BLEU: support tokenizer string alias and clarify input edge cases

metrics/bleu/bleu.py

@@ -60,6 +60,7 @@
     references: list of lists of or just a list of references for each translation.
     tokenizer : approach used for tokenizing `predictions` and `references`.
         The default tokenizer is `tokenizer_13a`, a minimal tokenization approach that is equivalent to `mteval-v13a`, used by WMT.
+        You can also pass tokenizer="13a" to use the default WMT tokenizer.
         This can be replaced by any function that takes a string as input and returns a list of tokens as output.
     max_order: Maximum n-gram order to use when computing BLEU score.
     smooth: Whether or not to apply Lin et al. 2004 smoothing.
@@ -112,8 +113,27 @@ def _info(self):
             ],
         )
 
-    def _compute(self, predictions, references, tokenizer=Tokenizer13a(), max_order=4, smooth=False):
+    def _compute(self, predictions, references, tokenizer=None, max_order=4, smooth=False):
         # if only one reference is provided make sure we still use list of lists
+        if len(predictions) != len(references):
+            raise ValueError(
+                f"Predictions and references must have the same length, got {len(predictions)} and {len(references)}."
+            )
+        if len(predictions) == 0:
+            raise ValueError("Predictions and references must be non-empty.")
+
+        if tokenizer is None:
+            tokenizer = Tokenizer13a() # Prevent instantiated defaults to follow best practices
+
+        if isinstance(tokenizer, str):
+            tok = tokenizer.strip().lower()
+            if tok in {"13a", "tokenizer_13a", "mteval-v13a"}:
+                tokenizer = Tokenizer13a()
+            else:
+                raise ValueError(
+                    f"Unknown tokenizer '{tokenizer}'. Supported string tokenizers: '13a'. Otherwise pass a callable."
+                )
+
         if isinstance(references[0], str):
             references = [[ref] for ref in references]