diff --git a/.gitignore b/.gitignore
index f3643de..bf6dfb4 100644
--- a/.gitignore
+++ b/.gitignore
@@ -1,11 +1,7 @@
 __pycache__
-build
+*.pyc
 trf.egg-info
-tests/faster-rnnlm
-tests/uniq.dat
-tests/test.input
-tests/rnnlm.output
-tests/__init__.pyc
-tests/test_acceptability.pyc
+
+build
 tools
 data
diff --git a/tests/test_acceptability.py b/tests/test_acceptability.py
index 4f12787..7b51d8e 100644
--- a/tests/test_acceptability.py
+++ b/tests/test_acceptability.py
@@ -1,6 +1,4 @@
 import unittest
-import warnings
-import tempfile
 
 from trf.acceptability import Acceptability
 from trf.util import check_executable
@@ -20,34 +18,34 @@ def setUp(self):
                           self.delimiter,
                           self.rnnlm_model_path)
 
-    def test_rnnlm_scores(self):
-        scores = self.acceptability.rnnlm_scores
+    def test_log_prob(self):
+        scores = self.acceptability._calc_log_prob_scores()
         self.assertAlmostEqual(scores[0], -11.571, places=2)
 
     def test_unigram_scores(self):
 
-        scores = self.acceptability.unigram_scores
+        scores = self.acceptability._calc_unigram_scores()
         self.assertAlmostEqual(scores[0], -31.457, places=2)
 
-    # def test_mean_unigram_scores(self):
+    def test_mean_lp_scores(self):
 
-    #    scores = self.acceptability.mean_unigram_scores
-    #    self.assertAlmostEqual(scores[0], -2.12, places=2)
+        score = self.acceptability.mean_lp
+        self.assertAlmostEqual(score, -2.892, places=2)
 
-    # def test_normalized_scores_div(self):
+    def test_norm_lp_div(self):
 
-    #    scores = self.acceptability.normalized_scores_div
-    #    self.assertAlmostEqual(scores[0], -5.446, places=2)
+        score = self.acceptability.norm_lp_div
+        self.assertAlmostEqual(score, -0.3678, places=2)
 
-    # def test_normalized_scores_sub(self):
+    def test_norm_lp_sub(self):
 
-    #    scores = self.acceptability.normalized_scores_sub
-    #    self.assertAlmostEqual(scores[0], -9.447, places=2)
+        score = self.acceptability.norm_lp_sub
+        self.assertAlmostEqual(score, 19.885, places=2)
 
-    # def test_normalized_scores_len(self):
+    def test_slor(self):
 
-    #    scores = self.acceptability.normalized_scores_len
-    #    self.assertAlmostEqual(scores[0], -0.9447, places=2)
+        score = self.acceptability.slor
+        self.assertAlmostEqual(score, 4.9713, places=2)
 
     def tearDown(self):
         pass
diff --git a/trf/__init__.pyc b/trf/__init__.pyc
new file mode 100644
index 0000000..ed8b29e
Binary files /dev/null and b/trf/__init__.pyc differ
diff --git a/trf/acceptability.py b/trf/acceptability.py
index 2328616..cb9f944 100644
--- a/trf/acceptability.py
+++ b/trf/acceptability.py
@@ -8,7 +8,6 @@
 from janome.tokenizer import Tokenizer
 
 import trf.constant as const
-from trf.analyser import Tree
 from trf.util import split_text
 
 
@@ -17,8 +16,8 @@ class Acceptability:
     def __init__(self, text: str, delimiter: str, rnnlm_model_path: str):
 
         self.text = text
-        self.sentences = split_text(text, delimiter)
-        self.tss = tokenize_by_janome(self.sentences)
+        self.sentences = split_text(text, delimiter)  # type: List[str]
+        lengths, self.tss = tokenize(self.sentences)
 
         if not os.path.isfile(rnnlm_model_path):
             raise FileNotFoundError(errno.ENOENT,
@@ -28,26 +27,27 @@ def __init__(self, text: str, delimiter: str, rnnlm_model_path: str):
 
         self.word_freq, self.n_total_words = self._load_word_freq(threshold=1)
 
-        self.rnnlm_scores = self.get_rnnlm_scores()
-        self.unigram_scores = self.calc_unigram_scores()
-
-        self.mean_unigram_scores = self.calc_mean_unigram_scores()
-
-        # self.normalized_scores_div = \
-        #     self.calc_normalized_scores('div')
-
-        # self.normalized_scores_sub = \
-        #     self.calc_normalized_scores('sub')
-
-        # self.normalized_scores_len = \
-        #     self.calc_normalized_scores('len')
-
-        self.mean_loglikelihood = \
-            None \
-            if None in self.rnnlm_scores \
-            else numpy.mean(self.rnnlm_scores)
-
-    def get_rnnlm_scores(self) -> List[Union[None, float]]:
+        log_prob_scores = \
+            self._calc_log_prob_scores()
+        unigram_scores = \
+            self._calc_unigram_scores()
+
+        mean_lp_scores = \
+            calc_mean_lp_scores(log_prob_scores, lengths)
+        norm_lp_div_scores = \
+            calc_norm_lp_div_scores(log_prob_scores, unigram_scores)
+        norm_lp_sub_scores = \
+            calc_norm_lp_sub_scores(log_prob_scores, unigram_scores)
+        slor_scores = \
+            calc_slor_scores(norm_lp_sub_scores, lengths)
+
+        self.log_prob = average(log_prob_scores)
+        self.mean_lp = average(mean_lp_scores)
+        self.norm_lp_div = average(norm_lp_div_scores)
+        self.norm_lp_sub = average(norm_lp_sub_scores)
+        self.slor = average(slor_scores)
+
+    def _calc_log_prob_scores(self) -> List[Union[None, float]]:
         """Get log likelihood scores by calling RNNLM
         """
 
@@ -62,7 +62,7 @@ def get_rnnlm_scores(self) -> List[Union[None, float]]:
                    '-test',
                    textfile.name]
         process = Popen(command, stdout=PIPE, stderr=PIPE)
-        output , err = process.communicate()
+        output, err = process.communicate()
         lines = [line.strip() for line in output.decode('UTF-8').split('\n')
                  if line.strip() != '']
         scores = []
@@ -95,7 +95,7 @@ def _load_word_freq(self, threshold: int) -> Tuple[Dict[str, int], int]:
 
         return (word_freq, n_total_words)
 
-    def calc_unigram_scores(self) -> List[float]:
+    def _calc_unigram_scores(self) -> List[float]:
 
         unigram_scores = []
         for ts in self.tss:
@@ -110,47 +110,120 @@ def calc_unigram_scores(self) -> List[float]:
 
         return unigram_scores
 
-    def calc_mean_unigram_scores(self) -> List[Union[None, float]]:
-        mean_unigram_scores = []
-        for score, sentence in zip(self.unigram_scores, self.sentences):
-            n = len(self.sentences)
-            x = None \
-                if score is None or n == 0 \
-                else float(score) / float(len(self.sentences))
-            mean_unigram_scores.append(x)
-        return mean_unigram_scores
-
-    def calc_normalized_scores(self, method: str) -> List[Union[None, float]]:
-
-        normalized_scores = []
-        for score, unigram_score, s in zip(self.rnnlm_scores,
-                                           self.unigram_scores,
-                                           self.sentences):
-            x = None \
-                if score is None or numpy.isclose(unigram_score,
-                                                  0.0, rtol=1e-05) \
-                else _f(score, unigram_score, len(s), method)
-            normalized_scores.append(x)
-        return normalized_scores
-
-
-def _f(score: float, unigram_score: float, length: int, method: str) -> float:
-
-    if method == 'div':
-        return (-1) * float(score) / float(unigram_score)
-    elif method == 'sub':
-        return float(score) - float(unigram_score)
-    elif method == 'len':
-        return (float(score) - float(unigram_score)) / length
-    else:
-        raise ValueError
-
-
-def tokenize_by_janome(sentences: List[str]) -> List[List[str]]:
+
+def average(xs: List[Union[None, float]]) -> float:
+    """Calculate the arithmetic mean of the given values (possibly None)
+    >>> '{:.2f}'.format(average([None, 1.0, 2.0]))
+    '1.50'
+    """
+    return numpy.mean([x for x in xs if x is not None])
+
+
+def calc_mean_lp_scores(log_prob_scores: List[float],
+                        lengths: List[int]) -> List[Union[None, float]]:
+    r"""
+    .. math:
+        \frac{%
+            \log P_\text{model}\left(\xi\right)
+            }{%
+              \text{length}\left(\xi\right)
+            }
+    >>> '{:.3f}'.format(calc_mean_lp_scores([-14.7579], [4])[0])
+    '-3.689'
+    """
+    mean_lp_scores = []
+    for score, length in zip(log_prob_scores, lengths):
+        x = None \
+            if score is None or length == 0 \
+            else float(score) / float(length)
+        mean_lp_scores.append(x)
+    return mean_lp_scores
+
+
+def calc_norm_lp_div_scores(
+        log_prob_scores: List[float],
+        unigram_scores: List[float]) -> List[Union[None, float]]:
+    r"""
+    .. math:
+        \frac{%
+            \log P_\text{model}\left(\xi\right)
+        }{%
+            \log P_\text{unigram}\left(\xi\right)
+        }
+    >>> '{:.3f}'.format(calc_norm_lp_div_scores([-14.7579], [-35.6325])[0])
+    '-0.414'
+    """
+    results = []
+    for log_prob, unigram_score in zip(log_prob_scores, unigram_scores):
+        if log_prob is None or numpy.isclose(unigram_score, 0.0, rtol=1e-05):
+            x = None
+        else:
+            x = (-1.0) * float(log_prob) / float(unigram_score)
+        results.append(x)
+    return results
+
+
+def calc_norm_lp_sub_scores(
+        log_prob_scores: List[float],
+        unigram_scores: List[float]) -> List[Union[None, float]]:
+    r"""
+    .. math:
+        \log P_\text{model}\left(\xi\right)
+            - \log P_\text{unigram}\left(\xi\right)
+    >>> '{:.3f}'.format(calc_norm_lp_sub_scores([-14.7579], [-35.6325])[0])
+    '20.875'
+    """
+
+    results = []
+    for log_prob, unigram_score in zip(log_prob_scores, unigram_scores):
+        if log_prob is None or numpy.isclose(unigram_score, 0.0, rtol=1e-05):
+            x = None
+        else:
+            x = float(log_prob) - float(unigram_score)
+        results.append(x)
+    return results
+
+
+def calc_slor_scores(norm_lp_sub_scores: List[float],
+                     lengths: List[int]) -> List[Union[None, float]]:
+    r"""Calculate SLOR (Syntactic Log-Odds Ratio)
+    .. math:
+        \frac{%
+            \log P_\text{model}\left(\xi\right)
+                - \log P_\text{unigram}\left(\xi\right)
+        }{%
+            \text{length}\left(\xi\right)
+        }
+    >>> '{:.3f}'.format(calc_slor_scores([20.8746], [4])[0])
+    '5.219'
+    """
+
+    results = []
+    for norm_lp_sub_score, length in zip(norm_lp_sub_scores, lengths):
+        if (norm_lp_sub_score is None) or length == 0:
+            x = None
+        else:
+            x = norm_lp_sub_score / length
+        results.append(x)
+    return results
+
+
+def tokenize(sentences: List[str]) -> Tuple[List[int], List[List[str]]]:
+
     tokenizer = Tokenizer()
-    tss = []
+    lengths = []
+    texts = []
     for s in sentences:
         result = tokenizer.tokenize(s)
-        ts = ' '.join([t.surface for t in result])
-        tss.append(ts)
-    return tss
+
+        surfaces = [t.surface for t in result]
+        lengths.append(len(surfaces))
+
+        text = ' '.join(surfaces)
+        texts.append(text)
+    return lengths, texts
+
+
+if __name__ == '__main__':
+    import doctest
+    doctest.testmod(verbose=True)
diff --git a/trf/cmdline.py b/trf/cmdline.py
index 7280e14..8717917 100644
--- a/trf/cmdline.py
+++ b/trf/cmdline.py
@@ -21,8 +21,16 @@ def translate(en: str):
         return '係り受け木の深さ'
     elif en == 'r_conditional':
         return '仮定節'
-    elif en == 'mean_loglikelihood':
-        return '言語モデルの尤度'
+    elif en == 'log_prob':
+        return '容認度 (LogProb)'
+    elif en == 'mean_lp':
+        return '容認度 (Mean LP)'
+    elif en == 'norm_lp_div':
+        return '容認度 (Norm LP (Div))'
+    elif en == 'norm_lp_sub':
+        return '容認度 (Norm LP (Sub))'
+    elif en == 'slor':
+        return '容認度 (SLOR)'
     else:
         return en
 
@@ -64,6 +72,11 @@ def show(self, lang: str='ja'):
             print('Unsupported language')
             sys.exit(1)
 
+
+def _f(score: float) -> str:
+    return 'None' if score is None else '{:.2f}'.format(score)
+
+
 def main():
 
     executables = ['juman', 'knp', 'rnnlm']
@@ -128,13 +141,14 @@ def main():
     Section('syntax', metrics).show()
 
     metrics = []
-    acceptability = \
-        Acceptability(text,
+    a = Acceptability(text,
                       args.delimiter,
                       args.rnnlm_model_path)
-    score = acceptability.mean_loglikelihood
-    score = 'None' if score is None else '{:.2f}'.format(score)
-    metrics.append(Metric('mean_loglikelihood', score))
+    metrics.append(Metric('log_prob', _f(a.log_prob)))
+    metrics.append(Metric('mean_lp', _f(a.mean_lp)))
+    metrics.append(Metric('norm_lp_div', _f(a.norm_lp_div)))
+    metrics.append(Metric('norm_lp_sub', _f(a.norm_lp_sub)))
+    metrics.append(Metric('slor', _f(a.slor)))
     Section('language_model', metrics).show()
 
 
diff --git a/trf/constant.pyc b/trf/constant.pyc
new file mode 100644
index 0000000..c71346c
Binary files /dev/null and b/trf/constant.pyc differ