diff --git a/itn/chinese/rules/fraction.py b/itn/chinese/rules/fraction.py
index f3c8117b..f61a1cdf 100644
--- a/itn/chinese/rules/fraction.py
+++ b/itn/chinese/rules/fraction.py
@@ -15,7 +15,8 @@
 from itn.chinese.rules.cardinal import Cardinal
 from tn.processor import Processor
 
-from pynini.lib.pynutil import delete, insert
+from pynini import string_file
+from pynini.lib.pynutil import delete, insert, add_weight
 
 
 class Fraction(Processor):
@@ -27,14 +28,21 @@ def __init__(self):
 
     def build_tagger(self):
         number = Cardinal().number
-
-        tagger = (insert('denominator: "') + number +
+        sign = string_file('itn/chinese/data/number/sign.tsv')    # + -
+
+        # NOTE(xcsong): default weight = 1.0,  set to -1.0 means higher priority
+        #   For example,
+        #       1.0, 负二分之三 -> { sign: "" denominator: "-2" numerator: "3" }
+        #       -1.0,负二分之三 -> { sign: "-" denominator: "2" numerator: "3" }
+        tagger = (insert('sign: "') + add_weight(sign, -1.0).ques +
+                  insert('" denominator: "') + number +
                   delete('分之') + insert('" numerator: "') +
                   number + insert('"'))
         self.tagger = self.add_tokens(tagger)
 
     def build_verbalizer(self):
-        numerator = delete('numerator: "') + self.SIGMA + delete('"')
+        sign = delete('sign: "') + self.SIGMA + delete('"')
+        numerator = delete(' numerator: "') + self.SIGMA + delete('"')
         denominator = delete(' denominator: "') + self.SIGMA + delete('"')
-        verbalizer = numerator + insert('/') + denominator
+        verbalizer = sign + numerator + insert('/') + denominator
         self.verbalizer = self.delete_tokens(verbalizer)
diff --git a/runtime/processor/token_parser.cc b/runtime/processor/token_parser.cc
index 4e640bdb..21a20ccc 100644
--- a/runtime/processor/token_parser.cc
+++ b/runtime/processor/token_parser.cc
@@ -34,7 +34,7 @@ const std::unordered_map<std::string, std::vector<std::string>> TN_ORDERS = {
     {"time", {"noon", "hour", "minute", "second"}}};
 const std::unordered_map<std::string, std::vector<std::string>> ITN_ORDERS = {
     {"date", {"year", "month", "day"}},
-    {"fraction", {"numerator", "denominator"}},
+    {"fraction", {"sign", "numerator", "denominator"}},
     {"measure", {"numerator", "denominator", "value"}},
     {"money", {"currency", "value"}},
     {"time", {"hour", "minute", "second", "noon"}}};
diff --git a/tn/token_parser.py b/tn/token_parser.py
index 5db6b606..ef32222f 100644
--- a/tn/token_parser.py
+++ b/tn/token_parser.py
@@ -23,7 +23,7 @@
     'time': ['noon', 'hour', 'minute', 'second']}
 ITN_ORDERS = {
     'date': ['year', 'month', 'day'],
-    'fraction': ['numerator', 'denominator'],
+    'fraction': ['sign', 'numerator', 'denominator'],
     'measure': ['numerator', 'denominator', 'value'],
     'money': ['currency', 'value'],
     'time': ['hour', 'minute', 'second', 'noon']}