Jeni's Musings

Just a couple of clarifications:

NXT’s data model doesn’t allow for any constraints that would require cross-hierarchy validation - if there are two hierarchies that draw on the same base layer of atoms, they are completely independent (they don’t even necessarily draw the atoms in the same order). We’ve never run into this sort of constraint as a requirement from our users (which isn’t to say there are no reasons for ever wanting this, just that our users’ needs are limited). Otherwise, validation is as you suggest - we create a set of XML trees that together cover all of the relationships expressed in our overall data graph, where the same node can occur redundantly in more than one tree, and validate each one separately (using a single schema that can cover any of them). This is simple-minded, but inefficient.

Many of our example uses are fairly flat because users come to NXT not just because of the overlapping hierarchies but also because we support multimodal annotation. However, the data model does support heavily structured documents, and there are some corpora that make use of that, with perhaps four different trees built on top of the same orthographic transcription, and some trees that decompose recursively through the same layer of tags. We don’t allow self-overlap for the same tag, though.