NLP start / end tagging

tl;dr For NLP, section tags are useful. Can end tags also be useful?

As discussed in Lesson 4, and later on in Lesson 10, for NLP it has been shown to be useful to include tags of various sorts to help the neural nets learn that there are differences between different sections of textual documents.

For example, labeling each text of an abstract from Arxiv with <CAT> for the category of the document, and <SUMM> for the summary section.

Are there any studies or general impressions about also including end tags? Is it generally harmful or even just redundant to also consider having </CAT> and </SUMM>? When a document only contains a category and a summary, it doesn’t strike me as helpful to include these tags, but with more complicated documents with sections and subsections, each of which may be of arbitrary and variable lengths, I can at least see that there may be some benefits in some situations.

Are there thoughts on this out there?

End tags are necessary when your markup language has arbitrary nesting. For example in HTML a table can contain another table, if you didn’t have end tags you wouldn’t know if it’s a sub-table or a whole new table. So I guess in NLP, in some situations they’d be necessary to get rid of ambiguity. Other than that, I’d say they’re spurious.