diff --git a/src/lighteval/metrics/metrics_corpus.py b/src/lighteval/metrics/metrics_corpus.py
index 92c2c574a..d88a3adda 100644
--- a/src/lighteval/metrics/metrics_corpus.py
+++ b/src/lighteval/metrics/metrics_corpus.py
@@ -28,6 +28,7 @@
 import logging
 import math
 from abc import ABC, abstractmethod
+from itertools import zip_longest
 from typing import Literal
 
 import numpy as np
@@ -142,7 +143,7 @@ def get_metric(self):
     def compute_corpus(self, items: list[GenerativeCorpusMetricInput]) -> float:
         """Computes the metric score over all the corpus generated items, by using the sacrebleu implementation."""
         metric = self.get_metric()
-        golds = [i.golds for i in items]
+        golds = [as_list(i.golds) for i in items]
         preds = []
         for i in items:
             pred = as_list(i.preds)
@@ -153,9 +154,12 @@ def compute_corpus(self, items: list[GenerativeCorpusMetricInput]) -> float:
             preds.append(pred[0])
 
         if self.metric_type == "bleu":
-            golds = [[gold[0] for gold in golds]]
+            references = [[gold[0] for gold in golds]]
+        else:
+            # SacreBLEU expects references as [reference_id][sample_id].
+            references = [list(ref_group) for ref_group in zip_longest(*golds, fillvalue=None)]
 
-        corpus_score = metric.corpus_score(hypotheses=preds, references=golds)
+        corpus_score = metric.corpus_score(hypotheses=preds, references=references)
         score = corpus_score.score
         results = float(score)
         return results
diff --git a/tests/unit/metrics/test_cases/chrf.json b/tests/unit/metrics/test_cases/chrf.json
index f55028674..1d11195d8 100644
--- a/tests/unit/metrics/test_cases/chrf.json
+++ b/tests/unit/metrics/test_cases/chrf.json
@@ -79,9 +79,9 @@
           "text": ["Das Wetter ist schön"]
         }
       ],
-      "expected_output": 100.0,
+      "expected_output": 89.1577277178731,
       "tolerance": 0.1,
-      "description": "High similarity - minor character differences (CHRF ≈ 88.0)"
+      "description": "High similarity - minor character differences (CHRF ≈ 89.16)"
     },
     {
       "name": "CHRF - Word Order Changes",
@@ -119,9 +119,9 @@
           "text": ["Lernen Maschinelles"]
         }
       ],
-      "expected_output": 78.84,
+      "expected_output": 82.21013454114589,
       "tolerance": 0.1,
-      "description": "Word order changes - same characters, different order (CHRF ≈ 75.0)"
+      "description": "Word order changes - same characters, different order (CHRF ≈ 82.21)"
     },
     {
       "name": "CHRF - Moderate Similarity",
@@ -159,9 +159,9 @@
           "text": ["Die Sterne"]
         }
       ],
-      "expected_output": 37.68,
+      "expected_output": 34.847705825542,
       "tolerance": 0.1,
-      "description": "Moderate similarity - partial character overlap (CHRF ≈ 50.0)"
+      "description": "Moderate similarity - partial character overlap (CHRF ≈ 34.85)"
     },
     {
       "name": "CHRF - Low Similarity",
@@ -199,9 +199,9 @@
           "text": ["Es sehr heiß"]
         }
       ],
-      "expected_output": 7.7,
+      "expected_output": 6.82996024778865,
       "tolerance": 0.1,
-      "description": "Low similarity - minimal character overlap (CHRF ≈ 20.0)"
+      "description": "Low similarity - minimal character overlap (CHRF ≈ 6.83)"
     }
   ]
 }
diff --git a/tests/unit/metrics/test_cases/chrf_plus.json b/tests/unit/metrics/test_cases/chrf_plus.json
index 29c45720d..89061449e 100644
--- a/tests/unit/metrics/test_cases/chrf_plus.json
+++ b/tests/unit/metrics/test_cases/chrf_plus.json
@@ -79,9 +79,9 @@
           "text": ["Das Wetter ist schön"]
         }
       ],
-      "expected_output": 100.0,
+      "expected_output": 88.06512778139042,
       "tolerance": 0.1,
-      "description": "High similarity - minor character differences (CHRF++ ≈ 85.0)"
+      "description": "High similarity - minor character differences (CHRF++ ≈ 88.07)"
     },
     {
       "name": "CHRF Plus - Moderate Similarity",
@@ -119,9 +119,9 @@
           "text": ["ML"]
         }
       ],
-      "expected_output": 58.82,
+      "expected_output": 24.450573381865112,
       "tolerance": 0.1,
-      "description": "Moderate similarity - significant character omissions (CHRF++ ≈ 45.0)"
+      "description": "Moderate similarity - significant character omissions (CHRF++ ≈ 24.45)"
     },
     {
       "name": "CHRF Plus - Low Similarity",
@@ -159,9 +159,9 @@
           "text": ["Es heiß"]
         }
       ],
-      "expected_output": 15.0,
-      "tolerance": 10.0,
-      "description": "Low similarity - minimal character overlap (CHRF++ ≈ 15.0)"
+      "expected_output": 2.733143823910822,
+      "tolerance": 0.1,
+      "description": "Low similarity - minimal character overlap (CHRF++ ≈ 2.73)"
     }
   ]
 }
diff --git a/tests/unit/metrics/test_cases/ter.json b/tests/unit/metrics/test_cases/ter.json
index 39b671b0f..2986df5c7 100644
--- a/tests/unit/metrics/test_cases/ter.json
+++ b/tests/unit/metrics/test_cases/ter.json
@@ -79,9 +79,9 @@
           "text": ["Das Wetter ist schön"]
         }
       ],
-      "expected_output": 0.0,
+      "expected_output": 9.090909090909092,
       "tolerance": 0.05,
-      "description": "Minor edits - small word differences"
+      "description": "Minor edits - small word differences (TER ≈ 9.09)"
     },
     {
       "name": "TER - Major Edits",
@@ -159,9 +159,9 @@
           "text": ["Es ist sehr heiß"]
         }
       ],
-      "expected_output": 80.0,
+      "expected_output": 100.0,
       "tolerance": 0.1,
-      "description": "Completely different translations - maximum edit distance"
+      "description": "Completely different translations - maximum edit distance (TER = 100.0)"
     }
   ]
 }
diff --git a/tests/unit/metrics/test_corpus_translation_metrics.py b/tests/unit/metrics/test_corpus_translation_metrics.py
new file mode 100644
index 000000000..cfada215f
--- /dev/null
+++ b/tests/unit/metrics/test_corpus_translation_metrics.py
@@ -0,0 +1,75 @@
+# MIT License
+#
+# Copyright (c) 2024 The HuggingFace Team
+#
+# Permission is hereby granted, free of charge, to any person obtaining a copy
+# of this software and associated documentation files (the "Software"), to deal
+# in the Software without restriction, including without limitation the rights
+# to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
+# copies of the Software, and to permit persons to whom the Software is
+# furnished to do so, subject to the following conditions:
+#
+# The above copyright notice and this permission notice shall be included in all
+# copies or substantial portions of the Software.
+#
+# THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
+# IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
+# FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
+# AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
+# LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
+# OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
+# SOFTWARE.
+
+from itertools import zip_longest
+
+import pytest
+
+from lighteval.metrics.metrics_corpus import CorpusLevelTranslationMetric
+from lighteval.metrics.sample_preparator import GenerativeCorpusMetricInput
+from lighteval.utils.utils import as_list
+
+
+def _transpose_references(items: list[GenerativeCorpusMetricInput]) -> list[list[str | None]]:
+    per_sample_references = [as_list(item.golds) for item in items]
+    return [list(ref_group) for ref_group in zip_longest(*per_sample_references, fillvalue=None)]
+
+
+def _first_prediction_per_sample(items: list[GenerativeCorpusMetricInput]) -> list[str]:
+    return [as_list(item.preds)[0] for item in items]
+
+
+@pytest.mark.parametrize("metric_type", ["chrf", "chrf++", "ter"])
+def test_translation_metrics_use_all_hypotheses(metric_type: str):
+    items = [
+        GenerativeCorpusMetricInput(golds=["GOOD"], preds=["GOOD"]),
+        GenerativeCorpusMetricInput(golds=["REF2"], preds=["PRED2"]),
+    ]
+    metric = CorpusLevelTranslationMetric(metric_type=metric_type)
+
+    hypotheses = _first_prediction_per_sample(items)
+    expected_references = _transpose_references(items)
+    wrong_orientation_references = [item.golds for item in items]
+
+    expected_score = metric.get_metric().corpus_score(hypotheses=hypotheses, references=expected_references).score
+    wrong_score = (
+        metric.get_metric().corpus_score(hypotheses=hypotheses, references=wrong_orientation_references).score
+    )
+    actual_score = metric.compute_corpus(items)
+
+    assert actual_score == pytest.approx(expected_score)
+    assert wrong_score != pytest.approx(expected_score)
+
+
+@pytest.mark.parametrize("metric_type", ["chrf", "chrf++", "ter"])
+def test_translation_metrics_support_variable_reference_counts(metric_type: str):
+    items = [
+        GenerativeCorpusMetricInput(golds=["the cat sits", "cat is sitting"], preds=["the cat sits"]),
+        GenerativeCorpusMetricInput(golds=["goodbye"], preds=["hello"]),
+    ]
+    metric = CorpusLevelTranslationMetric(metric_type=metric_type)
+
+    hypotheses = _first_prediction_per_sample(items)
+    references = _transpose_references(items)
+    expected_score = metric.get_metric().corpus_score(hypotheses=hypotheses, references=references).score
+
+    assert metric.compute_corpus(items) == pytest.approx(expected_score)