diff --git a/camphr/pipelines/pattern_search.py b/camphr/pipelines/pattern_search.py
index 2371a24f..67c9882d 100644
--- a/camphr/pipelines/pattern_search.py
+++ b/camphr/pipelines/pattern_search.py
@@ -1,10 +1,21 @@
 """Defines pattern search pipeline based on ahocorasik."""
-from typing import Callable, Dict, Iterable, Iterator, List, Optional, Tuple, cast
+from typing import (
+    Callable,
+    Dict,
+    Iterable,
+    Iterator,
+    List,
+    Optional,
+    Tuple,
+    Union,
+    cast,
+)
 
 import ahocorasick
 import spacy
 import textspan
 from spacy.tokens import Doc
+from spacy.tokens.span import Span
 from spacy.util import filter_spans
 from typing_extensions import Literal
 
@@ -106,7 +117,7 @@ def get_char_spans(self, text: str) -> Iterator[Tuple[int, int, str]]:
             i = j - len(word) + 1
             yield i, j + 1, word
 
-    def _to_text(self, doc: Doc) -> str:
+    def _to_text(self, doc: Union[Doc, Span]) -> str:
         if self.lemma:
             text = _to_lemma_text(doc)
         else:
@@ -133,7 +144,7 @@ def __call__(self, doc: Doc) -> Doc:
                 covering=not self.destructive,
                 label=self.get_label(text),
             )
-            if span:
+            if span and self._to_text(span) == text:
                 spans.append(span)
         [s.text for s in spans]  # TODO: resolve the evaluation bug and remove this line
         ents = filter_spans(doc.ents + tuple(spans))
@@ -141,7 +152,7 @@ def __call__(self, doc: Doc) -> Doc:
         return doc
 
 
-def _to_lemma_text(doc: Doc) -> str:
+def _to_lemma_text(doc: Union[Doc, Span]) -> str:
     ret = ""
     for token in doc:
         ret += token.lemma_
diff --git a/camphr/utils.py b/camphr/utils.py
index dc221506..374dc9ea 100644
--- a/camphr/utils.py
+++ b/camphr/utils.py
@@ -56,8 +56,8 @@ def token_from_char_pos(doc: Doc, i: int) -> Token:
 
 def _get_covering_span(doc: Doc, i: int, j: int) -> Span:
     token_idxs = [t.idx for t in doc]
-    i = doc[bisect.bisect(token_idxs, i) - 1].i
-    j = doc[bisect.bisect_left(token_idxs, j)].i
+    i = bisect.bisect(token_idxs, i) - 1
+    j = bisect.bisect_left(token_idxs, j)
     return doc[i:j]
 
 
diff --git a/tests/pipelines/test_pattern_search.py b/tests/pipelines/test_pattern_search.py
index 12a41151..e3400b46 100644
--- a/tests/pipelines/test_pattern_search.py
+++ b/tests/pipelines/test_pattern_search.py
@@ -23,7 +23,7 @@ def nlp(lang):
     _nlp = spacy.blank(lang)
     pipe = PatternSearcher.from_words(
         KEYWORDS,
-        destructive=True,
+        destructive=False,
         lower=True,
         lemma=True,
         normalizer=lambda x: re.sub(r"\W", "", x),
@@ -36,7 +36,8 @@ def nlp(lang):
     ("今日はいい天気だ", ["今日", "は"], "ja_mecab"),
     ("Mice is a plural form of mouse", ["mouse"], "en"),
     ("foo-bar", ["foo-bar"], "en"),
-    ("たくさん走った", ["走"], "ja_mecab"),
+    ("たくさん走った", ["走っ"], "ja_mecab"),
+    ("走れ", ["走れ"], "ja_mecab"),
 ]
 
 
@@ -49,11 +50,13 @@ def test_call(nlp, text, expected, target, lang):
     assert ents == expected
 
 
-def test_serialization(nlp, tmpdir):
+def test_serialization(nlp, tmpdir, lang):
+    text, expected, target = TESTCASES[0]
+    if lang != target:
+        pytest.skip(f"target lang is '{target}', but actual lang is {lang}")
     path = Path(tmpdir)
     nlp.to_disk(path)
     nlp = spacy.load(path)
-    text, expected, _ = TESTCASES[0]
     doc = nlp(text)
     ents = [span.text for span in doc.ents]
-    assert ents == expected
+    assert ents == expected, list(doc)