I had the same error but nothing worked for me. So, I replaced ntlk.download('punkt') with ntlk.download('punkt_tab') and it worked. Actually "punkt_tab" is a newer, more accurate version of the same tokenizer.