add SacreBLEU like soothing precision

chinesebleu.py

@@ -46,16 +46,17 @@ Returns:
     score: the Chinese BLEU score,
     counts: Counts in n-gram (1-4 grams),
     totals: Totals in n-gram,
-    bp: Brevity Penalty
+    bp: Brevity Penalty,
+    tokenizer: Selection of Tokenizer (either "chinese" or "char")
     
 Examples:
     Examples should be written in doctest format, and should illustrate how
     to use the function.
 
     >>> my_new_module = evaluate.load("chinesebleu")
-    >>> results = my_new_module.compute(references=["Reference Sentence in Chinese"], predictions=["Predicted Sentence in Chinese"])
+    >>> results = my_new_module.compute(references=["這裡就是香港都會大學"], predictions=["這裡是香港都會大學"])
     >>> print(results)
-    {'score': 71.89393375176813, 'counts': [9, 7, 5, 4], 'totals': [9, 8, 7, 6], 'bp': 1.0}
+    {'score': 71.89393375176813, 'counts': [9, 7, 5, 4], 'totals': [9, 8, 7, 6], 'bp': 1.0, tokenizer: 'chinese'}
 """
 
 @evaluate.utils.file_utils.add_start_docstrings(_DESCRIPTION, _KWARGS_DESCRIPTION)
@@ -72,8 +73,8 @@ class ChineseBLEU(evaluate.Metric):
                 'predictions': datasets.Value('string'),
                 'references': datasets.Value('string'),
             }),
-            homepage="https://github.com/shivanraptor/chinesebleu",
-            codebase_urls=["https://github.com/shivanraptor/chinesebleu"]
+            homepage="https://huggingface.co/spaces/raptorkwok/chinesebleu/",
+            codebase_urls=["https://huggingface.co/spaces/raptorkwok/chinesebleu/"]
         )
 
     def _download_and_prepare(self, dl_manager):
@@ -140,7 +141,7 @@ class ChineseBLEU(evaluate.Metric):
 
         # For total number of tokens < 4, fallback to character-level tokenizations
         if len(pred_tokens) < 4 or len(ref_tokens) < 4:
-            tokenizer = 'character'
+            tokenizer = 'char'
             pred_tokens = [self._tokenize_chinese(p, tokenizer) for p in predictions]
             ref_tokens = [self._tokenize_chinese(r, tokenizer) for r in references]
     
@@ -170,11 +171,15 @@ class ChineseBLEU(evaluate.Metric):
         
         # Compute precisions
         precisions = []
-        for c, t in zip(counts, totals):
+        for i, (c, t) in enumerate(zip(counts, totals)):
             if t == 0:
                 precisions.append(0.0)
             else:
-                precisions.append(float(c) / t)
+                # SacreBLEU floor smoothing for n=4 on short sentences (no matches, total=1)
+                p = float(c) / t
+                if i == 3 and c == 0 and t == 1: # i=3 is 4-gram (0-indexed)
+                    p = 1.0 / 2.0  # 0.5, matching SacreBLEU's heuristic
+                precisions.append(p)
         
         # Geometric mean of precisions
         if any(p == 0 for p in precisions):