BlenderbotSmallTokenizer throws tuple index out of range error for stopword

Using transformers==3.4.0

Script used:
```

from transformers import BlenderbotSmallTokenizer, BlenderbotForConditionalGeneration
mname = 'facebook/blenderbot-90M'
tokenizer = BlenderbotSmallTokenizer.from_pretrained(mname)

sentence = "."
tokenizer(sentence)['input_ids']
```

This throws `IndexError: tuple index out of range`