diff --git a/server/data.py b/server/data.py
index fdcf317..707a220 100644
--- a/server/data.py
+++ b/server/data.py
@@ -30,7 +30,7 @@ def sent_2_words(sent: str) -> List[str]:
     return words
 
 
-def corpus_to_words(file_path: str, vocab_path: str = str(VOCAB_PATH)):
+def corpus_to_words(file_path: str):
     my_counter: Counter = Counter()
     with open(file_path, "r", encoding="utf-8") as fl:
         for sent in tqdm(fl, desc="Precess file"):
@@ -40,7 +40,10 @@ def corpus_to_words(file_path: str, vocab_path: str = str(VOCAB_PATH)):
     min_count = max([10, max_cnt / 100])
 
     selected_words = [word for word, count in my_counter.items() if (min_count < count <= max_cnt)]
+    return selected_words
 
+
+def save_words(selected_words, vocab_path: str = str(VOCAB_PATH)):
     with open(vocab_path, "w", encoding="utf-8") as fl:
         for w in selected_words:
             fl.write(w + "\n")
@@ -179,7 +182,8 @@ def add_new_text(bucket_name=BUCKET_SPLIT_TEXTS, destination_bucket_name=BUCKET_
     to_copy = sorted(set(all_files) - set(copied_files))[0]
     print(to_copy)
     download_blob(bucket_name, to_copy, DATA_PATH / to_copy)
-    corpus_to_words(DATA_PATH / to_copy)
+    words = corpus_to_words(DATA_PATH / to_copy)
+    save_words(words)
     copy_blob(bucket_name, to_copy, destination_bucket_name, to_copy)
 
 
diff --git a/the_hat_game/data.py b/the_hat_game/data.py
new file mode 100644
index 0000000..28ae721
--- /dev/null
+++ b/the_hat_game/data.py
@@ -0,0 +1,32 @@
+import re
+from collections import Counter
+from typing import List
+
+from nltk.corpus import stopwords
+from nltk.stem import WordNetLemmatizer
+from nltk.tokenize import word_tokenize
+from tqdm.auto import tqdm
+
+STOP_WORDS = stopwords.words("english")
+LEMMATIZER = WordNetLemmatizer()
+
+
+def sent_2_words(sent: str) -> List[str]:
+    sent = sent.lower()
+    sent = re.sub("[^a-z]+", " ", sent)
+    words = word_tokenize(sent)
+    words = [LEMMATIZER.lemmatize(word) for word in words if ((word not in STOP_WORDS) and len(word.strip()) > 3)]
+    return words
+
+
+def corpus_to_words(file_path: str):
+    my_counter: Counter = Counter()
+    with open(file_path, "r", encoding="utf-8") as fl:
+        for sent in tqdm(fl, desc="Precess file"):
+            my_counter.update(sent_2_words(sent))
+
+    max_cnt = max(count for word, count in my_counter.items()) / 10
+    min_count = max([10, max_cnt / 100])
+
+    selected_words = [word for word, count in my_counter.items() if (min_count < count <= max_cnt)]
+    return selected_words
diff --git a/the_hat_game/game.py b/the_hat_game/game.py
index 008ad9f..3208a58 100644
--- a/the_hat_game/game.py
+++ b/the_hat_game/game.py
@@ -13,6 +13,7 @@
 import the_hat_game.nltk_setup  # noqa: F401
 from the_hat_game.loggers import c_handler, logger
 from the_hat_game.players import RemotePlayer
+from the_hat_game.data import corpus_to_words
 
 
 class Game:
@@ -24,8 +25,23 @@ def __init__(
         n_rounds,
         n_explain_words,
         n_guessing_words,
+        corpus_path=None,
+        vocab_path=None,
         random_state=None,
     ):
+        """Main class for Game.
+        params:
+            - players: list of AbstractPlayer - players in the game
+            - words: list of str - all used words for guessing
+            - criteria: 'hard' of 'soft' - game criteria
+            - n_rounds: int - number of rounds
+            - n_explain_words: int - number of words for explaining
+            - n_guessing_words: int - number of words for guessing
+            - corpus_path: str - path for the corpus to create vocabulary (for criteria='hard')
+            - vocab_path: str - path for vocabulary (for criteria='hard')
+            NOTE: only corpus_path or vocab_path must be defined
+            NOTE: if vocabulary is not defined nltk.wordnet will be used for filter not existing words
+        """
         assert len(players) >= 2
         assert criteria in ("hard", "soft")
         self.players = players
@@ -36,6 +52,21 @@ def __init__(
         self.n_guessing_words = n_guessing_words
         self.random_state = random_state
         self.stemmer = SnowballStemmer("english")
+        if corpus_path is not None:
+            assert vocab_path is None, "corpus and vocabulary cannot be defined at the same time"
+            self.whitelist = corpus_to_words(corpus_path)
+        elif vocab_path is not None:
+            with open(vocab_path, encoding="utf-8") as f:
+                vocab_words = f.readlines()
+                self.whitelist = [word.strip() for word in vocab_words]
+        else:
+            self.whitelist = None
+        # add all words for guessing to non-empty whitelist
+        if self.whitelist is not None:
+            self.whitelist += [
+                word for word in self.words
+                if word not in self.whitelist
+            ]
 
     def remove_repeated_words(self, words):
         unique_words = []
@@ -54,9 +85,11 @@ def remove_same_rooted_words(self, word, word_list):
         cleared_word_list = [w for w in word_list if self.stemmer.stem(w) != root]
         return cleared_word_list
 
-    @staticmethod
-    def remove_non_existing_words(words):
-        existing_words = [w for w in words if len(wordnet.synsets(w)) > 0]
+    def remove_non_existing_words(self, words):
+        if self.whitelist is not None:
+            existing_words = [w for w in words if w in self.whitelist]
+        else:
+            existing_words = [w for w in words if len(wordnet.synsets(w)) > 0]
         return existing_words
 
     def create_word_list(self, player, word, n_words):