diff --git a/native/jni/src/bigram_dictionary.cpp b/native/jni/src/bigram_dictionary.cpp
index 1da5212..b572278 100644
--- a/native/jni/src/bigram_dictionary.cpp
+++ b/native/jni/src/bigram_dictionary.cpp
@@ -26,6 +26,7 @@
 #include "suggest/core/dictionary/bloom_filter.h"
 #include "suggest/core/dictionary/char_utils.h"
 #include "suggest/core/dictionary/dictionary.h"
+#include "suggest/core/dictionary/probability_utils.h"
 
 namespace latinime {
 
@@ -136,7 +137,7 @@
             // resulting probability is 8 - although in the practice it's never bigger than 3 or 4
             // in very bad cases. This means that sometimes, we'll see some bigrams interverted
             // here, but it can't get too bad.
-            const int probability = BinaryFormat::computeProbabilityForBigram(
+            const int probability = ProbabilityUtils::computeProbabilityForBigram(
                     unigramProbability, bigramProbabilityTemp);
             addWordBigram(bigramBuffer, length, probability, bigramProbability, bigramCodePoints,
                     outputTypes);
diff --git a/native/jni/src/suggest/core/dicnode/dic_node_utils.cpp b/native/jni/src/suggest/core/dicnode/dic_node_utils.cpp
index 77afa9e..46dbe43 100644
--- a/native/jni/src/suggest/core/dicnode/dic_node_utils.cpp
+++ b/native/jni/src/suggest/core/dicnode/dic_node_utils.cpp
@@ -24,6 +24,7 @@
 #include "suggest/core/dictionary/binary_format.h"
 #include "suggest/core/dictionary/char_utils.h"
 #include "suggest/core/dictionary/multi_bigram_map.h"
+#include "suggest/core/dictionary/probability_utils.h"
 #include "suggest/core/layout/proximity_info.h"
 #include "suggest/core/layout/proximity_info_state.h"
 
@@ -226,7 +227,7 @@
     const int prevWordPos = node->getPrevWordPos();
     if (NOT_VALID_WORD == wordPos || NOT_VALID_WORD == prevWordPos) {
         // Note: Normally wordPos comes from the dictionary and should never equal NOT_VALID_WORD.
-        return backoff(unigramProbability);
+        return ProbabilityUtils::backoff(unigramProbability);
     }
     if (multiBigramMap) {
         return multiBigramMap->getBigramProbability(
diff --git a/native/jni/src/suggest/core/dictionary/binary_format.h b/native/jni/src/suggest/core/dictionary/binary_format.h
index ef9fd37..3bacc09 100644
--- a/native/jni/src/suggest/core/dictionary/binary_format.h
+++ b/native/jni/src/suggest/core/dictionary/binary_format.h
@@ -18,12 +18,12 @@
 #define LATINIME_BINARY_FORMAT_H
 
 #include <cstdlib>
-#include <map>
 #include <stdint.h>
 
 #include "hash_map_compat.h"
 #include "suggest/core/dictionary/bloom_filter.h"
 #include "suggest/core/dictionary/char_utils.h"
+#include "suggest/core/dictionary/probability_utils.h"
 
 namespace latinime {
 
@@ -91,10 +91,6 @@
             const int length, const bool forceLowerCaseSearch);
     static int getWordAtAddress(const uint8_t *const root, const int address, const int maxDepth,
             int *outWord, int *outUnigramProbability);
-    static int computeProbabilityForBigram(
-            const int unigramProbability, const int bigramProbability);
-    static int getProbability(const int position, const std::map<int, int> *bigramMap,
-            const uint8_t *bigramFilter, const int unigramProbability);
     static int getBigramProbabilityFromHashMap(const int position,
             const hash_map_compat<int, int> *bigramMap, const int unigramProbability);
     static float getMultiWordCostMultiplier(const uint8_t *const dict, const int dictSize);
@@ -678,51 +674,18 @@
     return 0;
 }
 
-static inline int backoff(const int unigramProbability) {
-    return unigramProbability;
-    // For some reason, applying the backoff weight gives bad results in tests. To apply the
-    // backoff weight, we divide the probability by 2, which in our storing format means
-    // decreasing the score by 8.
-    // TODO: figure out what's wrong with this.
-    // return unigramProbability > 8 ? unigramProbability - 8 : (0 == unigramProbability ? 0 : 8);
-}
-
-inline int BinaryFormat::computeProbabilityForBigram(
-        const int unigramProbability, const int bigramProbability) {
-    // We divide the range [unigramProbability..255] in 16.5 steps - in other words, we want the
-    // unigram probability to be the median value of the 17th step from the top. A value of
-    // 0 for the bigram probability represents the middle of the 16th step from the top,
-    // while a value of 15 represents the middle of the top step.
-    // See makedict.BinaryDictInputOutput for details.
-    const float stepSize = static_cast<float>(MAX_PROBABILITY - unigramProbability)
-            / (1.5f + MAX_BIGRAM_ENCODED_PROBABILITY);
-    return unigramProbability
-            + static_cast<int>(static_cast<float>(bigramProbability + 1) * stepSize);
-}
-
-// This returns a probability in log space.
-inline int BinaryFormat::getProbability(const int position, const std::map<int, int> *bigramMap,
-        const uint8_t *bigramFilter, const int unigramProbability) {
-    if (!bigramMap || !bigramFilter) return backoff(unigramProbability);
-    if (!isInFilter(bigramFilter, position)) return backoff(unigramProbability);
-    const std::map<int, int>::const_iterator bigramProbabilityIt = bigramMap->find(position);
-    if (bigramProbabilityIt != bigramMap->end()) {
-        const int bigramProbability = bigramProbabilityIt->second;
-        return computeProbabilityForBigram(unigramProbability, bigramProbability);
-    }
-    return backoff(unigramProbability);
-}
-
 // This returns a probability in log space.
 inline int BinaryFormat::getBigramProbabilityFromHashMap(const int position,
         const hash_map_compat<int, int> *bigramMap, const int unigramProbability) {
-    if (!bigramMap) return backoff(unigramProbability);
+    if (!bigramMap) {
+        return ProbabilityUtils::backoff(unigramProbability);
+    }
     const hash_map_compat<int, int>::const_iterator bigramProbabilityIt = bigramMap->find(position);
     if (bigramProbabilityIt != bigramMap->end()) {
         const int bigramProbability = bigramProbabilityIt->second;
-        return computeProbabilityForBigram(unigramProbability, bigramProbability);
+        return ProbabilityUtils::computeProbabilityForBigram(unigramProbability, bigramProbability);
     }
-    return backoff(unigramProbability);
+    return ProbabilityUtils::backoff(unigramProbability);
 }
 
 AK_FORCE_INLINE void BinaryFormat::fillBigramProbabilityToHashMap(
@@ -743,7 +706,9 @@
 AK_FORCE_INLINE int BinaryFormat::getBigramProbability(const uint8_t *const root, int position,
         const int nextPosition, const int unigramProbability) {
     position = getBigramListPositionForWordPosition(root, position);
-    if (0 == position) return backoff(unigramProbability);
+    if (0 == position) {
+        return ProbabilityUtils::backoff(unigramProbability);
+    }
 
     uint8_t bigramFlags;
     do {
@@ -752,10 +717,11 @@
                 root, bigramFlags, &position);
         if (bigramPos == nextPosition) {
             const int bigramProbability = MASK_ATTRIBUTE_PROBABILITY & bigramFlags;
-            return computeProbabilityForBigram(unigramProbability, bigramProbability);
+            return ProbabilityUtils::computeProbabilityForBigram(
+                    unigramProbability, bigramProbability);
         }
     } while (FLAG_ATTRIBUTE_HAS_NEXT & bigramFlags);
-    return backoff(unigramProbability);
+    return ProbabilityUtils::backoff(unigramProbability);
 }
 
 // Returns a pointer to the start of the bigram list.
diff --git a/native/jni/src/suggest/core/dictionary/probability_utils.h b/native/jni/src/suggest/core/dictionary/probability_utils.h
new file mode 100644
index 0000000..14d2f84
--- /dev/null
+++ b/native/jni/src/suggest/core/dictionary/probability_utils.h
@@ -0,0 +1,74 @@
+/*
+ * Copyright (C) 2013, 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.
+ */
+
+#ifndef LATINIME_PROBABILITY_UTILS_H
+#define LATINIME_PROBABILITY_UTILS_H
+
+#include <map>
+#include <stdint.h>
+
+#include "defines.h"
+
+namespace latinime {
+
+class ProbabilityUtils {
+ public:
+    static AK_FORCE_INLINE int backoff(const int unigramProbability) {
+        return unigramProbability;
+        // For some reason, applying the backoff weight gives bad results in tests. To apply the
+        // backoff weight, we divide the probability by 2, which in our storing format means
+        // decreasing the score by 8.
+        // TODO: figure out what's wrong with this.
+        // return unigramProbability > 8 ?
+        //         unigramProbability - 8 : (0 == unigramProbability ? 0 : 8);
+    }
+
+    static AK_FORCE_INLINE int computeProbabilityForBigram(
+            const int unigramProbability, const int bigramProbability) {
+        // We divide the range [unigramProbability..255] in 16.5 steps - in other words, we want
+        // the unigram probability to be the median value of the 17th step from the top. A value of
+        // 0 for the bigram probability represents the middle of the 16th step from the top,
+        // while a value of 15 represents the middle of the top step.
+        // See makedict.BinaryDictInputOutput for details.
+        const float stepSize = static_cast<float>(MAX_PROBABILITY - unigramProbability)
+                / (1.5f + MAX_BIGRAM_ENCODED_PROBABILITY);
+        return unigramProbability
+                + static_cast<int>(static_cast<float>(bigramProbability + 1) * stepSize);
+    }
+
+    // This returns a probability in log space.
+    static AK_FORCE_INLINE int getProbability(const int position,
+            const std::map<int, int> *const bigramMap,
+            const uint8_t *bigramFilter, const int unigramProbability) {
+        if (!bigramMap || !bigramFilter) {
+            return backoff(unigramProbability);
+        }
+        if (!isInFilter(bigramFilter, position)){
+            return backoff(unigramProbability);
+        }
+        const std::map<int, int>::const_iterator bigramProbabilityIt = bigramMap->find(position);
+        if (bigramProbabilityIt != bigramMap->end()) {
+            const int bigramProbability = bigramProbabilityIt->second;
+            return computeProbabilityForBigram(unigramProbability, bigramProbability);
+        }
+        return backoff(unigramProbability);
+    }
+
+ private:
+    DISALLOW_IMPLICIT_CONSTRUCTORS(ProbabilityUtils);
+};
+}
+#endif /* LATINIME_PROBABILITY_UTILS_H */
diff --git a/native/jni/src/unigram_dictionary.cpp b/native/jni/src/unigram_dictionary.cpp
index fa31538..8fd015c 100644
--- a/native/jni/src/unigram_dictionary.cpp
+++ b/native/jni/src/unigram_dictionary.cpp
@@ -24,6 +24,7 @@
 #include "suggest/core/dictionary/char_utils.h"
 #include "suggest/core/dictionary/dictionary.h"
 #include "suggest/core/dictionary/digraph_utils.h"
+#include "suggest/core/dictionary/probability_utils.h"
 #include "suggest/core/dictionary/terminal_attributes.h"
 #include "suggest/core/layout/proximity_info.h"
 #include "unigram_dictionary.h"
@@ -940,8 +941,8 @@
         TerminalAttributes terminalAttributes(mBinaryDicitonaryInfo, flags, attributesPos);
         // bigramMap contains the bigram frequencies indexed by addresses for fast lookup.
         // bigramFilter is a bloom filter of said frequencies for even faster rejection.
-        const int probability = BinaryFormat::getProbability(initialPos, bigramMap, bigramFilter,
-                unigramProbability);
+        const int probability = ProbabilityUtils::getProbability(
+                initialPos, bigramMap, bigramFilter, unigramProbability);
         onTerminal(probability, terminalAttributes, correction, queuePool, needsToInvokeOnTerminal,
                 currentWordIndex);