I have a question regarding the logic in utils/chair.py, specifically around lines 128 and 129.

I noticed a potential index misalignment issue. The idxs appear to be generated based on the sentence's original structure. However, these indices are then used to look up words in raw_words. The raw_words list has already been processed to handle double_words, which can alter its length and the positions of subsequent words.

My concern is that if the sentence contains double_words, the idxs may no longer correspond to the correct word positions in the modified raw_words list, leading to a dislocation.

Could you please clarify if this is the intended behavior or if I might be misunderstanding a previous step? Thank you!