#79 Textual content should perhaps percolate

[J Carletta]

The early NXT design documents suggest that textual
content should percolate up data element hierarchies -
that is, the textual content for an element should
be a concatenation of the textual content of its
children in order, presumably separated by a
configurable delimiter. This is one reason why
the NXT data model prohibits mixed content (the same
element having both children and text). Text percolation
would be useful for doing regular expressions over
phrases such as named entities; without it, we can
only search using regular expressions over individual
tokens (e.g., words) contained in the phrase. However,
it could also be confusing for some data uses - for
instance, with it, it would be more difficult with
some data set designs to render transcription/text
without accidentally doubling parts of it.

We've had our first enquiry about why we don't
have text percolation. Is this something we should
implement, or should we just make sure the documentation
doesn't include it?

[Jonathan Kilgour]

Our current thinking is that we should not switch this on by default but allow it via parameter or environment variable. in this way percolation of text is not permanent and can be controlled simply.

Details such as whether only to percolate text in circumstances where time would also be percolated are to be worked out.