spellcheck/AndroidWordLevelSpellCheckerSession.java
@@ -31,7 +31,7 @@
 import com.android.inputmethod.keyboard.Keyboard;
 import com.android.inputmethod.keyboard.ProximityInfo;
 import com.android.inputmethod.latin.Constants;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo;
 import com.android.inputmethod.latin.WordComposer;
 import com.android.inputmethod.latin.utils.BinaryDictionaryUtils;
@@ -73,27 +73,25 @@
         private final LruCache<String, SuggestionsParams> mUnigramSuggestionsInfoCache =
                 new LruCache<>(MAX_CACHE_SIZE);
 
-        // TODO: Support n-gram input
-        private static String generateKey(final String query, final PrevWordsInfo prevWordsInfo) {
-            if (TextUtils.isEmpty(query) || !prevWordsInfo.isValid()) {
+        private static String generateKey(final String query, final NgramContext ngramContext) {
+            if (TextUtils.isEmpty(query) || !ngramContext.isValid()) {
                 return query;
             }
-            return query + CHAR_DELIMITER + prevWordsInfo;
+            return query + CHAR_DELIMITER + ngramContext;
         }
 
         public SuggestionsParams getSuggestionsFromCache(String query,
-                final PrevWordsInfo prevWordsInfo) {
-            return mUnigramSuggestionsInfoCache.get(generateKey(query, prevWordsInfo));
+                final NgramContext ngramContext) {
+            return mUnigramSuggestionsInfoCache.get(generateKey(query, ngramContext));
         }
 
-        public void putSuggestionsToCache(
-                final String query, final PrevWordsInfo prevWordsInfo,
+        public void putSuggestionsToCache(final String query, final NgramContext ngramContext,
                 final String[] suggestions, final int flags) {
             if (suggestions == null || TextUtils.isEmpty(query)) {
                 return;
             }
             mUnigramSuggestionsInfoCache.put(
-                    generateKey(query, prevWordsInfo), new SuggestionsParams(suggestions, flags));
+                    generateKey(query, ngramContext), new SuggestionsParams(suggestions, flags));
         }
 
         public void clearCache() {
@@ -223,12 +221,11 @@
     }
 
     protected SuggestionsInfo onGetSuggestionsInternal(
-            final TextInfo textInfo, final PrevWordsInfo prevWordsInfo,
-            final int suggestionsLimit) {
+            final TextInfo textInfo, final NgramContext ngramContext, final int suggestionsLimit) {
         try {
             final String inText = textInfo.getText();
             final SuggestionsParams cachedSuggestionsParams =
-                    mSuggestionsCache.getSuggestionsFromCache(inText, prevWordsInfo);
+                    mSuggestionsCache.getSuggestionsFromCache(inText, ngramContext);
             if (cachedSuggestionsParams != null) {
                 if (DBG) {
                     Log.d(TAG, "Cache hit: " + inText + ", " + cachedSuggestionsParams.mFlags);
@@ -283,7 +280,7 @@
             composer.setComposingWord(codePoints, coordinates);
             // TODO: Don't gather suggestions if the limit is <= 0 unless necessary
             final SuggestionResults suggestionResults = mService.getSuggestionResults(
-                    mLocale, composer, prevWordsInfo, proximityInfo);
+                    mLocale, composer, ngramContext, proximityInfo);
             final Result result = getResult(capitalizeType, mLocale, suggestionsLimit,
                     mService.getRecommendedThreshold(), text, suggestionResults);
             isInDict = isInDictForAnyCapitalization(text, capitalizeType);
@@ -308,7 +305,7 @@
                                     .getValueOf_RESULT_ATTR_HAS_RECOMMENDED_SUGGESTIONS()
                             : 0);
             final SuggestionsInfo retval = new SuggestionsInfo(flags, result.mSuggestions);
-            mSuggestionsCache.putSuggestionsToCache(text, prevWordsInfo, result.mSuggestions,
+            mSuggestionsCache.putSuggestionsToCache(text, ngramContext, result.mSuggestions,
                     flags);
             return retval;
         } catch (RuntimeException e) {
diff --git a/java/src/com/android/inputmethod/latin/utils/DistracterFilter.java b/java/src/com/android/inputmethod/latin/utils/DistracterFilter.java
index 6fd241e..355d00d 100644
--- a/java/src/com/android/inputmethod/latin/utils/DistracterFilter.java
+++ b/java/src/com/android/inputmethod/latin/utils/DistracterFilter.java
@@ -22,23 +22,23 @@
 import android.view.inputmethod.InputMethodSubtype;
 
 import com.android.inputmethod.annotations.UsedForTesting;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 
 public interface DistracterFilter {
     /**
      * Determine whether a word is a distracter to words in dictionaries.
      *
-     * @param prevWordsInfo the information of previous words.
+     * @param ngramContext the n-gram context
      * @param testedWord the word that will be tested to see whether it is a distracter to words
      *                   in dictionaries.
      * @param locale the locale of word.
      * @return true if testedWord is a distracter, otherwise false.
      */
-    public boolean isDistracterToWordsInDictionaries(final PrevWordsInfo prevWordsInfo,
+    public boolean isDistracterToWordsInDictionaries(final NgramContext ngramContext,
             final String testedWord, final Locale locale);
 
     @UsedForTesting
-    public int getWordHandlingType(final PrevWordsInfo prevWordsInfo, final String testedWord,
+    public int getWordHandlingType(final NgramContext ngramContext, final String testedWord,
             final Locale locale);
 
     public void updateEnabledSubtypes(final List<InputMethodSubtype> enabledSubtypes);
@@ -72,13 +72,13 @@
 
     public static final DistracterFilter EMPTY_DISTRACTER_FILTER = new DistracterFilter() {
         @Override
-        public boolean isDistracterToWordsInDictionaries(PrevWordsInfo prevWordsInfo,
+        public boolean isDistracterToWordsInDictionaries(NgramContext ngramContext,
                 String testedWord, Locale locale) {
             return false;
         }
 
         @Override
-        public int getWordHandlingType(final PrevWordsInfo prevWordsInfo,
+        public int getWordHandlingType(final NgramContext ngramContext,
                 final String testedWord, final Locale locale) {
             return HandlingType.REQUIRE_NO_SPECIAL_HANDLINGS;
         }
diff --git a/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingExactMatchesAndSuggestions.java b/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingExactMatchesAndSuggestions.java
index f8a8453..8f0f9bb 100644
--- a/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingExactMatchesAndSuggestions.java
+++ b/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingExactMatchesAndSuggestions.java
@@ -36,7 +36,7 @@
 import com.android.inputmethod.keyboard.KeyboardLayoutSet;
 import com.android.inputmethod.latin.DictionaryFacilitator;
 import com.android.inputmethod.latin.DictionaryFacilitatorLruCache;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.RichInputMethodSubtype;
 import com.android.inputmethod.latin.SuggestedWords.SuggestedWordInfo;
 import com.android.inputmethod.latin.WordComposer;
@@ -156,14 +156,14 @@
     /**
      * Determine whether a word is a distracter to words in dictionaries.
      *
-     * @param prevWordsInfo the information of previous words. Not used for now.
+     * @param ngramContext the n-gram context. Not used for now.
      * @param testedWord the word that will be tested to see whether it is a distracter to words
      *                   in dictionaries.
      * @param locale the locale of word.
      * @return true if testedWord is a distracter, otherwise false.
      */
     @Override
-    public boolean isDistracterToWordsInDictionaries(final PrevWordsInfo prevWordsInfo,
+    public boolean isDistracterToWordsInDictionaries(final NgramContext ngramContext,
             final String testedWord, final Locale locale) {
         if (locale == null) {
             return false;
@@ -250,7 +250,7 @@
         final SuggestionResults suggestionResults;
         synchronized (mLock) {
             suggestionResults = dictionaryFacilitator.getSuggestionResults(
-                    composer, PrevWordsInfo.EMPTY_PREV_WORDS_INFO, keyboard.getProximityInfo(),
+                    composer, NgramContext.EMPTY_PREV_WORDS_INFO, keyboard.getProximityInfo(),
                     settingsValuesForSuggestion, 0 /* sessionId */);
         }
         if (suggestionResults.isEmpty()) {
@@ -283,7 +283,7 @@
         return false;
     }
 
-    private boolean shouldBeLowerCased(final PrevWordsInfo prevWordsInfo, final String testedWord,
+    private boolean shouldBeLowerCased(final NgramContext ngramContext, final String testedWord,
             final Locale locale) {
         final DictionaryFacilitator dictionaryFacilitator =
                 mDictionaryFacilitatorLruCache.get(locale);
@@ -298,7 +298,7 @@
             return true;
         }
         if (StringUtils.getCapitalizationType(testedWord) == StringUtils.CAPITALIZE_FIRST
-                && !prevWordsInfo.isValid()) {
+                && !ngramContext.isValid()) {
             // TODO: Check beginning-of-sentence.
             return true;
         }
@@ -306,13 +306,13 @@
     }
 
     @Override
-    public int getWordHandlingType(final PrevWordsInfo prevWordsInfo, final String testedWord,
+    public int getWordHandlingType(final NgramContext ngramContext, final String testedWord,
             final Locale locale) {
         // TODO: Use this method for user history dictionary.
         if (testedWord == null|| locale == null) {
             return HandlingType.getHandlingType(false /* shouldBeLowerCased */, false /* isOov */);
         }
-        final boolean shouldBeLowerCased = shouldBeLowerCased(prevWordsInfo, testedWord, locale);
+        final boolean shouldBeLowerCased = shouldBeLowerCased(ngramContext, testedWord, locale);
         final String caseModifiedWord =
                 shouldBeLowerCased ? testedWord.toLowerCase(locale) : testedWord;
         final boolean isOov = !mDictionaryFacilitatorLruCache.get(locale).isValidWord(
diff --git a/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingIsInDictionary.java b/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingIsInDictionary.java
index 349236f..df6e970 100644
--- a/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingIsInDictionary.java
+++ b/java/src/com/android/inputmethod/latin/utils/DistracterFilterCheckingIsInDictionary.java
@@ -22,7 +22,7 @@
 import android.view.inputmethod.InputMethodSubtype;
 
 import com.android.inputmethod.latin.Dictionary;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 
 public class DistracterFilterCheckingIsInDictionary implements DistracterFilter {
     private final DistracterFilter mDistracterFilter;
@@ -35,7 +35,7 @@
     }
 
     @Override
-    public boolean isDistracterToWordsInDictionaries(PrevWordsInfo prevWordsInfo,
+    public boolean isDistracterToWordsInDictionaries(NgramContext ngramContext,
             String testedWord, Locale locale) {
         if (mDictionary.isInDictionary(testedWord)) {
             // This filter treats entries that are already in the dictionary as non-distracters
@@ -43,14 +43,14 @@
             return false;
         } else {
             return mDistracterFilter.isDistracterToWordsInDictionaries(
-                    prevWordsInfo, testedWord, locale);
+                    ngramContext, testedWord, locale);
         }
     }
 
     @Override
-    public int getWordHandlingType(final PrevWordsInfo prevWordsInfo, final String testedWord,
+    public int getWordHandlingType(final NgramContext ngramContext, final String testedWord,
             final Locale locale) {
-        return mDistracterFilter.getWordHandlingType(prevWordsInfo, testedWord, locale);
+        return mDistracterFilter.getWordHandlingType(ngramContext, testedWord, locale);
     }
 
     @Override
diff --git a/java/src/com/android/inputmethod/latin/utils/LanguageModelParam.java b/java/src/com/android/inputmethod/latin/utils/LanguageModelParam.java
index 7955541..73aefb8 100644
--- a/java/src/com/android/inputmethod/latin/utils/LanguageModelParam.java
+++ b/java/src/com/android/inputmethod/latin/utils/LanguageModelParam.java
@@ -21,7 +21,7 @@
 import com.android.inputmethod.annotations.UsedForTesting;
 import com.android.inputmethod.latin.Dictionary;
 import com.android.inputmethod.latin.DictionaryFacilitator;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.settings.SpacingAndPunctuations;
 import com.android.inputmethod.latin.utils.DistracterFilter.HandlingType;
 
@@ -89,7 +89,7 @@
             final DistracterFilter distracterFilter) {
         final ArrayList<LanguageModelParam> languageModelParams = new ArrayList<>();
         final int N = tokens.size();
-        PrevWordsInfo prevWordsInfo = PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+        NgramContext ngramContext = NgramContext.EMPTY_PREV_WORDS_INFO;
         for (int i = 0; i < N; ++i) {
             final String tempWord = tokens.get(i);
             if (StringUtils.isEmptyStringOrWhiteSpaces(tempWord)) {
@@ -106,7 +106,7 @@
                             + tempWord + "\"");
                 }
                 // Sentence terminator found. Split.
-                prevWordsInfo = PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+                ngramContext = NgramContext.EMPTY_PREV_WORDS_INFO;
                 continue;
             }
             if (DEBUG_TOKEN) {
@@ -114,41 +114,41 @@
             }
             final LanguageModelParam languageModelParam =
                     detectWhetherVaildWordOrNotAndGetLanguageModelParam(
-                            prevWordsInfo, tempWord, timestamp, locale, distracterFilter);
+                            ngramContext, tempWord, timestamp, locale, distracterFilter);
             if (languageModelParam == null) {
                 continue;
             }
             languageModelParams.add(languageModelParam);
-            prevWordsInfo = prevWordsInfo.getNextPrevWordsInfo(
-                    new PrevWordsInfo.WordInfo(tempWord));
+            ngramContext = ngramContext.getNextNgramContext(
+                    new NgramContext.WordInfo(tempWord));
         }
         return languageModelParams;
     }
 
     private static LanguageModelParam detectWhetherVaildWordOrNotAndGetLanguageModelParam(
-            final PrevWordsInfo prevWordsInfo, final String targetWord, final int timestamp,
+            final NgramContext ngramContext, final String targetWord, final int timestamp,
             final Locale locale, final DistracterFilter distracterFilter) {
         if (locale == null) {
             return null;
         }
-        final int wordHandlingType = distracterFilter.getWordHandlingType(prevWordsInfo,
+        final int wordHandlingType = distracterFilter.getWordHandlingType(ngramContext,
                 targetWord, locale);
         final String word = HandlingType.shouldBeLowerCased(wordHandlingType) ?
                 targetWord.toLowerCase(locale) : targetWord;
-        if (distracterFilter.isDistracterToWordsInDictionaries(prevWordsInfo, targetWord, locale)) {
+        if (distracterFilter.isDistracterToWordsInDictionaries(ngramContext, targetWord, locale)) {
             // The word is a distracter.
             return null;
         }
-        return createAndGetLanguageModelParamOfWord(prevWordsInfo, word, timestamp,
+        return createAndGetLanguageModelParamOfWord(ngramContext, word, timestamp,
                 !HandlingType.shouldBeHandledAsOov(wordHandlingType));
     }
 
     private static LanguageModelParam createAndGetLanguageModelParamOfWord(
-            final PrevWordsInfo prevWordsInfo, final String word, final int timestamp,
+            final NgramContext ngramContext, final String word, final int timestamp,
             final boolean isValidWord) {
         final int unigramProbability = isValidWord ?
                 UNIGRAM_PROBABILITY_FOR_VALID_WORD : UNIGRAM_PROBABILITY_FOR_OOV_WORD;
-        if (!prevWordsInfo.isValid()) {
+        if (!ngramContext.isValid()) {
             if (DEBUG) {
                 Log.d(TAG, "--- add unigram: current("
                         + (isValidWord ? "Valid" : "OOV") + ") = " + word);
@@ -156,12 +156,12 @@
             return new LanguageModelParam(word, unigramProbability, timestamp);
         }
         if (DEBUG) {
-            Log.d(TAG, "--- add bigram: prev = " + prevWordsInfo + ", current("
+            Log.d(TAG, "--- add bigram: prev = " + ngramContext + ", current("
                     + (isValidWord ? "Valid" : "OOV") + ") = " + word);
         }
         final int bigramProbability = isValidWord ?
                 BIGRAM_PROBABILITY_FOR_VALID_WORD : BIGRAM_PROBABILITY_FOR_OOV_WORD;
-        return new LanguageModelParam(prevWordsInfo.mPrevWordsInfo[0].mWord, word,
+        return new LanguageModelParam(ngramContext.getNthPrevWord(1 /* n */), word,
                 unigramProbability, bigramProbability, timestamp);
     }
 }
diff --git a/java/src/com/android/inputmethod/latin/utils/PrevWordsInfoUtils.java b/java/src/com/android/inputmethod/latin/utils/NgramContextUtils.java
similarity index 90%
rename from java/src/com/android/inputmethod/latin/utils/PrevWordsInfoUtils.java
rename to java/src/com/android/inputmethod/latin/utils/NgramContextUtils.java
index 5720d93..34eeac2 100644
--- a/java/src/com/android/inputmethod/latin/utils/PrevWordsInfoUtils.java
+++ b/java/src/com/android/inputmethod/latin/utils/NgramContextUtils.java
@@ -20,12 +20,12 @@
 import java.util.regex.Pattern;
 
 import com.android.inputmethod.latin.Constants;
-import com.android.inputmethod.latin.PrevWordsInfo;
-import com.android.inputmethod.latin.PrevWordsInfo.WordInfo;
+import com.android.inputmethod.latin.NgramContext;
+import com.android.inputmethod.latin.NgramContext.WordInfo;
 import com.android.inputmethod.latin.settings.SpacingAndPunctuations;
 
-public final class PrevWordsInfoUtils {
-    private PrevWordsInfoUtils() {
+public final class NgramContextUtils {
+    private NgramContextUtils() {
         // Intentional empty constructor for utility class.
     }
 
@@ -44,7 +44,7 @@
     // (n = 2) "abc def|" -> beginning-of-sentence, abc
     // (n = 2) "abc def |" -> beginning-of-sentence, abc
     // (n = 2) "abc 'def|" -> empty. The context is different from "abc def", but we cannot
-    // represent this situation using PrevWordsInfo. See TODO in the method.
+    // represent this situation using NgramContext. See TODO in the method.
     // TODO: The next example's result should be "abc, def". This have to be fixed before we
     // retrieve the prior context of Beginning-of-Sentence.
     // (n = 2) "abc def. |" -> beginning-of-sentence, abc
@@ -52,9 +52,9 @@
     // (n = 2) "abc|" -> beginning-of-sentence
     // (n = 2) "abc |" -> beginning-of-sentence
     // (n = 2) "abc. def|" -> beginning-of-sentence
-    public static PrevWordsInfo getPrevWordsInfoFromNthPreviousWord(final CharSequence prev,
+    public static NgramContext getNgramContextFromNthPreviousWord(final CharSequence prev,
             final SpacingAndPunctuations spacingAndPunctuations, final int n) {
-        if (prev == null) return PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+        if (prev == null) return NgramContext.EMPTY_PREV_WORDS_INFO;
         final String[] w = SPACE_REGEX.split(prev);
         final WordInfo[] prevWordsInfo = new WordInfo[Constants.MAX_PREV_WORD_COUNT_FOR_N_GRAM];
         Arrays.fill(prevWordsInfo, WordInfo.EMPTY_WORD_INFO);
@@ -98,6 +98,6 @@
             }
             prevWordsInfo[i] = new WordInfo(focusedWord);
         }
-        return new PrevWordsInfo(prevWordsInfo);
+        return new NgramContext(prevWordsInfo);
     }
 }
diff --git a/java/src/com/android/inputmethod/latin/utils/SuggestionResults.java b/java/src/com/android/inputmethod/latin/utils/SuggestionResults.java
index d6f6442..4e2e396 100644
--- a/java/src/com/android/inputmethod/latin/utils/SuggestionResults.java
+++ b/java/src/com/android/inputmethod/latin/utils/SuggestionResults.java
@@ -31,7 +31,7 @@
 public final class SuggestionResults extends TreeSet<SuggestedWordInfo> {
     public final ArrayList<SuggestedWordInfo> mRawSuggestions;
     // TODO: Instead of a boolean , we may want to include the context of this suggestion results,
-    // such as {@link PrevWordsInfo}.
+    // such as {@link NgramContext}.
     public final boolean mIsBeginningOfSentence;
     private final int mCapacity;
 
diff --git a/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp b/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp
index e420f80..19aeb28 100644
--- a/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp
+++ b/native/jni/com_android_inputmethod_latin_BinaryDictionary.cpp
@@ -403,10 +403,10 @@
     jsize wordLength = env->GetArrayLength(word);
     int wordCodePoints[wordLength];
     env->GetIntArrayRegion(word, 0, wordLength, wordCodePoints);
-    // Use 1 for count to indicate the bigram has inputted.
-    const BigramProperty bigramProperty(CodePointArrayView(wordCodePoints, wordLength).toVector(),
+    // Use 1 for count to indicate the ngram has inputted.
+    const NgramProperty ngramProperty(CodePointArrayView(wordCodePoints, wordLength).toVector(),
             probability, timestamp, 0 /* level */, 1 /* count */);
-    return dictionary->addNgramEntry(&prevWordsInfo, &bigramProperty);
+    return dictionary->addNgramEntry(&prevWordsInfo, &ngramProperty);
 }
 
 static bool latinime_BinaryDictionary_removeNgramEntry(JNIEnv *env, jclass clazz, jlong dict,
@@ -501,12 +501,12 @@
         if (word0) {
             jint bigramProbability = env->GetIntField(languageModelParam, bigramProbabilityFieldId);
             // Use 1 for count to indicate the bigram has inputted.
-            const BigramProperty bigramProperty(
+            const NgramProperty ngramProperty(
                     CodePointArrayView(word1CodePoints, word1Length).toVector(),
                     bigramProbability, timestamp, 0 /* level */, 1 /* count */);
             const PrevWordsInfo prevWordsInfo(word0CodePoints, word0Length,
                     false /* isBeginningOfSentence */);
-            dictionary->addNgramEntry(&prevWordsInfo, &bigramProperty);
+            dictionary->addNgramEntry(&prevWordsInfo, &ngramProperty);
         }
         if (dictionary->needsToRunGC(true /* mindsBlockByGC */)) {
             return i + 1;
@@ -603,6 +603,7 @@
     } while (token != 0);
 
     // Add bigrams.
+    // TODO: Support ngrams.
     do {
         token = dictionary->getNextWordAndNextToken(token, wordCodePoints, &wordCodePointCount);
         const WordProperty wordProperty = dictionary->getWordProperty(
@@ -617,10 +618,10 @@
         }
         const PrevWordsInfo prevWordsInfo(wordCodePoints, wordCodePointCount,
                 wordProperty.getUnigramProperty()->representsBeginningOfSentence());
-        for (const BigramProperty &bigramProperty : *wordProperty.getBigramProperties()) {
+        for (const NgramProperty &ngramProperty : *wordProperty.getNgramProperties()) {
             if (!dictionaryStructureWithBufferPolicy->addNgramEntry(&prevWordsInfo,
-                    &bigramProperty)) {
-                LogUtils::logToJava(env, "Cannot add bigram to the new dict.");
+                    &ngramProperty)) {
+                LogUtils::logToJava(env, "Cannot add ngram to the new dict.");
                 return false;
             }
         }
diff --git a/native/jni/src/suggest/core/dictionary/dictionary.cpp b/native/jni/src/suggest/core/dictionary/dictionary.cpp
index a3bb408..c3f4229 100644
--- a/native/jni/src/suggest/core/dictionary/dictionary.cpp
+++ b/native/jni/src/suggest/core/dictionary/dictionary.cpp
@@ -144,9 +144,9 @@
 }
 
 bool Dictionary::addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-        const BigramProperty *const bigramProperty) {
+        const NgramProperty *const ngramProperty) {
     TimeKeeper::setCurrentTime();
-    return mDictionaryStructureWithBufferPolicy->addNgramEntry(prevWordsInfo, bigramProperty);
+    return mDictionaryStructureWithBufferPolicy->addNgramEntry(prevWordsInfo, ngramProperty);
 }
 
 bool Dictionary::removeNgramEntry(const PrevWordsInfo *const prevWordsInfo,
diff --git a/native/jni/src/suggest/core/dictionary/dictionary.h b/native/jni/src/suggest/core/dictionary/dictionary.h
index 324e350..09f8eac 100644
--- a/native/jni/src/suggest/core/dictionary/dictionary.h
+++ b/native/jni/src/suggest/core/dictionary/dictionary.h
@@ -85,7 +85,7 @@
     bool removeUnigramEntry(const CodePointArrayView codePoints);
 
     bool addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-            const BigramProperty *const bigramProperty);
+            const NgramProperty *const ngramProperty);
 
     bool removeNgramEntry(const PrevWordsInfo *const prevWordsInfo,
             const CodePointArrayView codePoints);
diff --git a/native/jni/src/suggest/core/dictionary/property/bigram_property.h b/native/jni/src/suggest/core/dictionary/property/ngram_property.h
similarity index 83%
rename from native/jni/src/suggest/core/dictionary/property/bigram_property.h
rename to native/jni/src/suggest/core/dictionary/property/ngram_property.h
index 9e0baa0..5d822fa 100644
--- a/native/jni/src/suggest/core/dictionary/property/bigram_property.h
+++ b/native/jni/src/suggest/core/dictionary/property/ngram_property.h
@@ -14,8 +14,8 @@
  * limitations under the License.
  */
 
-#ifndef LATINIME_BIGRAM_PROPERTY_H
-#define LATINIME_BIGRAM_PROPERTY_H
+#ifndef LATINIME_NGRAM_PROPERTY_H
+#define LATINIME_NGRAM_PROPERTY_H
 
 #include <vector>
 
@@ -23,10 +23,9 @@
 
 namespace latinime {
 
-// TODO: Change to NgramProperty.
-class BigramProperty {
+class NgramProperty {
  public:
-    BigramProperty(const std::vector<int> &&targetCodePoints, const int probability,
+    NgramProperty(const std::vector<int> &&targetCodePoints, const int probability,
             const int timestamp, const int level, const int count)
             : mTargetCodePoints(std::move(targetCodePoints)), mProbability(probability),
               mTimestamp(timestamp), mLevel(level), mCount(count) {}
@@ -53,7 +52,7 @@
 
  private:
     // Default copy constructor and assign operator are used for using in std::vector.
-    DISALLOW_DEFAULT_CONSTRUCTOR(BigramProperty);
+    DISALLOW_DEFAULT_CONSTRUCTOR(NgramProperty);
 
     // TODO: Make members const.
     std::vector<int> mTargetCodePoints;
@@ -63,4 +62,4 @@
     int mCount;
 };
 } // namespace latinime
-#endif // LATINIME_WORD_PROPERTY_H
+#endif // LATINIME_NGRAM_PROPERTY_H
diff --git a/native/jni/src/suggest/core/dictionary/property/word_property.cpp b/native/jni/src/suggest/core/dictionary/property/word_property.cpp
index 66daf3e..e3406fa 100644
--- a/native/jni/src/suggest/core/dictionary/property/word_property.cpp
+++ b/native/jni/src/suggest/core/dictionary/property/word_property.cpp
@@ -28,7 +28,7 @@
             MAX_WORD_LENGTH /* maxLength */, mCodePoints.data(), mCodePoints.size(),
             false /* needsNullTermination */);
     jboolean flags[] = {mUnigramProperty.isNotAWord(), mUnigramProperty.isBlacklisted(),
-            !mBigrams.empty(), mUnigramProperty.hasShortcuts(),
+            !mNgrams.empty(), mUnigramProperty.hasShortcuts(),
             mUnigramProperty.representsBeginningOfSentence()};
     env->SetBooleanArrayRegion(outFlags, 0 /* start */, NELEMS(flags), flags);
     int probabilityInfo[] = {mUnigramProperty.getProbability(), mUnigramProperty.getTimestamp(),
@@ -42,8 +42,9 @@
     jmethodID addMethodId = env->GetMethodID(arrayListClass, "add", "(Ljava/lang/Object;)Z");
 
     // Output bigrams.
-    for (const auto &bigramProperty : mBigrams) {
-        const std::vector<int> *const word1CodePoints = bigramProperty.getTargetCodePoints();
+    // TODO: Support n-gram
+    for (const auto &ngramProperty : mNgrams) {
+        const std::vector<int> *const word1CodePoints = ngramProperty.getTargetCodePoints();
         jintArray bigramWord1CodePointArray = env->NewIntArray(word1CodePoints->size());
         JniDataUtils::outputCodePoints(env, bigramWord1CodePointArray, 0 /* start */,
                 word1CodePoints->size(), word1CodePoints->data(), word1CodePoints->size(),
@@ -51,9 +52,9 @@
         env->CallBooleanMethod(outBigramTargets, addMethodId, bigramWord1CodePointArray);
         env->DeleteLocalRef(bigramWord1CodePointArray);
 
-        int bigramProbabilityInfo[] = {bigramProperty.getProbability(),
-                bigramProperty.getTimestamp(), bigramProperty.getLevel(),
-                bigramProperty.getCount()};
+        int bigramProbabilityInfo[] = {ngramProperty.getProbability(),
+                ngramProperty.getTimestamp(), ngramProperty.getLevel(),
+                ngramProperty.getCount()};
         jintArray bigramProbabilityInfoArray = env->NewIntArray(NELEMS(bigramProbabilityInfo));
         env->SetIntArrayRegion(bigramProbabilityInfoArray, 0 /* start */,
                 NELEMS(bigramProbabilityInfo), bigramProbabilityInfo);
diff --git a/native/jni/src/suggest/core/dictionary/property/word_property.h b/native/jni/src/suggest/core/dictionary/property/word_property.h
index 4e6febb..0c23e82 100644
--- a/native/jni/src/suggest/core/dictionary/property/word_property.h
+++ b/native/jni/src/suggest/core/dictionary/property/word_property.h
@@ -21,7 +21,7 @@
 
 #include "defines.h"
 #include "jni.h"
-#include "suggest/core/dictionary/property/bigram_property.h"
+#include "suggest/core/dictionary/property/ngram_property.h"
 #include "suggest/core/dictionary/property/unigram_property.h"
 
 namespace latinime {
@@ -31,12 +31,12 @@
  public:
     // Default constructor is used to create an instance that indicates an invalid word.
     WordProperty()
-            : mCodePoints(), mUnigramProperty(), mBigrams() {}
+            : mCodePoints(), mUnigramProperty(), mNgrams() {}
 
     WordProperty(const std::vector<int> &&codePoints, const UnigramProperty *const unigramProperty,
-            const std::vector<BigramProperty> *const bigrams)
+            const std::vector<NgramProperty> *const bigrams)
             : mCodePoints(std::move(codePoints)), mUnigramProperty(*unigramProperty),
-              mBigrams(*bigrams) {}
+              mNgrams(*bigrams) {}
 
     void outputProperties(JNIEnv *const env, jintArray outCodePoints, jbooleanArray outFlags,
             jintArray outProbabilityInfo, jobject outBigramTargets, jobject outBigramProbabilities,
@@ -46,8 +46,8 @@
         return &mUnigramProperty;
     }
 
-    const std::vector<BigramProperty> *getBigramProperties() const {
-        return &mBigrams;
+    const std::vector<NgramProperty> *getNgramProperties() const {
+        return &mNgrams;
     }
 
  private:
@@ -56,7 +56,7 @@
 
     const std::vector<int> mCodePoints;
     const UnigramProperty mUnigramProperty;
-    const std::vector<BigramProperty> mBigrams;
+    const std::vector<NgramProperty> mNgrams;
 };
 } // namespace latinime
 #endif // LATINIME_WORD_PROPERTY_H
diff --git a/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h b/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h
index 1546b26..f4b9798 100644
--- a/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h
+++ b/native/jni/src/suggest/core/policy/dictionary_structure_with_buffer_policy.h
@@ -81,7 +81,7 @@
 
     // Returns whether the update was success or not.
     virtual bool addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-            const BigramProperty *const bigramProperty) = 0;
+            const NgramProperty *const ngramProperty) = 0;
 
     // Returns whether the update was success or not.
     virtual bool removeNgramEntry(const PrevWordsInfo *const prevWordsInfo,
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.cpp
index 3e8e059..c639770 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.cpp
@@ -24,7 +24,7 @@
 
 #include "suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.h"
 
-#include "suggest/core/dictionary/property/bigram_property.h"
+#include "suggest/core/dictionary/property/ngram_property.h"
 #include "suggest/policyimpl/dictionary/header/header_policy.h"
 #include "suggest/policyimpl/dictionary/structure/pt_common/bigram/bigram_list_read_write_utils.h"
 #include "suggest/policyimpl/dictionary/structure/backward/v402/content/bigram_dict_content.h"
@@ -60,7 +60,7 @@
 }
 
 bool Ver4BigramListPolicy::addNewEntry(const int terminalId, const int newTargetTerminalId,
-        const BigramProperty *const bigramProperty, bool *const outAddedNewEntry) {
+        const NgramProperty *const ngramProperty, bool *const outAddedNewEntry) {
     // 1. The word has no bigrams yet.
     // 2. The word has bigrams, and there is the target in the list.
     // 3. The word has bigrams, and there is an invalid entry that can be reclaimed.
@@ -79,7 +79,7 @@
         const BigramEntry newBigramEntry(false /* hasNext */, NOT_A_PROBABILITY,
                 newTargetTerminalId);
         const BigramEntry bigramEntryToWrite = createUpdatedBigramEntryFrom(&newBigramEntry,
-                bigramProperty);
+                ngramProperty);
         // Write an entry.
         const int writingPos =  mBigramDictContent->getBigramListHeadPos(terminalId);
         if (!mBigramDictContent->writeBigramEntry(&bigramEntryToWrite, writingPos)) {
@@ -112,7 +112,7 @@
         const BigramEntry newBigramEntry(false /* hasNext */, NOT_A_PROBABILITY,
                 newTargetTerminalId);
         const BigramEntry bigramEntryToWrite = createUpdatedBigramEntryFrom(
-                &newBigramEntry, bigramProperty);
+                &newBigramEntry, ngramProperty);
         if (!mBigramDictContent->writeBigramEntryAtTail(&bigramEntryToWrite)) {
             return false;
         }
@@ -138,7 +138,7 @@
     const BigramEntry updatedBigramEntry =
             originalBigramEntry.updateTargetTerminalIdAndGetEntry(newTargetTerminalId);
     const BigramEntry bigramEntryToWrite = createUpdatedBigramEntryFrom(
-            &updatedBigramEntry, bigramProperty);
+            &updatedBigramEntry, ngramProperty);
     return mBigramDictContent->writeBigramEntry(&bigramEntryToWrite, entryPosToUpdate);
 }
 
@@ -264,18 +264,18 @@
 
 const BigramEntry Ver4BigramListPolicy::createUpdatedBigramEntryFrom(
         const BigramEntry *const originalBigramEntry,
-        const BigramProperty *const bigramProperty) const {
+        const NgramProperty *const ngramProperty) const {
     // TODO: Consolidate historical info and probability.
     if (mHeaderPolicy->hasHistoricalInfoOfWords()) {
-        const HistoricalInfo historicalInfoForUpdate(bigramProperty->getTimestamp(),
-                bigramProperty->getLevel(), bigramProperty->getCount());
+        const HistoricalInfo historicalInfoForUpdate(ngramProperty->getTimestamp(),
+                ngramProperty->getLevel(), ngramProperty->getCount());
         const HistoricalInfo updatedHistoricalInfo =
                 ForgettingCurveUtils::createUpdatedHistoricalInfo(
-                        originalBigramEntry->getHistoricalInfo(), bigramProperty->getProbability(),
+                        originalBigramEntry->getHistoricalInfo(), ngramProperty->getProbability(),
                         &historicalInfoForUpdate, mHeaderPolicy);
         return originalBigramEntry->updateHistoricalInfoAndGetEntry(&updatedHistoricalInfo);
     } else {
-        return originalBigramEntry->updateProbabilityAndGetEntry(bigramProperty->getProbability());
+        return originalBigramEntry->updateProbabilityAndGetEntry(ngramProperty->getProbability());
     }
 }
 
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.h b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.h
index 50a4c97..aac6f54 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/bigram/ver4_bigram_list_policy.h
@@ -36,7 +36,7 @@
 class BigramDictContent;
 } // namespace v402
 } // namespace backward
-class BigramProperty;
+class NgramProperty;
 namespace backward {
 namespace v402 {
 } // namespace v402
@@ -64,7 +64,7 @@
     }
 
     bool addNewEntry(const int terminalId, const int newTargetTerminalId,
-            const BigramProperty *const bigramProperty, bool *const outAddedNewEntry);
+            const NgramProperty *const ngramProperty, bool *const outAddedNewEntry);
 
     bool removeEntry(const int terminalId, const int targetTerminalId);
 
@@ -80,7 +80,7 @@
             int *const outTailEntryPos) const;
 
     const BigramEntry createUpdatedBigramEntryFrom(const BigramEntry *const originalBigramEntry,
-            const BigramProperty *const bigramProperty) const;
+            const NgramProperty *const ngramProperty) const;
 
     bool updateHasNextFlag(const bool hasNext, const int bigramEntryPos);
 
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.cpp
index 97a8bcc..6827c3e 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.cpp
@@ -232,8 +232,8 @@
 }
 
 bool Ver4PatriciaTrieNodeWriter::addNgramEntry(const WordIdArrayView prevWordIds, const int wordId,
-        const BigramProperty *const bigramProperty, bool *const outAddedNewEntry) {
-    if (!mBigramPolicy->addNewEntry(prevWordIds[0], wordId, bigramProperty, outAddedNewEntry)) {
+        const NgramProperty *const ngramProperty, bool *const outAddedNewEntry) {
+    if (!mBigramPolicy->addNewEntry(prevWordIds[0], wordId, ngramProperty, outAddedNewEntry)) {
         AKLOGE("Cannot add new bigram entry. prevWordId: %d, wordId: %d",
                 prevWordIds[0], wordId);
         return false;
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.h b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.h
index 9d8a55b..d0bab50 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_node_writer.h
@@ -94,7 +94,7 @@
             const UnigramProperty *const unigramProperty, int *const ptNodeWritingPos);
 
     virtual bool addNgramEntry(const WordIdArrayView prevWordIds, const int wordId,
-            const BigramProperty *const bigramProperty, bool *const outAddedNewEntry);
+            const NgramProperty *const ngramProperty, bool *const outAddedNewEntry);
 
     virtual bool removeNgramEntry(const WordIdArrayView prevWordIds, const int wordId);
 
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.cpp
index f752f89..9631cf1 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.cpp
@@ -30,7 +30,7 @@
 #include "suggest/core/dicnode/dic_node_vector.h"
 #include "suggest/core/dictionary/multi_bigram_map.h"
 #include "suggest/core/dictionary/ngram_listener.h"
-#include "suggest/core/dictionary/property/bigram_property.h"
+#include "suggest/core/dictionary/property/ngram_property.h"
 #include "suggest/core/dictionary/property/unigram_property.h"
 #include "suggest/core/dictionary/property/word_property.h"
 #include "suggest/core/session/prev_words_info.h"
@@ -312,7 +312,7 @@
 }
 
 bool Ver4PatriciaTriePolicy::addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-        const BigramProperty *const bigramProperty) {
+        const NgramProperty *const ngramProperty) {
     if (!mBuffers->isUpdatable()) {
         AKLOGI("Warning: addNgramEntry() is called for non-updatable dictionary.");
         return false;
@@ -326,9 +326,9 @@
         AKLOGE("prev words info is not valid for adding n-gram entry to the dictionary.");
         return false;
     }
-    if (bigramProperty->getTargetCodePoints()->size() > MAX_WORD_LENGTH) {
+    if (ngramProperty->getTargetCodePoints()->size() > MAX_WORD_LENGTH) {
         AKLOGE("The word is too long to insert the ngram to the dictionary. "
-                "length: %zd", bigramProperty->getTargetCodePoints()->size());
+                "length: %zd", ngramProperty->getTargetCodePoints()->size());
         return false;
     }
     WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
@@ -356,7 +356,7 @@
         }
     }
     const int wordPos = getTerminalPtNodePosFromWordId(getWordId(
-            CodePointArrayView(*bigramProperty->getTargetCodePoints()),
+            CodePointArrayView(*ngramProperty->getTargetCodePoints()),
                     false /* forceLowerCaseSearch */));
     if (wordPos == NOT_A_DICT_POS) {
         return false;
@@ -364,7 +364,7 @@
     bool addedNewBigram = false;
     const int prevWordPtNodePos = getTerminalPtNodePosFromWordId(prevWordIds[0]);
     if (mUpdatingHelper.addNgramEntry(PtNodePosArrayView::singleElementView(&prevWordPtNodePos),
-            wordPos, bigramProperty, &addedNewBigram)) {
+            wordPos, ngramProperty, &addedNewBigram)) {
         if (addedNewBigram) {
             mBigramCount++;
         }
@@ -499,7 +499,7 @@
                     ptNodeParams.getTerminalId());
     const HistoricalInfo *const historicalInfo = probabilityEntry.getHistoricalInfo();
     // Fetch bigram information.
-    std::vector<BigramProperty> bigrams;
+    std::vector<NgramProperty> ngrams;
     const int bigramListPos = getBigramsPositionOfPtNode(ptNodePos);
     if (bigramListPos != NOT_A_DICT_POS) {
         int bigramWord1CodePoints[MAX_WORD_LENGTH];
@@ -526,7 +526,7 @@
                     ForgettingCurveUtils::decodeProbability(
                             bigramEntry.getHistoricalInfo(), mHeaderPolicy) :
                     bigramEntry.getProbability();
-            bigrams.emplace_back(
+            ngrams.emplace_back(
                     CodePointArrayView(bigramWord1CodePoints, codePointCount).toVector(),
                     probability, historicalInfo->getTimeStamp(), historicalInfo->getLevel(),
                     historicalInfo->getCount());
@@ -554,7 +554,7 @@
             ptNodeParams.isNotAWord(), ptNodeParams.isBlacklisted(), ptNodeParams.getProbability(),
             historicalInfo->getTimeStamp(), historicalInfo->getLevel(),
             historicalInfo->getCount(), &shortcuts);
-    return WordProperty(wordCodePoints.toVector(), &unigramProperty, &bigrams);
+    return WordProperty(wordCodePoints.toVector(), &unigramProperty, &ngrams);
 }
 
 int Ver4PatriciaTriePolicy::getNextWordAndNextToken(const int token, int *const outCodePoints,
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.h b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.h
index 8420c94..324a53e 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/backward/v402/ver4_patricia_trie_policy.h
@@ -59,6 +59,7 @@
 namespace v402 {
 
 // Word id = Position of a PtNode that represents the word.
+// Max supported n-gram is bigram.
 class Ver4PatriciaTriePolicy : public DictionaryStructureWithBufferPolicy {
  public:
     Ver4PatriciaTriePolicy(Ver4DictBuffers::Ver4DictBuffersPtr buffers)
@@ -112,7 +113,7 @@
     bool removeUnigramEntry(const CodePointArrayView wordCodePoints);
 
     bool addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-            const BigramProperty *const bigramProperty);
+            const NgramProperty *const ngramProperty);
 
     bool removeNgramEntry(const PrevWordsInfo *const prevWordsInfo,
             const CodePointArrayView wordCodePoints);
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_gc_event_listeners.h b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_gc_event_listeners.h
index 2aa4027..b8a4a92 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_gc_event_listeners.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_gc_event_listeners.h
@@ -76,6 +76,7 @@
         int mValidUnigramCount;
     };
 
+    // TODO: Remove when we stop supporting v402 format.
     // Updates all bigram entries that are held by valid PtNodes. This removes useless bigram
     // entries.
     class TraversePolicyToUpdateBigramProbability
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.cpp
index 3b58d7d..92fd6f2 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.cpp
@@ -82,7 +82,7 @@
 }
 
 bool DynamicPtUpdatingHelper::addNgramEntry(const PtNodePosArrayView prevWordsPtNodePos,
-        const int wordPos, const BigramProperty *const bigramProperty,
+        const int wordPos, const NgramProperty *const ngramProperty,
         bool *const outAddedNewEntry) {
     if (prevWordsPtNodePos.empty()) {
         return false;
@@ -96,7 +96,7 @@
     const WordIdArrayView prevWordIds(prevWordTerminalIds, prevWordsPtNodePos.size());
     const int wordId =
             mPtNodeReader->fetchPtNodeParamsInBufferFromPtNodePos(wordPos).getTerminalId();
-    return mPtNodeWriter->addNgramEntry(prevWordIds, wordId, bigramProperty, outAddedNewEntry);
+    return mPtNodeWriter->addNgramEntry(prevWordIds, wordId, ngramProperty, outAddedNewEntry);
 }
 
 bool DynamicPtUpdatingHelper::removeNgramEntry(const PtNodePosArrayView prevWordsPtNodePos,
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.h b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.h
index 710047e..2bbe2f4 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/dynamic_pt_updating_helper.h
@@ -23,7 +23,7 @@
 
 namespace latinime {
 
-class BigramProperty;
+class NgramProperty;
 class BufferWithExtendableBuffer;
 class DynamicPtReadingHelper;
 class PtNodeReader;
@@ -46,7 +46,7 @@
     // TODO: Remove after stopping supporting v402.
     // Add an n-gram entry.
     bool addNgramEntry(const PtNodePosArrayView prevWordsPtNodePos, const int wordPos,
-            const BigramProperty *const bigramProperty, bool *const outAddedNewEntry);
+            const NgramProperty *const ngramProperty, bool *const outAddedNewEntry);
 
     // TODO: Remove after stopping supporting v402.
     // Remove an n-gram entry.
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/pt_node_writer.h b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/pt_node_writer.h
index 955d779..954db9b 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/pt_node_writer.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/pt_common/pt_node_writer.h
@@ -25,7 +25,7 @@
 
 namespace latinime {
 
-class BigramProperty;
+class NgramProperty;
 class UnigramProperty;
 
 // Interface class used to write PtNode information.
@@ -72,7 +72,7 @@
             const UnigramProperty *const unigramProperty, int *const ptNodeWritingPos) = 0;
 
     virtual bool addNgramEntry(const WordIdArrayView prevWordIds, const int wordId,
-            const BigramProperty *const bigramProperty, bool *const outAddedNewEntry) = 0;
+            const NgramProperty *const ngramProperty, bool *const outAddedNewEntry) = 0;
 
     virtual bool removeNgramEntry(const WordIdArrayView prevWordIds, const int wordId) = 0;
 
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.cpp
index 13cf9a5..49095d0 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.cpp
@@ -436,7 +436,7 @@
     const PtNodeParams ptNodeParams =
             mPtNodeReader.fetchPtNodeParamsInBufferFromPtNodePos(ptNodePos);
     // Fetch bigram information.
-    std::vector<BigramProperty> bigrams;
+    std::vector<NgramProperty> ngrams;
     const int bigramListPos = getBigramsPositionOfPtNode(ptNodePos);
     int bigramWord1CodePoints[MAX_WORD_LENGTH];
     BinaryDictionaryBigramsIterator bigramsIt(&mBigramListPolicy, bigramListPos);
@@ -450,7 +450,7 @@
                     getWordIdFromTerminalPtNodePos(bigramsIt.getBigramPos()), MAX_WORD_LENGTH,
                     bigramWord1CodePoints, &word1Probability);
             const int probability = getProbability(word1Probability, bigramsIt.getProbability());
-            bigrams.emplace_back(
+            ngrams.emplace_back(
                     CodePointArrayView(bigramWord1CodePoints, word1CodePointCount).toVector(),
                     probability, NOT_A_TIMESTAMP /* timestamp */, 0 /* level */, 0 /* count */);
         }
@@ -478,7 +478,7 @@
     const UnigramProperty unigramProperty(ptNodeParams.representsBeginningOfSentence(),
             ptNodeParams.isNotAWord(), ptNodeParams.isBlacklisted(), ptNodeParams.getProbability(),
             NOT_A_TIMESTAMP /* timestamp */, 0 /* level */, 0 /* count */, &shortcuts);
-    return WordProperty(wordCodePoints.toVector(), &unigramProperty, &bigrams);
+    return WordProperty(wordCodePoints.toVector(), &unigramProperty, &ngrams);
 }
 
 int PatriciaTriePolicy::getNextWordAndNextToken(const int token, int *const outCodePoints,
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.h b/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.h
index 0d679c5..a912d03 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v2/patricia_trie_policy.h
@@ -38,6 +38,7 @@
 class DicNodeVector;
 
 // Word id = Position of a PtNode that represents the word.
+// Max supported n-gram is bigram.
 class PatriciaTriePolicy : public DictionaryStructureWithBufferPolicy {
  public:
     PatriciaTriePolicy(MmappedBuffer::MmappedBufferPtr mmappedBuffer)
@@ -93,7 +94,7 @@
     }
 
     bool addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-            const BigramProperty *const bigramProperty) {
+            const NgramProperty *const ngramProperty) {
         // This method should not be called for non-updatable dictionary.
         AKLOGI("Warning: addNgramEntry() is called for non-updatable dictionary.");
         return false;
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/content/probability_entry.h b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/content/probability_entry.h
index e1e10ca..fd52e57 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/content/probability_entry.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/content/probability_entry.h
@@ -21,7 +21,7 @@
 #include <cstdint>
 
 #include "defines.h"
-#include "suggest/core/dictionary/property/bigram_property.h"
+#include "suggest/core/dictionary/property/ngram_property.h"
 #include "suggest/core/dictionary/property/unigram_property.h"
 #include "suggest/policyimpl/dictionary/structure/v4/ver4_dict_constants.h"
 #include "suggest/policyimpl/dictionary/utils/historical_info.h"
@@ -56,12 +56,12 @@
               mHistoricalInfo(unigramProperty->getTimestamp(), unigramProperty->getLevel(),
                       unigramProperty->getCount()) {}
 
-    // Create from bigram property.
+    // Create from ngram property.
     // TODO: Set flags.
-    ProbabilityEntry(const BigramProperty *const bigramProperty)
-            : mFlags(0), mProbability(bigramProperty->getProbability()),
-              mHistoricalInfo(bigramProperty->getTimestamp(), bigramProperty->getLevel(),
-                      bigramProperty->getCount()) {}
+    ProbabilityEntry(const NgramProperty *const ngramProperty)
+            : mFlags(0), mProbability(ngramProperty->getProbability()),
+              mHistoricalInfo(ngramProperty->getTimestamp(), ngramProperty->getLevel(),
+                      ngramProperty->getCount()) {}
 
     bool isValid() const {
         return (mFlags & Ver4DictConstants::FLAG_NOT_A_VALID_ENTRY) == 0;
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.cpp
index a1a33d2..f13512d 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.cpp
@@ -61,6 +61,7 @@
     }
 }
 
+// TODO: Quit using bigramLinkedNodePos.
 bool Ver4PatriciaTrieNodeWriter::markPtNodeAsMoved(
         const PtNodeParams *const toBeUpdatedPtNodeParams,
         const int movedPos, const int bigramLinkedNodePos) {
@@ -208,15 +209,16 @@
             terminalId, &probabilityEntryToWrite);
 }
 
+// TODO: Support counting ngram entries.
 bool Ver4PatriciaTrieNodeWriter::addNgramEntry(const WordIdArrayView prevWordIds, const int wordId,
-        const BigramProperty *const bigramProperty, bool *const outAddedNewBigram) {
+        const NgramProperty *const ngramProperty, bool *const outAddedNewBigram) {
     LanguageModelDictContent *const languageModelDictContent =
             mBuffers->getMutableLanguageModelDictContent();
     const ProbabilityEntry probabilityEntry =
             languageModelDictContent->getNgramProbabilityEntry(prevWordIds, wordId);
-    const ProbabilityEntry probabilityEntryOfBigramProperty(bigramProperty);
+    const ProbabilityEntry probabilityEntryOfNgramProperty(ngramProperty);
     const ProbabilityEntry updatedProbabilityEntry = createUpdatedEntryFrom(
-            &probabilityEntry, &probabilityEntryOfBigramProperty);
+            &probabilityEntry, &probabilityEntryOfNgramProperty);
     if (!languageModelDictContent->setNgramProbabilityEntry(
             prevWordIds, wordId, &updatedProbabilityEntry)) {
         AKLOGE("Cannot add new ngram entry. prevWordId[0]: %d, prevWordId.size(): %zd, wordId: %d",
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.h b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.h
index 1791527..ea4f099 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_node_writer.h
@@ -74,7 +74,7 @@
             const UnigramProperty *const unigramProperty, int *const ptNodeWritingPos);
 
     virtual bool addNgramEntry(const WordIdArrayView prevWordIds, const int wordId,
-            const BigramProperty *const bigramProperty, bool *const outAddedNewEntry);
+            const NgramProperty *const ngramProperty, bool *const outAddedNewEntry);
 
     virtual bool removeNgramEntry(const WordIdArrayView prevWordIds, const int wordId);
 
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.cpp
index 0f06964..d46acf5 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.cpp
@@ -23,7 +23,7 @@
 #include "suggest/core/dicnode/dic_node_vector.h"
 #include "suggest/core/dictionary/multi_bigram_map.h"
 #include "suggest/core/dictionary/ngram_listener.h"
-#include "suggest/core/dictionary/property/bigram_property.h"
+#include "suggest/core/dictionary/property/ngram_property.h"
 #include "suggest/core/dictionary/property/unigram_property.h"
 #include "suggest/core/dictionary/property/word_property.h"
 #include "suggest/core/session/prev_words_info.h"
@@ -266,7 +266,7 @@
 }
 
 bool Ver4PatriciaTriePolicy::addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-        const BigramProperty *const bigramProperty) {
+        const NgramProperty *const ngramProperty) {
     if (!mBuffers->isUpdatable()) {
         AKLOGI("Warning: addNgramEntry() is called for non-updatable dictionary.");
         return false;
@@ -280,9 +280,9 @@
         AKLOGE("prev words info is not valid for adding n-gram entry to the dictionary.");
         return false;
     }
-    if (bigramProperty->getTargetCodePoints()->size() > MAX_WORD_LENGTH) {
+    if (ngramProperty->getTargetCodePoints()->size() > MAX_WORD_LENGTH) {
         AKLOGE("The word is too long to insert the ngram to the dictionary. "
-                "length: %zd", bigramProperty->getTargetCodePoints()->size());
+                "length: %zd", ngramProperty->getTargetCodePoints()->size());
         return false;
     }
     WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
@@ -311,13 +311,13 @@
         // Refresh word ids.
         prevWordsInfo->getPrevWordIds(this, &prevWordIdArray, false /* tryLowerCaseSearch */);
     }
-    const int wordId = getWordId(CodePointArrayView(*bigramProperty->getTargetCodePoints()),
+    const int wordId = getWordId(CodePointArrayView(*ngramProperty->getTargetCodePoints()),
             false /* forceLowerCaseSearch */);
     if (wordId == NOT_A_WORD_ID) {
         return false;
     }
     bool addedNewEntry = false;
-    if (mNodeWriter.addNgramEntry(prevWordIds, wordId, bigramProperty, &addedNewEntry)) {
+    if (mNodeWriter.addNgramEntry(prevWordIds, wordId, ngramProperty, &addedNewEntry)) {
         if (addedNewEntry) {
             mBigramCount++;
         }
@@ -451,7 +451,7 @@
     const HistoricalInfo *const historicalInfo = probabilityEntry.getHistoricalInfo();
     // Fetch bigram information.
     // TODO: Support n-gram.
-    std::vector<BigramProperty> bigrams;
+    std::vector<NgramProperty> ngrams;
     const WordIdArrayView prevWordIds = WordIdArrayView::singleElementView(&wordId);
     int bigramWord1CodePoints[MAX_WORD_LENGTH];
     for (const auto entry : mBuffers->getLanguageModelDictContent()->getProbabilityEntries(
@@ -463,7 +463,7 @@
         const int probability = probabilityEntry.hasHistoricalInfo() ?
                 ForgettingCurveUtils::decodeProbability(historicalInfo, mHeaderPolicy) :
                 probabilityEntry.getProbability();
-        bigrams.emplace_back(CodePointArrayView(bigramWord1CodePoints, codePointCount).toVector(),
+        ngrams.emplace_back(CodePointArrayView(bigramWord1CodePoints, codePointCount).toVector(),
                 probability, historicalInfo->getTimeStamp(), historicalInfo->getLevel(),
                 historicalInfo->getCount());
     }
@@ -489,7 +489,7 @@
             probabilityEntry.isNotAWord(), probabilityEntry.isBlacklisted(),
             probabilityEntry.getProbability(), historicalInfo->getTimeStamp(),
             historicalInfo->getLevel(), historicalInfo->getCount(), &shortcuts);
-    return WordProperty(wordCodePoints.toVector(), &unigramProperty, &bigrams);
+    return WordProperty(wordCodePoints.toVector(), &unigramProperty, &ngrams);
 }
 
 int Ver4PatriciaTriePolicy::getNextWordAndNextToken(const int token, int *const outCodePoints,
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.h b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.h
index c9bde2c..598122b 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_policy.h
@@ -37,6 +37,7 @@
 class DicNode;
 class DicNodeVector;
 
+// TODO: Support counting ngram entries.
 // Word id = Artificial id that is stored in the PtNode looked up by the word.
 class Ver4PatriciaTriePolicy : public DictionaryStructureWithBufferPolicy {
  public:
@@ -92,7 +93,7 @@
     bool removeUnigramEntry(const CodePointArrayView wordCodePoints);
 
     bool addNgramEntry(const PrevWordsInfo *const prevWordsInfo,
-            const BigramProperty *const bigramProperty);
+            const NgramProperty *const ngramProperty);
 
     bool removeNgramEntry(const PrevWordsInfo *const prevWordsInfo,
             const CodePointArrayView wordCodePoints);
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.cpp b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.cpp
index 442abad..e1ff973 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.cpp
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.cpp
@@ -114,14 +114,6 @@
         return false;
     }
 
-    readingHelper.initWithPtNodeArrayPos(rootPtNodeArrayPos);
-    DynamicPtGcEventListeners::TraversePolicyToUpdateBigramProbability
-            traversePolicyToUpdateBigramProbability(&ptNodeWriter);
-    if (!readingHelper.traverseAllPtNodesInPostorderDepthFirstManner(
-            &traversePolicyToUpdateBigramProbability)) {
-        return false;
-    }
-
     // Mapping from positions in mBuffer to positions in bufferToWrite.
     PtNodeWriter::DictPositionRelocationMap dictPositionRelocationMap;
     readingHelper.initWithPtNodeArrayPos(rootPtNodeArrayPos);
diff --git a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.h b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.h
index b6278c4..3569d05 100644
--- a/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.h
+++ b/native/jni/src/suggest/policyimpl/dictionary/structure/v4/ver4_patricia_trie_writing_helper.h
@@ -33,6 +33,7 @@
     Ver4PatriciaTrieWritingHelper(Ver4DictBuffers *const buffers)
             : mBuffers(buffers) {}
 
+    // TODO: Support counting ngram entries.
     bool writeToDictFile(const char *const dictDirPath, const int unigramCount,
             const int bigramCount) const;
 
@@ -70,11 +71,6 @@
             Ver4DictBuffers *const buffersToWrite, int *const outUnigramCount,
             int *const outBigramCount);
 
-    bool truncateUnigrams(const Ver4PatriciaTrieNodeReader *const ptNodeReader,
-            Ver4PatriciaTrieNodeWriter *const ptNodeWriter, const int maxUnigramCount);
-
-    bool truncateBigrams(const int maxBigramCount);
-
     Ver4DictBuffers *const mBuffers;
 };
 } // namespace latinime
diff --git a/tests/src/com/android/inputmethod/latin/BinaryDictionaryDecayingTests.java b/tests/src/com/android/inputmethod/latin/BinaryDictionaryDecayingTests.java
index f3bbe4a..8d5d6cc 100644
--- a/tests/src/com/android/inputmethod/latin/BinaryDictionaryDecayingTests.java
+++ b/tests/src/com/android/inputmethod/latin/BinaryDictionaryDecayingTests.java
@@ -20,7 +20,7 @@
 import android.test.suitebuilder.annotation.LargeTest;
 import android.util.Pair;
 
-import com.android.inputmethod.latin.PrevWordsInfo.WordInfo;
+import com.android.inputmethod.latin.NgramContext.WordInfo;
 import com.android.inputmethod.latin.makedict.BinaryDictIOUtils;
 import com.android.inputmethod.latin.makedict.CodePointUtils;
 import com.android.inputmethod.latin.makedict.DictDecoder;
@@ -78,13 +78,13 @@
 
     private void addBigramWords(final BinaryDictionary binaryDictionary, final String word0,
             final String word1, final int probability) {
-        binaryDictionary.addNgramEntry(new PrevWordsInfo(new WordInfo(word0)), word1, probability,
+        binaryDictionary.addNgramEntry(new NgramContext(new WordInfo(word0)), word1, probability,
                 mCurrentTime /* timestamp */);
     }
 
     private static boolean isValidBigram(final BinaryDictionary binaryDictionary,
             final String word0, final String word1) {
-        return binaryDictionary.isValidNgram(new PrevWordsInfo(new WordInfo(word0)), word1);
+        return binaryDictionary.isValidNgram(new NgramContext(new WordInfo(word0)), word1);
     }
 
     private void forcePassingShortTime(final BinaryDictionary binaryDictionary) {
@@ -661,31 +661,31 @@
                 BinaryDictionary.NOT_A_PROBABILITY /* shortcutProbability */,
                 true /* isBeginningOfSentence */, true /* isNotAWord */, false /* isBlacklisted */,
                 mCurrentTime);
-        final PrevWordsInfo prevWordsInfoStartOfSentence = PrevWordsInfo.BEGINNING_OF_SENTENCE;
+        final NgramContext beginningOfSentenceContext = NgramContext.BEGINNING_OF_SENTENCE;
         addUnigramWord(binaryDictionary, "aaa", DUMMY_PROBABILITY);
-        binaryDictionary.addNgramEntry(prevWordsInfoStartOfSentence, "aaa", DUMMY_PROBABILITY,
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "aaa", DUMMY_PROBABILITY,
                 mCurrentTime);
-        assertTrue(binaryDictionary.isValidNgram(prevWordsInfoStartOfSentence, "aaa"));
-        binaryDictionary.addNgramEntry(prevWordsInfoStartOfSentence, "aaa", DUMMY_PROBABILITY,
+        assertTrue(binaryDictionary.isValidNgram(beginningOfSentenceContext, "aaa"));
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "aaa", DUMMY_PROBABILITY,
                 mCurrentTime);
         addUnigramWord(binaryDictionary, "bbb", DUMMY_PROBABILITY);
-        binaryDictionary.addNgramEntry(prevWordsInfoStartOfSentence, "bbb", DUMMY_PROBABILITY,
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "bbb", DUMMY_PROBABILITY,
                 mCurrentTime);
-        assertTrue(binaryDictionary.isValidNgram(prevWordsInfoStartOfSentence, "aaa"));
-        assertTrue(binaryDictionary.isValidNgram(prevWordsInfoStartOfSentence, "bbb"));
+        assertTrue(binaryDictionary.isValidNgram(beginningOfSentenceContext, "aaa"));
+        assertTrue(binaryDictionary.isValidNgram(beginningOfSentenceContext, "bbb"));
 
         forcePassingLongTime(binaryDictionary);
-        assertFalse(binaryDictionary.isValidNgram(prevWordsInfoStartOfSentence, "aaa"));
-        assertFalse(binaryDictionary.isValidNgram(prevWordsInfoStartOfSentence, "bbb"));
+        assertFalse(binaryDictionary.isValidNgram(beginningOfSentenceContext, "aaa"));
+        assertFalse(binaryDictionary.isValidNgram(beginningOfSentenceContext, "bbb"));
 
         addUnigramWord(binaryDictionary, "aaa", DUMMY_PROBABILITY);
-        binaryDictionary.addNgramEntry(prevWordsInfoStartOfSentence, "aaa", DUMMY_PROBABILITY,
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "aaa", DUMMY_PROBABILITY,
                 mCurrentTime);
         addUnigramWord(binaryDictionary, "bbb", DUMMY_PROBABILITY);
-        binaryDictionary.addNgramEntry(prevWordsInfoStartOfSentence, "bbb", DUMMY_PROBABILITY,
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "bbb", DUMMY_PROBABILITY,
                 mCurrentTime);
-        assertTrue(binaryDictionary.isValidNgram(prevWordsInfoStartOfSentence, "aaa"));
-        assertTrue(binaryDictionary.isValidNgram(prevWordsInfoStartOfSentence, "bbb"));
+        assertTrue(binaryDictionary.isValidNgram(beginningOfSentenceContext, "aaa"));
+        assertTrue(binaryDictionary.isValidNgram(beginningOfSentenceContext, "bbb"));
         binaryDictionary.close();
         dictFile.delete();
     }
diff --git a/tests/src/com/android/inputmethod/latin/BinaryDictionaryTests.java b/tests/src/com/android/inputmethod/latin/BinaryDictionaryTests.java
index 4025744..e6f00b6 100644
--- a/tests/src/com/android/inputmethod/latin/BinaryDictionaryTests.java
+++ b/tests/src/com/android/inputmethod/latin/BinaryDictionaryTests.java
@@ -21,7 +21,7 @@
 import android.text.TextUtils;
 import android.util.Pair;
 
-import com.android.inputmethod.latin.PrevWordsInfo.WordInfo;
+import com.android.inputmethod.latin.NgramContext.WordInfo;
 import com.android.inputmethod.latin.makedict.CodePointUtils;
 import com.android.inputmethod.latin.makedict.FormatSpec;
 import com.android.inputmethod.latin.makedict.WeightedString;
@@ -208,45 +208,45 @@
 
     private static void addBigramWords(final BinaryDictionary binaryDictionary, final String word0,
             final String word1, final int probability) {
-        binaryDictionary.addNgramEntry(new PrevWordsInfo(new WordInfo(word0)), word1, probability,
+        binaryDictionary.addNgramEntry(new NgramContext(new WordInfo(word0)), word1, probability,
                 BinaryDictionary.NOT_A_VALID_TIMESTAMP /* timestamp */);
     }
 
     private static void addTrigramEntry(final BinaryDictionary binaryDictionary, final String word0,
             final String word1, final String word2, final int probability) {
-        final PrevWordsInfo prevWordsInfo =
-                new PrevWordsInfo(new WordInfo[] { new WordInfo(word1), new WordInfo(word0) } );
-        binaryDictionary.addNgramEntry(prevWordsInfo, word2, probability,
+        final NgramContext ngramContext =
+                new NgramContext(new WordInfo[] { new WordInfo(word1), new WordInfo(word0) } );
+        binaryDictionary.addNgramEntry(ngramContext, word2, probability,
                 BinaryDictionary.NOT_A_VALID_TIMESTAMP /* timestamp */);
     }
 
     private static boolean isValidBigram(final BinaryDictionary binaryDictionary,
             final String word0, final String word1) {
-        return binaryDictionary.isValidNgram(new PrevWordsInfo(new WordInfo(word0)), word1);
+        return binaryDictionary.isValidNgram(new NgramContext(new WordInfo(word0)), word1);
     }
 
     private static void removeBigramEntry(final BinaryDictionary binaryDictionary,
             final String word0, final String word1) {
-        binaryDictionary.removeNgramEntry(new PrevWordsInfo(new WordInfo(word0)), word1);
+        binaryDictionary.removeNgramEntry(new NgramContext(new WordInfo(word0)), word1);
     }
 
     private static void removeTrigramEntry(final BinaryDictionary binaryDictionary,
             final String word0, final String word1, final String word2) {
-        final PrevWordsInfo prevWordsInfo =
-                new PrevWordsInfo(new WordInfo[] { new WordInfo(word1), new WordInfo(word0) } );
-        binaryDictionary.removeNgramEntry(prevWordsInfo, word2);
+        final NgramContext ngramContext =
+                new NgramContext(new WordInfo[] { new WordInfo(word1), new WordInfo(word0) } );
+        binaryDictionary.removeNgramEntry(ngramContext, word2);
     }
 
     private static int getBigramProbability(final BinaryDictionary binaryDictionary,
             final String word0,  final String word1) {
-        return binaryDictionary.getNgramProbability(new PrevWordsInfo(new WordInfo(word0)), word1);
+        return binaryDictionary.getNgramProbability(new NgramContext(new WordInfo(word0)), word1);
     }
 
     private static int getTrigramProbability(final BinaryDictionary binaryDictionary,
             final String word0, final String word1, final String word2) {
-        final PrevWordsInfo prevWordsInfo =
-                new PrevWordsInfo(new WordInfo[] { new WordInfo(word1), new WordInfo(word0) } );
-        return binaryDictionary.getNgramProbability(prevWordsInfo, word2);
+        final NgramContext ngramContext =
+                new NgramContext(new WordInfo[] { new WordInfo(word1), new WordInfo(word0) } );
+        return binaryDictionary.getNgramProbability(ngramContext, word2);
     }
 
     public void testAddUnigramWord() {
@@ -1422,7 +1422,7 @@
         binaryDictionary.addUnigramEntry("ddd", unigramProbability, null /* shortcutTarget */,
                 Dictionary.NOT_A_PROBABILITY, false /* isBeginningOfSentence */,
                 true /* isNotAWord */, true /* isBlacklisted */, 0 /* timestamp */);
-        binaryDictionary.addNgramEntry(PrevWordsInfo.BEGINNING_OF_SENTENCE,
+        binaryDictionary.addNgramEntry(NgramContext.BEGINNING_OF_SENTENCE,
                 "aaa", bigramProbability, 0 /* timestamp */);
         assertEquals(unigramProbability, binaryDictionary.getFrequency("aaa"));
         assertEquals(unigramProbability, binaryDictionary.getFrequency("bbb"));
@@ -1436,7 +1436,7 @@
         if (canCheckBigramProbability(toFormatVersion)) {
             assertEquals(bigramProbability, getBigramProbability(binaryDictionary, "aaa", "bbb"));
             assertEquals(bigramProbability, binaryDictionary.getNgramProbability(
-                    PrevWordsInfo.BEGINNING_OF_SENTENCE, "aaa"));
+                    NgramContext.BEGINNING_OF_SENTENCE, "aaa"));
         }
         assertTrue(isValidBigram(binaryDictionary, "aaa", "bbb"));
         WordProperty wordProperty = binaryDictionary.getWordProperty("ccc",
@@ -1546,23 +1546,23 @@
                 0 /* offset */, dictFile.length(), true /* useFullEditDistance */,
                 Locale.getDefault(), TEST_LOCALE, true /* isUpdatable */);
         final int dummyProbability = 0;
-        final PrevWordsInfo prevWordsInfoBeginningOfSentence = PrevWordsInfo.BEGINNING_OF_SENTENCE;
+        final NgramContext beginningOfSentenceContext = NgramContext.BEGINNING_OF_SENTENCE;
         final int bigramProbability = 200;
         addUnigramWord(binaryDictionary, "aaa", dummyProbability);
-        binaryDictionary.addNgramEntry(prevWordsInfoBeginningOfSentence, "aaa", bigramProbability,
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "aaa", bigramProbability,
                 BinaryDictionary.NOT_A_VALID_TIMESTAMP /* timestamp */);
         assertEquals(bigramProbability,
-                binaryDictionary.getNgramProbability(prevWordsInfoBeginningOfSentence, "aaa"));
-        binaryDictionary.addNgramEntry(prevWordsInfoBeginningOfSentence, "aaa", bigramProbability,
+                binaryDictionary.getNgramProbability(beginningOfSentenceContext, "aaa"));
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "aaa", bigramProbability,
                 BinaryDictionary.NOT_A_VALID_TIMESTAMP /* timestamp */);
         addUnigramWord(binaryDictionary, "bbb", dummyProbability);
-        binaryDictionary.addNgramEntry(prevWordsInfoBeginningOfSentence, "bbb", bigramProbability,
+        binaryDictionary.addNgramEntry(beginningOfSentenceContext, "bbb", bigramProbability,
                 BinaryDictionary.NOT_A_VALID_TIMESTAMP /* timestamp */);
         binaryDictionary.flushWithGC();
         assertEquals(bigramProbability,
-                binaryDictionary.getNgramProbability(prevWordsInfoBeginningOfSentence, "aaa"));
+                binaryDictionary.getNgramProbability(beginningOfSentenceContext, "aaa"));
         assertEquals(bigramProbability,
-                binaryDictionary.getNgramProbability(prevWordsInfoBeginningOfSentence, "bbb"));
+                binaryDictionary.getNgramProbability(beginningOfSentenceContext, "bbb"));
     }
 
     public void testGetMaxFrequencyOfExactMatches() {
diff --git a/tests/src/com/android/inputmethod/latin/BlueUnderlineTests.java b/tests/src/com/android/inputmethod/latin/BlueUnderlineTests.java
index 6e894de..30b0881 100644
--- a/tests/src/com/android/inputmethod/latin/BlueUnderlineTests.java
+++ b/tests/src/com/android/inputmethod/latin/BlueUnderlineTests.java
@@ -63,7 +63,7 @@
         final int typedLength = STRING_TO_TYPE.length();
         final int EXPECTED_SUGGESTION_SPAN_START = -1;
         final int EXPECTED_UNDERLINE_SPAN_START = 0;
-        final int EXPECTED_UNDERLINE_SPAN_END = 4;
+        final int EXPECTED_UNDERLINE_SPAN_END = 3;
         type(STRING_TO_TYPE);
         sleep(DELAY_TO_WAIT_FOR_UNDERLINE);
         runMessages();
diff --git a/tests/src/com/android/inputmethod/latin/InputLogicTests.java b/tests/src/com/android/inputmethod/latin/InputLogicTests.java
index 59b858d..ec249da 100644
--- a/tests/src/com/android/inputmethod/latin/InputLogicTests.java
+++ b/tests/src/com/android/inputmethod/latin/InputLogicTests.java
@@ -159,8 +159,11 @@
     }
 
     public void testAutoCorrectWithSpaceThenRevert() {
+        // Backspacing to cancel the "tgis"->"this" autocorrection should result in
+        // a "phantom space": if the user presses space immediately after,
+        // only one space will be inserted in total.
         final String STRING_TO_TYPE = "tgis ";
-        final String EXPECTED_RESULT = "tgis ";
+        final String EXPECTED_RESULT = "tgis";
         type(STRING_TO_TYPE);
         mLatinIME.onUpdateSelection(0, 0, STRING_TO_TYPE.length(), STRING_TO_TYPE.length(), -1, -1);
         type(Constants.CODE_DELETE);
@@ -168,6 +171,24 @@
                 mEditText.getText().toString());
     }
 
+    public void testAutoCorrectWithSpaceThenRevertThenTypeMore() {
+        final String STRING_TO_TYPE_FIRST = "tgis ";
+        final String STRING_TO_TYPE_SECOND = "a";
+        final String EXPECTED_RESULT = "tgis a";
+        type(STRING_TO_TYPE_FIRST);
+        mLatinIME.onUpdateSelection(0, 0,
+                STRING_TO_TYPE_FIRST.length(), STRING_TO_TYPE_FIRST.length(), -1, -1);
+        type(Constants.CODE_DELETE);
+
+        type(STRING_TO_TYPE_SECOND);
+        mLatinIME.onUpdateSelection(STRING_TO_TYPE_FIRST.length(), STRING_TO_TYPE_FIRST.length(),
+                STRING_TO_TYPE_FIRST.length() - 1 + STRING_TO_TYPE_SECOND.length(),
+                STRING_TO_TYPE_FIRST.length() - 1 + STRING_TO_TYPE_SECOND.length(),
+                -1, -1);
+        assertEquals("auto-correct with space then revert then type more", EXPECTED_RESULT,
+                mEditText.getText().toString());
+    }
+
     public void testAutoCorrectToSelfDoesNotRevert() {
         final String STRING_TO_TYPE = "this ";
         final String EXPECTED_RESULT = "this";
diff --git a/tests/src/com/android/inputmethod/latin/NgramContextTests.java b/tests/src/com/android/inputmethod/latin/NgramContextTests.java
new file mode 100644
index 0000000..ecc2c63
--- /dev/null
+++ b/tests/src/com/android/inputmethod/latin/NgramContextTests.java
@@ -0,0 +1,66 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.inputmethod.latin;
+
+import com.android.inputmethod.latin.NgramContext.WordInfo;
+
+import android.test.AndroidTestCase;
+import android.test.suitebuilder.annotation.SmallTest;
+
+@SmallTest
+public class NgramContextTests extends AndroidTestCase {
+    public void testConstruct() {
+        assertEquals(new NgramContext(new WordInfo("a")), new NgramContext(new WordInfo("a")));
+        assertEquals(new NgramContext(WordInfo.BEGINNING_OF_SENTENCE),
+                new NgramContext(WordInfo.BEGINNING_OF_SENTENCE));
+        assertEquals(new NgramContext(WordInfo.EMPTY_WORD_INFO),
+                new NgramContext(WordInfo.EMPTY_WORD_INFO));
+        assertEquals(new NgramContext(WordInfo.EMPTY_WORD_INFO),
+                new NgramContext(WordInfo.EMPTY_WORD_INFO));
+    }
+
+    public void testIsBeginningOfSentenceContext() {
+        assertFalse(new NgramContext().isBeginningOfSentenceContext());
+        assertTrue(new NgramContext(WordInfo.BEGINNING_OF_SENTENCE)
+                .isBeginningOfSentenceContext());
+        assertTrue(NgramContext.BEGINNING_OF_SENTENCE.isBeginningOfSentenceContext());
+        assertFalse(new NgramContext(new WordInfo("a")).isBeginningOfSentenceContext());
+        assertFalse(new NgramContext(new WordInfo("")).isBeginningOfSentenceContext());
+        assertFalse(new NgramContext(WordInfo.EMPTY_WORD_INFO).isBeginningOfSentenceContext());
+        assertTrue(new NgramContext(WordInfo.BEGINNING_OF_SENTENCE, new WordInfo("a"))
+                .isBeginningOfSentenceContext());
+        assertFalse(new NgramContext(new WordInfo("a"), WordInfo.BEGINNING_OF_SENTENCE)
+                .isBeginningOfSentenceContext());
+        assertFalse(new NgramContext(WordInfo.EMPTY_WORD_INFO, WordInfo.BEGINNING_OF_SENTENCE)
+                .isBeginningOfSentenceContext());
+    }
+
+    public void testGetNextNgramContext() {
+        final NgramContext ngramContext_a = new NgramContext(new WordInfo("a"));
+        final NgramContext ngramContext_b_a =
+                ngramContext_a.getNextNgramContext(new WordInfo("b"));
+        assertEquals("b", ngramContext_b_a.getNthPrevWord(1));
+        assertEquals("a", ngramContext_b_a.getNthPrevWord(2));
+        final NgramContext ngramContext_bos_b =
+                ngramContext_b_a.getNextNgramContext(WordInfo.BEGINNING_OF_SENTENCE);
+        assertTrue(ngramContext_bos_b.isBeginningOfSentenceContext());
+        assertEquals("b", ngramContext_bos_b.getNthPrevWord(2));
+        final NgramContext ngramContext_c_bos =
+                ngramContext_b_a.getNextNgramContext(new WordInfo("c"));
+        assertEquals("c", ngramContext_c_bos.getNthPrevWord(1));
+    }
+}
diff --git a/tests/src/com/android/inputmethod/latin/RichInputConnectionAndTextRangeTests.java b/tests/src/com/android/inputmethod/latin/RichInputConnectionAndTextRangeTests.java
index f9d7226..7a32336 100644
--- a/tests/src/com/android/inputmethod/latin/RichInputConnectionAndTextRangeTests.java
+++ b/tests/src/com/android/inputmethod/latin/RichInputConnectionAndTextRangeTests.java
@@ -30,9 +30,8 @@
 import android.view.inputmethod.InputConnection;
 import android.view.inputmethod.InputConnectionWrapper;
 
-import com.android.inputmethod.latin.PrevWordsInfo.WordInfo;
 import com.android.inputmethod.latin.settings.SpacingAndPunctuations;
-import com.android.inputmethod.latin.utils.PrevWordsInfoUtils;
+import com.android.inputmethod.latin.utils.NgramContextUtils;
 import com.android.inputmethod.latin.utils.RunInLocale;
 import com.android.inputmethod.latin.utils.ScriptUtils;
 import com.android.inputmethod.latin.utils.StringUtils;
@@ -158,26 +157,25 @@
      */
     public void testGetPreviousWord() {
         // If one of the following cases breaks, the bigram suggestions won't work.
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def", mSpacingAndPunctuations, 2).mPrevWordsInfo[0].mWord, "abc");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc", mSpacingAndPunctuations, 2), PrevWordsInfo.BEGINNING_OF_SENTENCE);
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc. def", mSpacingAndPunctuations, 2), PrevWordsInfo.BEGINNING_OF_SENTENCE);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def", mSpacingAndPunctuations, 2).getNthPrevWord(1), "abc");
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc", mSpacingAndPunctuations, 2), NgramContext.BEGINNING_OF_SENTENCE);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc. def", mSpacingAndPunctuations, 2), NgramContext.BEGINNING_OF_SENTENCE);
 
-        assertFalse(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def", mSpacingAndPunctuations, 2).mPrevWordsInfo[0].mIsBeginningOfSentence);
-        assertTrue(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc", mSpacingAndPunctuations, 2).mPrevWordsInfo[0].mIsBeginningOfSentence);
+        assertFalse(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def", mSpacingAndPunctuations, 2).isBeginningOfSentenceContext());
+        assertTrue(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc", mSpacingAndPunctuations, 2).isBeginningOfSentenceContext());
 
         // For n-gram
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def", mSpacingAndPunctuations, 1).mPrevWordsInfo[0].mWord, "def");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def", mSpacingAndPunctuations, 1).mPrevWordsInfo[1].mWord, "abc");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def", mSpacingAndPunctuations, 2).mPrevWordsInfo[1],
-                WordInfo.BEGINNING_OF_SENTENCE);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def", mSpacingAndPunctuations, 1).getNthPrevWord(1), "def");
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def", mSpacingAndPunctuations, 1).getNthPrevWord(2), "abc");
+        assertTrue(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def", mSpacingAndPunctuations, 2).isNthPrevWordBeginningOfSontence(2));
 
         // The following tests reflect the current behavior of the function
         // RichInputConnection#getNthPreviousWord.
@@ -186,33 +184,33 @@
         // this function if needed - especially since it does not seem very
         // logical. These tests are just there to catch any unintentional
         // changes in the behavior of the RichInputConnection#getPreviousWord method.
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def ", mSpacingAndPunctuations, 2).mPrevWordsInfo[0].mWord, "abc");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def.", mSpacingAndPunctuations, 2).mPrevWordsInfo[0].mWord, "abc");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def .", mSpacingAndPunctuations, 2).mPrevWordsInfo[0].mWord, "def");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc ", mSpacingAndPunctuations, 2), PrevWordsInfo.BEGINNING_OF_SENTENCE);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def ", mSpacingAndPunctuations, 2).getNthPrevWord(1), "abc");
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def.", mSpacingAndPunctuations, 2).getNthPrevWord(1), "abc");
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def .", mSpacingAndPunctuations, 2).getNthPrevWord(1), "def");
+        assertTrue(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc ", mSpacingAndPunctuations, 2).isBeginningOfSentenceContext());
 
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def", mSpacingAndPunctuations, 1).mPrevWordsInfo[0].mWord, "def");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def ", mSpacingAndPunctuations, 1).mPrevWordsInfo[0].mWord, "def");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc 'def", mSpacingAndPunctuations, 1).mPrevWordsInfo[0].mWord, "'def");
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def.", mSpacingAndPunctuations, 1), PrevWordsInfo.BEGINNING_OF_SENTENCE);
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc def .", mSpacingAndPunctuations, 1), PrevWordsInfo.BEGINNING_OF_SENTENCE);
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc, def", mSpacingAndPunctuations, 2), PrevWordsInfo.EMPTY_PREV_WORDS_INFO);
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc? def", mSpacingAndPunctuations, 2), PrevWordsInfo.EMPTY_PREV_WORDS_INFO);
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc! def", mSpacingAndPunctuations, 2), PrevWordsInfo.EMPTY_PREV_WORDS_INFO);
-        assertEquals(PrevWordsInfoUtils.getPrevWordsInfoFromNthPreviousWord(
-                "abc 'def", mSpacingAndPunctuations, 2), PrevWordsInfo.EMPTY_PREV_WORDS_INFO);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def", mSpacingAndPunctuations, 1).getNthPrevWord(1), "def");
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def ", mSpacingAndPunctuations, 1).getNthPrevWord(1), "def");
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc 'def", mSpacingAndPunctuations, 1).getNthPrevWord(1), "'def");
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def.", mSpacingAndPunctuations, 1), NgramContext.BEGINNING_OF_SENTENCE);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc def .", mSpacingAndPunctuations, 1), NgramContext.BEGINNING_OF_SENTENCE);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc, def", mSpacingAndPunctuations, 2), NgramContext.EMPTY_PREV_WORDS_INFO);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc? def", mSpacingAndPunctuations, 2), NgramContext.EMPTY_PREV_WORDS_INFO);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc! def", mSpacingAndPunctuations, 2), NgramContext.EMPTY_PREV_WORDS_INFO);
+        assertEquals(NgramContextUtils.getNgramContextFromNthPreviousWord(
+                "abc 'def", mSpacingAndPunctuations, 2), NgramContext.EMPTY_PREV_WORDS_INFO);
     }
 
     public void testGetWordRangeAtCursor() {
diff --git a/tests/src/com/android/inputmethod/latin/makedict/BinaryDictEncoderUtils.java b/tests/src/com/android/inputmethod/latin/makedict/BinaryDictEncoderUtils.java
index 2b3fd89..12290e6 100644
--- a/tests/src/com/android/inputmethod/latin/makedict/BinaryDictEncoderUtils.java
+++ b/tests/src/com/android/inputmethod/latin/makedict/BinaryDictEncoderUtils.java
@@ -27,6 +27,8 @@
 import java.io.IOException;
 import java.io.OutputStream;
 import java.util.ArrayList;
+import java.util.HashMap;
+import java.util.Map.Entry;
 
 /**
  * Encodes binary files for a FusionDictionary.
@@ -791,10 +793,12 @@
      * @param destination the stream to write the file header to.
      * @param dict the dictionary to write.
      * @param formatOptions file format options.
+     * @param codePointOccurrenceArray code points ordered by occurrence count.
      * @return the size of the header.
      */
     /* package */ static int writeDictionaryHeader(final OutputStream destination,
-            final FusionDictionary dict, final FormatOptions formatOptions)
+            final FusionDictionary dict, final FormatOptions formatOptions,
+            final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray)
                     throws IOException, UnsupportedFormatException {
         final int version = formatOptions.mVersion;
         if (version < FormatSpec.MINIMUM_SUPPORTED_VERSION
@@ -833,6 +837,9 @@
             CharEncoding.writeString(headerBuffer, key);
             CharEncoding.writeString(headerBuffer, value);
         }
+
+        // TODO: Write out the code point table.
+
         final int size = headerBuffer.size();
         final byte[] bytes = headerBuffer.toByteArray();
         // Write out the header size.
@@ -845,4 +852,15 @@
         headerBuffer.close();
         return size;
     }
+
+    static final class CodePointTable {
+        final HashMap<Integer, Integer> mCodePointToOneByteCodeMap;
+        final ArrayList<Entry<Integer, Integer>> mCodePointOccurrenceArray;
+
+        CodePointTable(final HashMap<Integer, Integer> codePointToOneByteCodeMap,
+                final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray) {
+            mCodePointToOneByteCodeMap = codePointToOneByteCodeMap;
+            mCodePointOccurrenceArray = codePointOccurrenceArray;
+        }
+    }
 }
diff --git a/tests/src/com/android/inputmethod/latin/makedict/Ver2DictEncoder.java b/tests/src/com/android/inputmethod/latin/makedict/Ver2DictEncoder.java
index 0fa75e8..c471901 100644
--- a/tests/src/com/android/inputmethod/latin/makedict/Ver2DictEncoder.java
+++ b/tests/src/com/android/inputmethod/latin/makedict/Ver2DictEncoder.java
@@ -18,6 +18,7 @@
 
 import com.android.inputmethod.annotations.UsedForTesting;
 import com.android.inputmethod.latin.makedict.BinaryDictDecoderUtils.CharEncoding;
+import com.android.inputmethod.latin.makedict.BinaryDictEncoderUtils.CodePointTable;
 import com.android.inputmethod.latin.makedict.FormatSpec.FormatOptions;
 import com.android.inputmethod.latin.makedict.FusionDictionary.PtNode;
 import com.android.inputmethod.latin.makedict.FusionDictionary.PtNodeArray;
@@ -28,7 +29,11 @@
 import java.io.IOException;
 import java.io.OutputStream;
 import java.util.ArrayList;
+import java.util.Collections;
+import java.util.Comparator;
+import java.util.HashMap;
 import java.util.Iterator;
+import java.util.Map.Entry;
 
 /**
  * An implementation of DictEncoder for version 2 binary dictionary.
@@ -73,6 +78,49 @@
         }
     }
 
+    // Package for testing
+    static CodePointTable makeCodePointTable(final FusionDictionary dict) {
+        final HashMap<Integer, Integer> codePointOccurrenceCounts = new HashMap<>();
+        for (final WordProperty word : dict) {
+            // Store per code point occurrence
+            final String wordString = word.mWord;
+            for (int i = 0; i < wordString.length(); ++i) {
+                final int codePoint = Character.codePointAt(wordString, i);
+                if (codePointOccurrenceCounts.containsKey(codePoint)) {
+                    codePointOccurrenceCounts.put(codePoint,
+                            codePointOccurrenceCounts.get(codePoint) + 1);
+                } else {
+                    codePointOccurrenceCounts.put(codePoint, 1);
+                }
+            }
+        }
+        final ArrayList<Entry<Integer, Integer>> codePointOccurrenceArray =
+                new ArrayList<>(codePointOccurrenceCounts.entrySet());
+        // Descending order sort by occurrence (value side)
+        Collections.sort(codePointOccurrenceArray, new Comparator<Entry<Integer, Integer>>() {
+            @Override
+            public int compare(final Entry<Integer, Integer> a, final Entry<Integer, Integer> b) {
+                if (a.getValue() != b.getValue()) {
+                    return b.getValue().compareTo(a.getValue());
+                }
+                return b.getKey().compareTo(a.getKey());
+            }
+        });
+        int currentCodePointTableIndex = FormatSpec.MINIMAL_ONE_BYTE_CHARACTER_VALUE;
+        // Temporary map for writing of nodes
+        final HashMap<Integer, Integer> codePointToOneByteCodeMap = new HashMap<>();
+        for (final Entry<Integer, Integer> entry : codePointOccurrenceArray) {
+            // Put a relation from the original code point to the one byte code.
+            codePointToOneByteCodeMap.put(entry.getKey(), currentCodePointTableIndex);
+            if (FormatSpec.MAXIMAL_ONE_BYTE_CHARACTER_VALUE < ++currentCodePointTableIndex) {
+                break;
+            }
+        }
+        // codePointToOneByteCodeMap for writing the trie
+        // codePointOccurrenceArray for writing the header
+        return new CodePointTable(codePointToOneByteCodeMap, codePointOccurrenceArray);
+    }
+
     @Override
     public void writeDictionary(final FusionDictionary dict, final FormatOptions formatOptions)
             throws IOException, UnsupportedFormatException {
@@ -85,7 +133,12 @@
         if (mOutStream == null) {
             openStream();
         }
-        BinaryDictEncoderUtils.writeDictionaryHeader(mOutStream, dict, formatOptions);
+
+        // Make code point conversion table ordered by occurrence of code points
+        final CodePointTable codePointTable = makeCodePointTable(dict);
+
+        BinaryDictEncoderUtils.writeDictionaryHeader(mOutStream, dict, formatOptions,
+                codePointTable.mCodePointOccurrenceArray);
 
         // Addresses are limited to 3 bytes, but since addresses can be relative to each node
         // array, the structure itself is not limited to 16MB. However, if it is over 16MB deciding
diff --git a/tests/src/com/android/inputmethod/latin/makedict/Ver2DictEncoderTests.java b/tests/src/com/android/inputmethod/latin/makedict/Ver2DictEncoderTests.java
new file mode 100644
index 0000000..9104c2f
--- /dev/null
+++ b/tests/src/com/android/inputmethod/latin/makedict/Ver2DictEncoderTests.java
@@ -0,0 +1,91 @@
+/*
+ * Copyright (C) 2014 The Android Open Source Project
+ *
+ * Licensed under the Apache License, Version 2.0 (the "License");
+ * you may not use this file except in compliance with the License.
+ * You may obtain a copy of the License at
+ *
+ *      http://www.apache.org/licenses/LICENSE-2.0
+ *
+ * Unless required by applicable law or agreed to in writing, software
+ * distributed under the License is distributed on an "AS IS" BASIS,
+ * WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
+ * See the License for the specific language governing permissions and
+ * limitations under the License.
+ */
+
+package com.android.inputmethod.latin.makedict;
+
+import java.util.ArrayList;
+import java.util.Arrays;
+import java.util.HashMap;
+import java.util.List;
+import java.util.Map.Entry;
+
+import com.android.inputmethod.latin.makedict.BinaryDictEncoderUtils.CodePointTable;
+import com.android.inputmethod.latin.makedict.FusionDictionary.PtNodeArray;
+
+import android.test.AndroidTestCase;
+import android.test.suitebuilder.annotation.LargeTest;
+import android.util.Log;
+
+/**
+ * Unit tests for Ver2DictEncoder
+ */
+@LargeTest
+public class Ver2DictEncoderTests extends AndroidTestCase {
+    private static final String TAG = Ver2DictEncoderTests.class.getSimpleName();
+    private static final int UNIGRAM_FREQ = 10;
+
+    public void testCodePointTable() {
+        final String[] wordSource = {"words", "used", "for", "testing", "a", "code point", "table"};
+        final List<String> words = Arrays.asList(wordSource);
+        final String correctCodePointTable = "eotdsanirfg bclwup";
+        final String correctCodePointOccurrenceArrayString =
+                "11641114101411531003114211021052972119111711121108110311021991981321";
+        final String correctCodePointExpectedMapString = "343332363540383937464549484744414243";
+        final String dictName = "codePointTableTest";
+        final String dictVersion = Long.toString(System.currentTimeMillis());
+
+        final FormatSpec.FormatOptions formatOptions =
+                new FormatSpec.FormatOptions(FormatSpec.VERSION2);
+        final FusionDictionary sourcedict = new FusionDictionary(new PtNodeArray(),
+                BinaryDictUtils.makeDictionaryOptions(dictName, dictVersion, formatOptions));
+        addUnigrams(sourcedict, words, null /* shortcutMap */);
+        final CodePointTable codePointTable = Ver2DictEncoder.makeCodePointTable(sourcedict);
+
+        // Check if mCodePointOccurrenceArray is correct
+        final StringBuilder codePointOccurrenceArrayString = new StringBuilder();
+        for (Entry<Integer, Integer> entry : codePointTable.mCodePointOccurrenceArray) {
+            codePointOccurrenceArrayString.append(entry.getKey());
+            codePointOccurrenceArrayString.append(entry.getValue());
+        }
+        assertEquals(correctCodePointOccurrenceArrayString,
+                codePointOccurrenceArrayString.toString());
+
+        // Check if mCodePointToOneByteCodeMap is correct
+        final StringBuilder codePointExpectedMapString = new StringBuilder();
+        for (int i = 0; i < correctCodePointTable.length(); ++i) {
+            codePointExpectedMapString.append(codePointTable.mCodePointToOneByteCodeMap.get(
+                    correctCodePointTable.codePointAt(i)));
+        }
+        assertEquals(correctCodePointExpectedMapString, codePointExpectedMapString.toString());
+    }
+
+    /**
+     * Adds unigrams to the dictionary.
+     */
+    private void addUnigrams(final FusionDictionary dict, final List<String> words,
+            final HashMap<String, List<String>> shortcutMap) {
+        for (final String word : words) {
+            final ArrayList<WeightedString> shortcuts = new ArrayList<>();
+            if (shortcutMap != null && shortcutMap.containsKey(word)) {
+                for (final String shortcut : shortcutMap.get(word)) {
+                    shortcuts.add(new WeightedString(shortcut, UNIGRAM_FREQ));
+                }
+            }
+            dict.add(word, new ProbabilityInfo(UNIGRAM_FREQ),
+                    (shortcutMap == null) ? null : shortcuts, false /* isNotAWord */);
+        }
+    }
+}
diff --git a/tests/src/com/android/inputmethod/latin/makedict/Ver4DictEncoder.java b/tests/src/com/android/inputmethod/latin/makedict/Ver4DictEncoder.java
index 2e54351..74da937 100644
--- a/tests/src/com/android/inputmethod/latin/makedict/Ver4DictEncoder.java
+++ b/tests/src/com/android/inputmethod/latin/makedict/Ver4DictEncoder.java
@@ -19,7 +19,7 @@
 import com.android.inputmethod.annotations.UsedForTesting;
 import com.android.inputmethod.latin.BinaryDictionary;
 import com.android.inputmethod.latin.Dictionary;
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.makedict.FormatSpec.FormatOptions;
 import com.android.inputmethod.latin.makedict.FusionDictionary.PtNode;
 import com.android.inputmethod.latin.utils.BinaryDictionaryUtils;
@@ -104,12 +104,12 @@
         for (final WordProperty word0Property : dict) {
             if (null == word0Property.mBigrams) continue;
             for (final WeightedString word1 : word0Property.mBigrams) {
-                final PrevWordsInfo prevWordsInfo =
-                        new PrevWordsInfo(new PrevWordsInfo.WordInfo(word0Property.mWord));
-                if (!binaryDict.addNgramEntry(prevWordsInfo, word1.mWord,
+                final NgramContext ngramContext =
+                        new NgramContext(new NgramContext.WordInfo(word0Property.mWord));
+                if (!binaryDict.addNgramEntry(ngramContext, word1.mWord,
                         word1.getProbability(), 0 /* timestamp */)) {
                     MakedictLog.e("Cannot add n-gram entry for "
-                            + prevWordsInfo + " -> " + word1.mWord);
+                            + ngramContext + " -> " + word1.mWord);
                     return;
                 }
                 if (binaryDict.needsToRunGC(true /* mindsBlockByGC */)) {
diff --git a/tests/src/com/android/inputmethod/latin/personalization/UserHistoryDictionaryTests.java b/tests/src/com/android/inputmethod/latin/personalization/UserHistoryDictionaryTests.java
index 6162096..7666273 100644
--- a/tests/src/com/android/inputmethod/latin/personalization/UserHistoryDictionaryTests.java
+++ b/tests/src/com/android/inputmethod/latin/personalization/UserHistoryDictionaryTests.java
@@ -21,8 +21,8 @@
 import android.util.Log;
 
 import com.android.inputmethod.latin.ExpandableBinaryDictionary;
-import com.android.inputmethod.latin.PrevWordsInfo;
-import com.android.inputmethod.latin.PrevWordsInfo.WordInfo;
+import com.android.inputmethod.latin.NgramContext;
+import com.android.inputmethod.latin.NgramContext.WordInfo;
 import com.android.inputmethod.latin.utils.BinaryDictionaryUtils;
 import com.android.inputmethod.latin.utils.DistracterFilter;
 import com.android.inputmethod.latin.utils.FileUtils;
@@ -159,13 +159,13 @@
         return new ArrayList<>(wordSet);
     }
 
-    private static void addToDict(final UserHistoryDictionary dict, final List<String> words) {
-        PrevWordsInfo prevWordsInfo = PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+    private static void addToDict(final UserHistoryDictionary dict, final List<String> words,
+            final int timestamp) {
+        NgramContext ngramContext = NgramContext.EMPTY_PREV_WORDS_INFO;
         for (String word : words) {
-            UserHistoryDictionary.addToDictionary(dict, prevWordsInfo, word, true,
-                    (int)TimeUnit.MILLISECONDS.toSeconds(System.currentTimeMillis()),
+            UserHistoryDictionary.addToDictionary(dict, ngramContext, word, true, timestamp,
                     DistracterFilter.EMPTY_DISTRACTER_FILTER);
-            prevWordsInfo = prevWordsInfo.getNextPrevWordsInfo(new WordInfo(word));
+            ngramContext = ngramContext.getNextNgramContext(new WordInfo(word));
         }
     }
 
@@ -177,7 +177,7 @@
             final int numberOfWords, final Random random, final boolean checkContents) {
         final List<String> words = generateWords(numberOfWords, random);
         // Add random words to the user history dictionary.
-        addToDict(dict, words);
+        addToDict(dict, words, mCurrentTime);
         if (checkContents) {
             dict.waitAllTasksForTests();
             for (int i = 0; i < numberOfWords; ++i) {
@@ -288,11 +288,11 @@
         clearHistory(dict);
         final List<String> words = generateWords(numberOfWords, random);
         dict.waitAllTasksForTests();
-        PrevWordsInfo prevWordsInfo = PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+        NgramContext ngramContext = NgramContext.EMPTY_PREV_WORDS_INFO;
         for (final String word : words) {
-            UserHistoryDictionary.addToDictionary(dict, prevWordsInfo, word, true, mCurrentTime,
+            UserHistoryDictionary.addToDictionary(dict, ngramContext, word, true, mCurrentTime,
                     DistracterFilter.EMPTY_DISTRACTER_FILTER);
-            prevWordsInfo = prevWordsInfo.getNextPrevWordsInfo(new WordInfo(word));
+            ngramContext = ngramContext.getNextNgramContext(new WordInfo(word));
             dict.waitAllTasksForTests();
             assertTrue(dict.isInDictionary(word));
         }
@@ -308,6 +308,5 @@
         for (final String word : words) {
             assertFalse(dict.isInDictionary(word));
         }
-        stopTestModeInNativeCode();
     }
 }
diff --git a/tests/src/com/android/inputmethod/latin/utils/DistracterFilterTest.java b/tests/src/com/android/inputmethod/latin/utils/DistracterFilterTest.java
index 6ed9120..8360d53 100644
--- a/tests/src/com/android/inputmethod/latin/utils/DistracterFilterTest.java
+++ b/tests/src/com/android/inputmethod/latin/utils/DistracterFilterTest.java
@@ -24,7 +24,7 @@
 import android.test.suitebuilder.annotation.LargeTest;
 import android.view.inputmethod.InputMethodSubtype;
 
-import com.android.inputmethod.latin.PrevWordsInfo;
+import com.android.inputmethod.latin.NgramContext;
 import com.android.inputmethod.latin.RichInputMethodManager;
 import com.android.inputmethod.latin.utils.DistracterFilter.HandlingType;
 
@@ -58,7 +58,7 @@
     }
 
     public void testIsDistracterToWordsInDictionaries() {
-        final PrevWordsInfo EMPTY_PREV_WORDS_INFO = PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+        final NgramContext EMPTY_PREV_WORDS_INFO = NgramContext.EMPTY_PREV_WORDS_INFO;
 
         final Locale localeEnUs = new Locale("en", "US");
         String typedWord;
@@ -204,7 +204,7 @@
 
     public void testGetWordHandlingType() {
         final Locale localeEnUs = new Locale("en", "US");
-        final PrevWordsInfo EMPTY_PREV_WORDS_INFO = PrevWordsInfo.EMPTY_PREV_WORDS_INFO;
+        final NgramContext EMPTY_PREV_WORDS_INFO = NgramContext.EMPTY_PREV_WORDS_INFO;
         int handlingType = 0;
 
         handlingType = mDistracterFilter.getWordHandlingType(EMPTY_PREV_WORDS_INFO,
diff --git a/tools/dicttool/Android.mk b/tools/dicttool/Android.mk
index 3e3d419..7f34ccf 100644
--- a/tools/dicttool/Android.mk
+++ b/tools/dicttool/Android.mk
@@ -50,7 +50,7 @@
         latin/Dictionary.java \
         latin/InputPointers.java \
         latin/LastComposedWord.java \
-        latin/PrevWordsInfo.java \
+        latin/NgramContext.java \
         latin/SuggestedWords.java \
         latin/WordComposer.java \
         latin/settings/NativeSuggestOptions.java \