diff --git a/tests/test_normalizer.py b/tests/test_normalizer.py
index 7f89da0ec..fb0d064b3 100644
--- a/tests/test_normalizer.py
+++ b/tests/test_normalizer.py
@@ -89,6 +89,29 @@ def test_spelling_normalizer():
     assert std("cancelation") == "cancellation"
 
 
+@pytest.mark.parametrize("std", [EnglishNumberNormalizer(), EnglishTextNormalizer()])
+@pytest.mark.parametrize(
+    "text, expected",
+    [
+        ("the positive and negative effects", "the positive and negative effects"),
+        ("the plus and minus signs", "the plus and minus signs"),
+        ("a positive point of view", "a positive point of view"),
+        ("a negative percent change", "a negative percent change"),
+        ("plus dollars for shipping", "plus dollars for shipping"),
+        ("positive double vision", "positive double vision"),
+        ("positive and negative five", "positive and -5"),
+        ("minus point five", "-.5"),
+        ("negative point 5", "-.5"),
+        ("minus double five", "-55"),
+        ("plus triple oh", "+000"),
+        ("positive twenty thousand", "+20000"),
+        ("negative zero", "-0"),
+    ],
+)
+def test_numeric_prefix_requires_a_number(std, text, expected):
+    assert std(text) == expected
+
+
 def test_text_normalizer():
     std = EnglishTextNormalizer()
     assert std("Let's") == "let us"
diff --git a/whisper/normalizers/english.py b/whisper/normalizers/english.py
index 21c3a54e5..5ac76d6e0 100644
--- a/whisper/normalizers/english.py
+++ b/whisper/normalizers/english.py
@@ -161,6 +161,12 @@ def __init__(self):
             ]
         )
         self.literal_words = {"one", "ones"}
+        self.number_starters = self.words.difference(
+            self.preceding_prefixers,
+            self.following_prefixers,
+            self.suffixers,
+            self.specials,
+        )
 
     def process_words(self, words: List[str]) -> Iterator[str]:
         prefix: Optional[str] = None
@@ -185,7 +191,9 @@ def output(result: Union[str, int]):
         if len(words) == 0:
             return
 
-        for prev, current, next in windowed([None] + words + [None], 3):
+        for prev, current, next, after_next in windowed(
+            [None] + words + [None, None], 4
+        ):
             if skip:
                 skip = False
                 continue
@@ -324,7 +332,18 @@ def output(result: Union[str, int]):
                 if value is not None:
                     yield output(value)
 
-                if next in self.words or next_is_numeric:
+                next_starts_number = next in self.number_starters or next_is_numeric
+                if next == "point":
+                    next_starts_number = after_next in self.decimals or (
+                        after_next is not None
+                        and re.match(r"^\d+(\.\d+)?$", after_next)
+                    )
+                elif next in {"double", "triple"}:
+                    next_starts_number = (
+                        after_next in self.ones or after_next in self.zeros
+                    )
+
+                if next_starts_number:
                     prefix = self.preceding_prefixers[current]
                 else:
                     yield output(current)