In general, transform punctuation with >1 character to single-character

[kayaulai]

Perhaps use some list of ascii characters to transform them into?

[kayaulai]

My suggestion for this:

1. BEFORE getting the boundary lists, add all the non-boundary markers to the Utterance column. This means every line MUST end with a boundary marker.
2. Thereafter, we can assign any random Unicode character to correspond to the boundary types, since there is no risk of the final character not being a boundary marker.

[kayaulai]

我 説 了 三 句 <PERIOD>

<COMMA> > 逗
<PERIOD> > 句

我 説 了 三 句 句

transform back to (in the record)

我 説 了 三 句 <PERIOD>