When individual sentences are summarized, decoding from tokens should apply the `skip_special_tokens` parameter to limit output to English.