structure tree
The structure tree is the list, held inside a tagged PDF, that says what each part of the page is and what it belongs to.
In more detail
It is called a tree because the parts nest: a document holds sections, a section holds a heading and paragraphs, a table holds rows and a row holds cells. Every branch says what it is, and the leaves point at the marks on the page that belong to it.
Nothing of it shows. Two documents can look identical on screen and on paper while one carries a tree and the other carries none, and the difference only appears when software tries to make sense of the page.
What it does on a real document
Software reading a page aloud walks the tree rather than the ink, which is what gives it headings to jump between and a table it can read column heading by column heading. A program pulling figures back out does the same, and so does a search engine deciding what the document is about.
Much of the tree follows from what a page was asked to draw — a table knows its own rows, a paragraph knows where it ends — but nothing is marked of its own accord: every branch is one the document was told to write. What no program could work out from the ink is the outline: which block is a section, what a picture means, and in what order it all reads. That part is declared too.
Where it comes from, and why
It arrived with the tagged PDF in 2001, for two needs at once: reading a page aloud, and reflowing a page onto a small screen.
It is what the level of the archiving rules written for documents that must be read aloud asks for, and what the whole accessibility standard is built on. Both refuse a document that claims them and carries no tree.
See it on a finished document
A document that can be read aloud A list whose pictures speak too The page that goes further on this word