diff --git a/spacy/pipeline/edit_tree_lemmatizer.py b/spacy/pipeline/edit_tree_lemmatizer.py
index d8279da31..b19511790 100644
--- a/spacy/pipeline/edit_tree_lemmatizer.py
+++ b/spacy/pipeline/edit_tree_lemmatizer.py
@@ -137,15 +137,15 @@ class EditTreeLemmatizer(TrainablePipe):
             ):
                 if gold_lemma is None:
                     label = -1
-                    eg_lowercasing_truths.append([0])
+                    eg_lowercasing_truths.append([0.0])
                 else:
                     if self.lowercasing and _should_lowercased(
                         predicted.text, gold_lemma
                     ):
-                        eg_lowercasing_truths.append([1])
+                        eg_lowercasing_truths.append([1.0])
                         text = predicted.lower_
                     else:
-                        eg_lowercasing_truths.append([0])
+                        eg_lowercasing_truths.append([0.0])
                         text = predicted.text
                     tree_id = self.trees.add(text, gold_lemma)
                     label = self.tree2label.get(tree_id, 0)
@@ -163,7 +163,8 @@ class EditTreeLemmatizer(TrainablePipe):
             for i, doc_d_tree_scores in enumerate(d_tree_scores):
                 eg_lowercasing_flags = lowercasing_flags[i]
                 eg_d_lowercasing_flags, eg_lowercasing_loss = lowercasing_loss_func(
-                    eg_lowercasing_flags, self.model.ops.asarray2i(lowercasing_truths[i])
+                    eg_lowercasing_flags,
+                    self.model.ops.asarray2f(lowercasing_truths[i]),  # type: ignore[arg-type]
                 )
                 doc_d_scores = self.model.ops.xp.hstack(
                     [