diff --git a/native/jni/src/defines.h b/native/jni/src/defines.h
index a80c975..24d04e5 100644
--- a/native/jni/src/defines.h
+++ b/native/jni/src/defines.h
@@ -336,6 +336,9 @@
 #define MAX_POINTER_COUNT 1
 #define MAX_POINTER_COUNT_G 2
 
+// (MAX_PREV_WORD_COUNT_FOR_N_GRAM + 1)-gram is supported.
+#define MAX_PREV_WORD_COUNT_FOR_N_GRAM 1
+
 #define DISALLOW_DEFAULT_CONSTRUCTOR(TypeName) \
   TypeName() = delete
 
diff --git a/native/jni/src/suggest/core/session/dic_traverse_session.cpp b/native/jni/src/suggest/core/session/dic_traverse_session.cpp
index 7406fbe..b9e9db7 100644
--- a/native/jni/src/suggest/core/session/dic_traverse_session.cpp
+++ b/native/jni/src/suggest/core/session/dic_traverse_session.cpp
@@ -36,17 +36,17 @@
             ->getMultiWordCostMultiplier();
     mSuggestOptions = suggestOptions;
     if (!prevWordsInfo->getPrevWordCodePoints()) {
-        mPrevWordPtNodePos = NOT_A_DICT_POS;
+        mPrevWordsPtNodePos[0] = NOT_A_DICT_POS;
         return;
     }
     // TODO: merge following similar calls to getTerminalPosition into one case-insensitive call.
-    mPrevWordPtNodePos = getDictionaryStructurePolicy()->getTerminalPtNodePositionOfWord(
+    mPrevWordsPtNodePos[0] = getDictionaryStructurePolicy()->getTerminalPtNodePositionOfWord(
             prevWordsInfo->getPrevWordCodePoints(), prevWordsInfo->getPrevWordCodePointCount(),
             false /* forceLowerCaseSearch */);
-    if (mPrevWordPtNodePos == NOT_A_DICT_POS) {
+    if (mPrevWordsPtNodePos[0] == NOT_A_DICT_POS) {
         // Check bigrams for lower-cased previous word if original was not found. Useful for
         // auto-capitalized words like "The [current_word]".
-        mPrevWordPtNodePos = getDictionaryStructurePolicy()->getTerminalPtNodePositionOfWord(
+        mPrevWordsPtNodePos[0] = getDictionaryStructurePolicy()->getTerminalPtNodePositionOfWord(
                 prevWordsInfo->getPrevWordCodePoints(), prevWordsInfo->getPrevWordCodePointCount(),
                 true /* forceLowerCaseSearch */);
     }
diff --git a/native/jni/src/suggest/core/session/dic_traverse_session.h b/native/jni/src/suggest/core/session/dic_traverse_session.h
index 50c60d0..90aff06 100644
--- a/native/jni/src/suggest/core/session/dic_traverse_session.h
+++ b/native/jni/src/suggest/core/session/dic_traverse_session.h
@@ -50,12 +50,14 @@
     }
 
     AK_FORCE_INLINE DicTraverseSession(JNIEnv *env, jstring localeStr, bool usesLargeCache)
-            : mPrevWordPtNodePos(NOT_A_DICT_POS), mProximityInfo(nullptr),
-              mDictionary(nullptr), mSuggestOptions(nullptr), mDicNodesCache(usesLargeCache),
-              mMultiBigramMap(), mInputSize(0), mMaxPointerCount(1),
+            : mProximityInfo(nullptr), mDictionary(nullptr), mSuggestOptions(nullptr),
+              mDicNodesCache(usesLargeCache), mMultiBigramMap(), mInputSize(0), mMaxPointerCount(1),
               mMultiWordCostMultiplier(1.0f) {
         // NOTE: mProximityInfoStates is an array of instances.
         // No need to initialize it explicitly here.
+        for (size_t i = 0; i < NELEMS(mPrevWordsPtNodePos); ++i) {
+            mPrevWordsPtNodePos[i] = NOT_A_DICT_POS;
+        }
     }
 
     // Non virtual inline destructor -- never inherit this class
@@ -77,7 +79,7 @@
     //--------------------
     const ProximityInfo *getProximityInfo() const { return mProximityInfo; }
     const SuggestOptions *getSuggestOptions() const { return mSuggestOptions; }
-    int getPrevWordPtNodePos() const { return mPrevWordPtNodePos; }
+    int getPrevWordPtNodePos() const { return mPrevWordsPtNodePos[0]; }
     DicNodesCache *getDicTraverseCache() { return &mDicNodesCache; }
     MultiBigramMap *getMultiBigramMap() { return &mMultiBigramMap; }
     const ProximityInfoState *getProximityInfoState(int id) const {
@@ -164,7 +166,7 @@
             const int *const inputYs, const int *const times, const int *const pointerIds,
             const int inputSize, const float maxSpatialDistance, const int maxPointerCount);
 
-    int mPrevWordPtNodePos;
+    int mPrevWordsPtNodePos[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
     const ProximityInfo *mProximityInfo;
     const Dictionary *mDictionary;
     const SuggestOptions *mSuggestOptions;
diff --git a/native/jni/src/suggest/core/session/prev_words_info.h b/native/jni/src/suggest/core/session/prev_words_info.h
index df7e7e3..bc68594 100644
--- a/native/jni/src/suggest/core/session/prev_words_info.h
+++ b/native/jni/src/suggest/core/session/prev_words_info.h
@@ -27,27 +27,39 @@
 class PrevWordsInfo {
  public:
     // No prev word information.
-    PrevWordsInfo()
-            : mPrevWordCodePoints(nullptr), mPrevWordCodePointCount(0) {}
+    PrevWordsInfo() {
+        clear();
+    }
 
     PrevWordsInfo(const int *const prevWordCodePoints, const int prevWordCodePointCount,
-            const bool isBeginningOfSentence)
-            : mPrevWordCodePoints(prevWordCodePoints),
-              mPrevWordCodePointCount(prevWordCodePointCount) {}
-
+            const bool isBeginningOfSentence) {
+        clear();
+        mPrevWordCodePoints[0] = prevWordCodePoints;
+        mPrevWordCodePointCount[0] = prevWordCodePointCount;
+        mIsBeginningOfSentence[0] = isBeginningOfSentence;
+    }
     const int *getPrevWordCodePoints() const {
-        return mPrevWordCodePoints;
+        return mPrevWordCodePoints[0];
     }
 
     int getPrevWordCodePointCount() const {
-        return mPrevWordCodePointCount;
+        return mPrevWordCodePointCount[0];
     }
 
  private:
     DISALLOW_COPY_AND_ASSIGN(PrevWordsInfo);
 
-    const int *const mPrevWordCodePoints;
-    const int mPrevWordCodePointCount;
+    void clear() {
+        for (size_t i = 0; i < NELEMS(mPrevWordCodePoints); ++i) {
+            mPrevWordCodePoints[i] = nullptr;
+            mPrevWordCodePointCount[i] = 0;
+            mIsBeginningOfSentence[i] = false;
+        }
+    }
+
+    const int *mPrevWordCodePoints[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
+    int mPrevWordCodePointCount[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
+    bool mIsBeginningOfSentence[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
 };
 } // namespace latinime
 #endif // LATINIME_PREV_WORDS_INFO_H