diff --git a/flair/models/entity_linker_model.py b/flair/models/entity_linker_model.py
index 306a4318e4..1a29c9e2fc 100644
--- a/flair/models/entity_linker_model.py
+++ b/flair/models/entity_linker_model.py
@@ -91,7 +91,7 @@ def _filter_data_point(self, data_point: Sentence) -> bool:
         return bool(data_point.get_labels(self.label_type))
 
     def _embed_prediction_data_point(self, prediction_data_point: Span) -> torch.Tensor:
-        return self.aggregated_embedding(prediction_data_point, self.word_embeddings.get_names()).unsqueeze(0)
+        return self.aggregated_embedding(prediction_data_point, self.word_embeddings.get_names())
 
     def _get_state_dict(self):
         model_state = {
diff --git a/tests/test_entity_linker.py b/tests/test_entity_linker.py
index 54fee84398..72933b8c5f 100644
--- a/tests/test_entity_linker.py
+++ b/tests/test_entity_linker.py
@@ -10,3 +10,15 @@ def test_entity_linker_with_no_candidates():
 
     sentence = Sentence("I live in Berlin")
     linker.predict(sentence)
+
+
+def test_forward_loss():
+    sentence = Sentence("I love NYC and hate OYC")
+    sentence[2:3].add_label("nel", "New York City")
+    sentence[5:6].add_label("nel", "Old York City")
+
+    # init tagger and do a forward pass
+    tagger = EntityLinker(TransformerWordEmbeddings("distilbert-base-uncased"), label_dictionary=Dictionary())
+    loss, count = tagger.forward_loss([sentence])
+    assert count == 1
+    assert loss.size() == ()