diff --git a/native/jni/src/suggest/core/dictionary/dictionary.cpp b/native/jni/src/suggest/core/dictionary/dictionary.cpp
index ec261cf..b843791 100644
--- a/native/jni/src/suggest/core/dictionary/dictionary.cpp
+++ b/native/jni/src/suggest/core/dictionary/dictionary.cpp
@@ -93,13 +93,13 @@
 void Dictionary::getPredictions(const PrevWordsInfo *const prevWordsInfo,
         SuggestionResults *const outSuggestionResults) const {
     TimeKeeper::setCurrentTime();
-    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIds;
-    prevWordsInfo->getPrevWordIds(mDictionaryStructureWithBufferPolicy.get(), prevWordIds.data(),
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
+    const WordIdArrayView prevWordIds = prevWordsInfo->getPrevWordIds(
+            mDictionaryStructureWithBufferPolicy.get(), &prevWordIdArray,
             true /* tryLowerCaseSearch */);
-    const WordIdArrayView prevWordIdArrayView = WordIdArrayView::fromArray(prevWordIds);
-    NgramListenerForPrediction listener(prevWordsInfo, prevWordIdArrayView, outSuggestionResults,
+    NgramListenerForPrediction listener(prevWordsInfo, prevWordIds, outSuggestionResults,
             mDictionaryStructureWithBufferPolicy.get());
-    mDictionaryStructureWithBufferPolicy->iterateNgramEntries(prevWordIdArrayView, &listener);
+    mDictionaryStructureWithBufferPolicy->iterateNgramEntries(prevWordIds, &listener);
 }
 
 int Dictionary::getProbability(const int *word, int length) const {
@@ -121,11 +121,11 @@
     if (!prevWordsInfo) {
         return getDictionaryStructurePolicy()->getProbabilityOfWord(WordIdArrayView(), wordId);
     }
-    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIds;
-    prevWordsInfo->getPrevWordIds(mDictionaryStructureWithBufferPolicy.get(), prevWordIds.data(),
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
+    const WordIdArrayView prevWordIds = prevWordsInfo->getPrevWordIds
+            (mDictionaryStructureWithBufferPolicy.get(), &prevWordIdArray,
             true /* tryLowerCaseSearch */);
-    return getDictionaryStructurePolicy()->getProbabilityOfWord(
-            IntArrayView::fromArray(prevWordIds), wordId);
+    return getDictionaryStructurePolicy()->getProbabilityOfWord(prevWordIds, wordId);
 }
 
 bool Dictionary::addUnigramEntry(const int *const word, const int length,
diff --git a/native/jni/src/suggest/core/dictionary/dictionary_utils.cpp b/native/jni/src/suggest/core/dictionary/dictionary_utils.cpp
index 617bf5a..d09266e 100644
--- a/native/jni/src/suggest/core/dictionary/dictionary_utils.cpp
+++ b/native/jni/src/suggest/core/dictionary/dictionary_utils.cpp
@@ -35,12 +35,11 @@
 
     // No prev words information.
     PrevWordsInfo emptyPrevWordsInfo;
-    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIds;
-    emptyPrevWordsInfo.getPrevWordIds(dictionaryStructurePolicy, prevWordIds.data(),
-            false /* tryLowerCaseSearch */);
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
+    const WordIdArrayView prevWordIds = emptyPrevWordsInfo.getPrevWordIds(
+            dictionaryStructurePolicy, &prevWordIdArray, false /* tryLowerCaseSearch */);
     current.emplace_back();
-    DicNodeUtils::initAsRoot(dictionaryStructurePolicy,
-            IntArrayView::fromArray(prevWordIds), &current.front());
+    DicNodeUtils::initAsRoot(dictionaryStructurePolicy, prevWordIds, &current.front());
     for (int i = 0; i < codePointCount; ++i) {
         // The base-lower input is used to ignore case errors and accent errors.
         const int codePoint = CharUtils::toBaseLowerCase(codePoints[i]);
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 4f58c0c..4d7505a 100644
--- a/native/jni/src/suggest/core/session/dic_traverse_session.cpp
+++ b/native/jni/src/suggest/core/session/dic_traverse_session.cpp
@@ -35,8 +35,8 @@
     mMultiWordCostMultiplier = getDictionaryStructurePolicy()->getHeaderStructurePolicy()
             ->getMultiWordCostMultiplier();
     mSuggestOptions = suggestOptions;
-    prevWordsInfo->getPrevWordIds(getDictionaryStructurePolicy(), mPrevWordsIds.data(),
-            true /* tryLowerCaseSearch */);
+    mPrevWordIdCount = prevWordsInfo->getPrevWordIds(getDictionaryStructurePolicy(),
+            &mPrevWordIdArray, true /* tryLowerCaseSearch */).size();
 }
 
 void DicTraverseSession::setupForGetSuggestions(const ProximityInfo *pInfo,
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 af19907..9f841aa 100644
--- a/native/jni/src/suggest/core/session/dic_traverse_session.h
+++ b/native/jni/src/suggest/core/session/dic_traverse_session.h
@@ -51,12 +51,11 @@
     }
 
     AK_FORCE_INLINE DicTraverseSession(JNIEnv *env, jstring localeStr, bool usesLargeCache)
-            : mProximityInfo(nullptr), mDictionary(nullptr), mSuggestOptions(nullptr),
-              mDicNodesCache(usesLargeCache), mMultiBigramMap(), mInputSize(0), mMaxPointerCount(1),
-              mMultiWordCostMultiplier(1.0f) {
+            : mPrevWordIdCount(0), 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.
-        mPrevWordsIds.fill(NOT_A_DICT_POS);
     }
 
     // Non virtual inline destructor -- never inherit this class
@@ -78,7 +77,9 @@
     //--------------------
     const ProximityInfo *getProximityInfo() const { return mProximityInfo; }
     const SuggestOptions *getSuggestOptions() const { return mSuggestOptions; }
-    const WordIdArrayView getPrevWordIds() const { return IntArrayView::fromArray(mPrevWordsIds); }
+    const WordIdArrayView getPrevWordIds() const {
+        return WordIdArrayView::fromArray(mPrevWordIdArray).limit(mPrevWordIdCount);
+    }
     DicNodesCache *getDicTraverseCache() { return &mDicNodesCache; }
     MultiBigramMap *getMultiBigramMap() { return &mMultiBigramMap; }
     const ProximityInfoState *getProximityInfoState(int id) const {
@@ -165,7 +166,8 @@
             const int *const inputYs, const int *const times, const int *const pointerIds,
             const int inputSize, const float maxSpatialDistance, const int maxPointerCount);
 
-    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> mPrevWordsIds;
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> mPrevWordIdArray;
+    size_t mPrevWordIdCount;
     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 fc9a359..02e82a8 100644
--- a/native/jni/src/suggest/core/session/prev_words_info.h
+++ b/native/jni/src/suggest/core/session/prev_words_info.h
@@ -17,6 +17,8 @@
 #ifndef LATINIME_PREV_WORDS_INFO_H
 #define LATINIME_PREV_WORDS_INFO_H
 
+#include <array>
+
 #include "defines.h"
 #include "suggest/core/policy/dictionary_structure_with_buffer_policy.h"
 #include "utils/char_utils.h"
@@ -27,12 +29,13 @@
 class PrevWordsInfo {
  public:
     // No prev word information.
-    PrevWordsInfo() {
+    PrevWordsInfo() : mPrevWordCount(0) {
         clear();
     }
 
-    PrevWordsInfo(PrevWordsInfo &&prevWordsInfo) {
-        for (size_t i = 0; i < NELEMS(mPrevWordCodePoints); ++i) {
+    PrevWordsInfo(PrevWordsInfo &&prevWordsInfo)
+            : mPrevWordCount(prevWordsInfo.mPrevWordCount) {
+        for (size_t i = 0; i < mPrevWordCount; ++i) {
             mPrevWordCodePointCount[i] = prevWordsInfo.mPrevWordCodePointCount[i];
             memmove(mPrevWordCodePoints[i], prevWordsInfo.mPrevWordCodePoints[i],
                     sizeof(mPrevWordCodePoints[i][0]) * mPrevWordCodePointCount[i]);
@@ -43,9 +46,10 @@
     // Construct from previous words.
     PrevWordsInfo(const int prevWordCodePoints[][MAX_WORD_LENGTH],
             const int *const prevWordCodePointCount, const bool *const isBeginningOfSentence,
-            const size_t prevWordCount) {
+            const size_t prevWordCount)
+            : mPrevWordCount(std::min(NELEMS(mPrevWordCodePoints), prevWordCount)) {
         clear();
-        for (size_t i = 0; i < std::min(NELEMS(mPrevWordCodePoints), prevWordCount); ++i) {
+        for (size_t i = 0; i < mPrevWordCount; ++i) {
             if (prevWordCodePointCount[i] < 0 || prevWordCodePointCount[i] > MAX_WORD_LENGTH) {
                 continue;
             }
@@ -58,7 +62,7 @@
 
     // Construct from a previous word.
     PrevWordsInfo(const int *const prevWordCodePoints, const int prevWordCodePointCount,
-            const bool isBeginningOfSentence) {
+            const bool isBeginningOfSentence) : mPrevWordCount(1) {
         clear();
         if (prevWordCodePointCount > MAX_WORD_LENGTH || !prevWordCodePoints) {
             return;
@@ -79,26 +83,29 @@
         return false;
     }
 
-    void getPrevWordIds(const DictionaryStructureWithBufferPolicy *const dictStructurePolicy,
-            int *const outPrevWordIds, const bool tryLowerCaseSearch) const {
-        for (size_t i = 0; i < NELEMS(mPrevWordCodePoints); ++i) {
-            outPrevWordIds[i] = getWordId(dictStructurePolicy,
+    template<size_t N>
+    const WordIdArrayView getPrevWordIds(
+            const DictionaryStructureWithBufferPolicy *const dictStructurePolicy,
+            std::array<int, N> *const prevWordIdBuffer, const bool tryLowerCaseSearch) const {
+        for (size_t i = 0; i < std::min(mPrevWordCount, N); ++i) {
+            prevWordIdBuffer->at(i) = getWordId(dictStructurePolicy,
                     mPrevWordCodePoints[i], mPrevWordCodePointCount[i],
                     mIsBeginningOfSentence[i], tryLowerCaseSearch);
         }
+        return WordIdArrayView::fromArray(*prevWordIdBuffer).limit(mPrevWordCount);
     }
 
     // n is 1-indexed.
-    const CodePointArrayView getNthPrevWordCodePoints(const int n) const {
-        if (n <= 0 || n > MAX_PREV_WORD_COUNT_FOR_N_GRAM) {
+    const CodePointArrayView getNthPrevWordCodePoints(const size_t n) const {
+        if (n <= 0 || n > mPrevWordCount) {
             return CodePointArrayView();
         }
         return CodePointArrayView(mPrevWordCodePoints[n - 1], mPrevWordCodePointCount[n - 1]);
     }
 
     // n is 1-indexed.
-    bool isNthPrevWordBeginningOfSentence(const int n) const {
-        if (n <= 0 || n > MAX_PREV_WORD_COUNT_FOR_N_GRAM) {
+    bool isNthPrevWordBeginningOfSentence(const size_t n) const {
+        if (n <= 0 || n > mPrevWordCount) {
             return false;
         }
         return mIsBeginningOfSentence[n - 1];
@@ -142,6 +149,7 @@
         }
     }
 
+    const size_t mPrevWordCount;
     int mPrevWordCodePoints[MAX_PREV_WORD_COUNT_FOR_N_GRAM][MAX_WORD_LENGTH];
     int mPrevWordCodePointCount[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
     bool mIsBeginningOfSentence[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
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 dfc3d2d..41b9a11 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
@@ -332,8 +332,12 @@
                 "length: %zd", bigramProperty->getTargetCodePoints()->size());
         return false;
     }
-    int prevWordIds[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
-    prevWordsInfo->getPrevWordIds(this, prevWordIds, false /* tryLowerCaseSearch */);
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
+    const WordIdArrayView prevWordIds = prevWordsInfo->getPrevWordIds(this, &prevWordIdArray,
+            false /* tryLowerCaseSearch */);
+    if (prevWordIds.empty()) {
+        return false;
+    }
     if (prevWordIds[0] == NOT_A_WORD_ID) {
         if (prevWordsInfo->isNthPrevWordBeginningOfSentence(1 /* n */)) {
             const std::vector<UnigramProperty::ShortcutProperty> shortcuts;
@@ -347,7 +351,7 @@
                 return false;
             }
             // Refresh word ids.
-            prevWordsInfo->getPrevWordIds(this, prevWordIds, false /* tryLowerCaseSearch */);
+            prevWordsInfo->getPrevWordIds(this, &prevWordIdArray, false /* tryLowerCaseSearch */);
         } else {
             return false;
         }
@@ -390,9 +394,10 @@
         AKLOGE("word is too long to remove n-gram entry form the dictionary. length: %zd",
                 wordCodePoints.size());
     }
-    int prevWordIds[MAX_PREV_WORD_COUNT_FOR_N_GRAM];
-    prevWordsInfo->getPrevWordIds(this, prevWordIds, false /* tryLowerCaseSerch */);
-    if (prevWordIds[0] == NOT_A_WORD_ID) {
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
+    const WordIdArrayView prevWordIds = prevWordsInfo->getPrevWordIds(this, &prevWordIdArray,
+            false /* tryLowerCaseSerch */);
+    if (prevWordIds.empty() || prevWordIds[0] == NOT_A_WORD_ID) {
         return false;
     }
     const int wordPos = getTerminalPtNodePosFromWordId(getWordId(wordCodePoints,
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 5b1907f..e624bf3 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
@@ -303,8 +303,12 @@
                 "length: %zd", bigramProperty->getTargetCodePoints()->size());
         return false;
     }
-    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIds;
-    prevWordsInfo->getPrevWordIds(this, prevWordIds.data(), false /* tryLowerCaseSearch */);
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
+    const WordIdArrayView prevWordIds = prevWordsInfo->getPrevWordIds(this, &prevWordIdArray,
+            false /* tryLowerCaseSearch */);
+    if (prevWordIds.empty()) {
+        return false;
+    }
     // TODO: Support N-gram.
     if (prevWordIds[0] == NOT_A_WORD_ID) {
         if (prevWordsInfo->isNthPrevWordBeginningOfSentence(1 /* n */)) {
@@ -319,7 +323,7 @@
                 return false;
             }
             // Refresh word ids.
-            prevWordsInfo->getPrevWordIds(this, prevWordIds.data(), false /* tryLowerCaseSearch */);
+            prevWordsInfo->getPrevWordIds(this, &prevWordIdArray, false /* tryLowerCaseSearch */);
         } else {
             return false;
         }
@@ -367,10 +371,11 @@
         AKLOGE("word is too long to remove n-gram entry form the dictionary. length: %zd",
                 wordCodePoints.size());
     }
-    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIds;
-    prevWordsInfo->getPrevWordIds(this, prevWordIds.data(), false /* tryLowerCaseSerch */);
+    WordIdArray<MAX_PREV_WORD_COUNT_FOR_N_GRAM> prevWordIdArray;
+    const WordIdArrayView prevWordIds = prevWordsInfo->getPrevWordIds(this, &prevWordIdArray,
+            false /* tryLowerCaseSerch */);
     // TODO: Support N-gram.
-    if (prevWordIds[0] == NOT_A_WORD_ID) {
+    if (prevWordIds.empty() || prevWordIds[0] == NOT_A_WORD_ID) {
         return false;
     }
     const int wordId = getWordId(wordCodePoints, false /* forceLowerCaseSearch */);