diff --git a/spacy/ml/models/coref_util.py b/spacy/ml/models/coref_util.py
index 38af629d5..3be0bd835 100644
--- a/spacy/ml/models/coref_util.py
+++ b/spacy/ml/models/coref_util.py
@@ -205,7 +205,7 @@ def create_gold_scores(
     return out
 
 
-def _spans_to_offsets(doc):
+def _spans_to_offsets(doc: Doc) -> List[List[Tuple[int, int]]]:
     """Convert doc.spans to nested list of ints for comparison.
     The ints are character indices, and the spans groups are sorted by key first.
 
diff --git a/spacy/tests/pipeline/test_span_predictor.py b/spacy/tests/pipeline/test_span_predictor.py
index a79756d88..c0e59e914 100644
--- a/spacy/tests/pipeline/test_span_predictor.py
+++ b/spacy/tests/pipeline/test_span_predictor.py
@@ -154,7 +154,7 @@ def test_tokenization_mismatch(nlp):
         for key, cluster in ref.spans.items():
             char_spans[key] = []
             for span in cluster:
-                char_spans[key].append((span[0].idx, span[-1].idx + len(span[-1])))
+                char_spans[key].append((span.start_char, span.end_char))
         with ref.retokenize() as retokenizer:
             # merge "picked up"
             retokenizer.merge(ref[2:4])