fix bleu bug (#5073)

* fix bleu bug

PaddleNLP/paddlenlp/metrics/bleu.py

@@ -28,15 +28,15 @@ def get_match_size(cand_ngram, refs_ngram):
     for ref_ngram in refs_ngram:
         tmp_ref_set = defaultdict(int)
         for ngram in ref_ngram:
-            tmp_ref_set[ngram] += tmp_ref_set.get(ngram, 0) + 1
+            tmp_ref_set[tuple(ngram)] += 1
         for ngram, count in tmp_ref_set.items():
-            ref_set[ngram] = max(ref_set[ngram], count)
+            ref_set[tuple(ngram)] = max(ref_set[tuple(ngram)], count)
     cand_set = defaultdict(int)
     for ngram in cand_ngram:
-        cand_set[ngram] += 1
+        cand_set[tuple(ngram)] += 1
     match_size = 0
     for ngram, count in cand_set.items():
-        match_size += min(count, ref_set.get(ngram, 0))
+        match_size += min(count, ref_set.get(tuple(ngram), 0))
     cand_size = len(cand_ngram)
     return match_size, cand_size
 
@@ -101,10 +101,10 @@ class BLEU(paddle.metric.Metric):
         .. code-block:: python
             from paddlenlp.metrics import BLEU
             bleu = BLEU()
-            cand = "Welcome to PaddleNLP."
-            ref_list = ["Welcome PaddleNLP"]
+            cand = ["The","cat","The","cat","on","the","mat"]
+            ref_list = [["The","cat","is","on","the","mat"],["There","is","a","cat","on","the","mat"]]
             bleu.add_inst(cand, ref_list)
-            print(bleu.score()) # 0.7510186074254295
+            print(bleu.score()) # 0.4671379777282001
 
         2. Using as an instance of `paddle.metric.Metric`.
                 
@@ -156,7 +156,7 @@ class BLEU(paddle.metric.Metric):
         Update the states based on the a pair of candidate and references.
 
         Args:
-            cand (str): The candidate sentence generated by model.
+            cand (list): Tokenized candidate sentence generated by model.
             ref_list (list): List of ground truth sentences.
         '''
         for n_size in range(self.n_size):

PaddleNLP/paddlenlp/metrics/utils.py

@@ -26,8 +26,8 @@ def default_trans_func(output, label, seq_mask, vocab):
             if seq_mask[i][j][0] == 0:
                 break
             token_list.append(vocab[idx[i][j]])
-        token_str = " ".join(token_list)
-        ref_list.append([token_str])
+
+        ref_list.append([token_list])
 
     label = np.squeeze(label, axis=2)
     for i in range(label.shape[0]):
@@ -36,6 +36,6 @@ def default_trans_func(output, label, seq_mask, vocab):
             if seq_mask[i][j][0] == 0:
                 break
             token_list.append(vocab[label[i][j]])
-        token_str = " ".join(token_list)
-        cand.append(token_str)
+
+        cand.append(token_list)
     return cand, ref_list