diff --git a/docs/source/creating_and_sharing.mdx b/docs/source/creating_and_sharing.mdx
index c1b96d3fc..f09a34174 100644
--- a/docs/source/creating_and_sharing.mdx
+++ b/docs/source/creating_and_sharing.mdx
@@ -54,12 +54,12 @@ def _download_and_prepare(self, dl_manager):
     self.scorer = score.BleurtScorer(os.path.join(model_path, self.config_name))
 ```
 
-Or if you need to download the NLTK `"punkt"` resources:
+Or if you need to download the NLTK `"punkt_tab"` resources:
 
 ```py
 def _download_and_prepare(self, dl_manager):
     import nltk
-    nltk.download("punkt")
+    nltk.download("punkt_tab")
 ```
 
 Next, we need to define how the computation of the evaluation module works.
diff --git a/docs/source/transformers_integrations.mdx b/docs/source/transformers_integrations.mdx
index 7993b2f28..bca014187 100644
--- a/docs/source/transformers_integrations.mdx
+++ b/docs/source/transformers_integrations.mdx
@@ -81,7 +81,7 @@ def preprocess_function(examples):
 tokenized_billsum = billsum.map(preprocess_function, batched=True)
 
 # Setup evaluation
-nltk.download("punkt", quiet=True)
+nltk.download("punkt_tab", quiet=True)
 metric = evaluate.load("rouge")
 
 def compute_metrics(eval_preds):
diff --git a/measurements/word_length/word_length.py b/measurements/word_length/word_length.py
index 3448a679b..46a840879 100644
--- a/measurements/word_length/word_length.py
+++ b/measurements/word_length/word_length.py
@@ -16,10 +16,19 @@
 
 import datasets
 from nltk import word_tokenize
+from packaging import version
 
 import evaluate
 
 
+if evaluate.config.PY_VERSION < version.parse("3.8"):
+    import importlib_metadata
+else:
+    import importlib.metadata as importlib_metadata
+
+
+NLTK_VERSION = version.parse(importlib_metadata.version("nltk"))
+
 _DESCRIPTION = """
 Returns the average length (in terms of the number of words) of the input data.
 """
@@ -75,7 +84,10 @@ def _info(self):
     def _download_and_prepare(self, dl_manager):
         import nltk
 
-        nltk.download("punkt")
+        if NLTK_VERSION >= version.Version("3.9.0"):
+            nltk.download("punkt_tab")
+        else:
+            nltk.download("punkt")
 
     def _compute(self, data, tokenizer=word_tokenize):
         """Returns the average word length of the input data"""
diff --git a/metrics/meteor/meteor.py b/metrics/meteor/meteor.py
index 058ee80ed..bc023a8e2 100644
--- a/metrics/meteor/meteor.py
+++ b/metrics/meteor/meteor.py
@@ -120,7 +120,9 @@ def _download_and_prepare(self, dl_manager):
         import nltk
 
         nltk.download("wordnet")
-        if NLTK_VERSION >= version.Version("3.6.5"):
+        if NLTK_VERSION >= version.Version("3.9.0"):
+            nltk.download("punkt_tab")
+        elif NLTK_VERSION >= version.Version("3.6.5"):
             nltk.download("punkt")
         if NLTK_VERSION >= version.Version("3.6.6"):
             nltk.download("omw-1.4")
diff --git a/setup.py b/setup.py
index 5a97a847f..bf4680a8a 100644
--- a/setup.py
+++ b/setup.py
@@ -74,7 +74,7 @@
     "absl-py",
     "charcut>=1.1.1",  # for charcut_mt
     "cer>=1.2.0",  # for characTER
-    "nltk<3.9",  # for NIST and probably others; temporarily pin < 3.9 to avoid "Resource punkt_tab not found" (GH-622)
+    "nltk",  # for NIST and probably others
     "pytest",
     "pytest-datadir",
     "pytest-xdist",