Hello!

Thanks for this suggestion!

Having numbered levels may result in some difficulties to insert any extra
step between two levels, but it seems there is a more important issue, there.

> indicate the union of those portions in Tennessee and North Carolina:
Here we have

@level="2" for Tennessee

and

@level="2" for North Carolina

and get the UNION of the elements at the same level. The following example
with Chestnut Hill works in a similar way. But when we reach the

> section of Mass Ave within the park
we have

@level="5" Lincoln Park

@level="5" Massachusetts Avenue

to get the INTERSECTION of the elements at the same level, which I consider
is not consistent with the preceding example. I consider that one question
here is whether we should rely on the order of the elements within
hierarchicalGeographic or if we should rely on their nesting (or both). One
suggestion would be to avoid the @level attributes and rely on the order of
the geographical elements:

hierarchicalGeographic

country United States /country

geographicUnion

state Tennessee /state

state North Carolina /state

/geographicUnion

placeOther placeType="national park" Smoky Mountain National Park /placeOther

/hierarchicalGeographic

thus adding the geographicUnion element when needed. In this case, every
child element of hierarchicalGeographic intersects with all its preceding
siblings (if any), whereas all the children of each geographicUnion element
are concatenated (gathered) with each other before their intersection with
the elements preceding this "geographicUnion" (if any) within the
hierarchicalGeographic element.

Of course, my suggestion assumes a context-free and unambiguous
identification of the first child of the hierarchicalGeographic element or
(in cases where this first child is a geographicUnion element) each child of
the first child of the hierarchicalGeographic element.

Each of the non-first children of the hierarchicalGeographic element has to
be (entirely or partially) geographically included in (and unambiguously
identified within) its immediately preceding sibling.

This assumes that every geographicUnion element is geographically
unambiguously identified by its children (within the context of its
immediately preceding sibling if any).

An other (somewhat similar) alternative suggestion would be to use a
geographicIntersection too and thus relax a few elements' by-order
identification, but I consider that this would not make things really easier.

As an other alternative suggestion, we could relax the ordering
requirements, but this would require an improved identification strategy for
the geographic elements (which in turn could reduce the number of children
of the hierarchicalGeographic element). I don't consider that this would
make things really easier either (and it would not improve human
readability), though.

I did not take into account any comparison of the processing speeds for
these different alternative suggestions.

Regards!

Saašha,