"Experiments in overlap: lessons from implementing the LMNL model in XML"

In many ways, the idea of implementing LMNL in XML is backwards. The
whole idea of LMNL, a data model for describing marked up text, is to
transcend limitations in XML by allowing arbitrary overlap of marked up
ranges of text, and annotations to markup (the LMNL equivalent of XML
attributes) to have structure. Yet two factors make the idea of
implementing LMNL in XML compelling: 1) XML tools are powerful, flexible
and ubiquitous, and 2) implementing pathways to get LMNL data out of and
into XML (a natural adjunct of this activity) allow us to interface with
XML datasets, which further takes advantage of strengths in XML (many
kinds of operations are easy; data is available).

Conventions for representing LMNL data using XML markup, the ECLIX and
CLIX formats, provide a basis for this work. Any XML representation of
overlap can be transformed into ECLIX, which is basically XML
supplemented with a milestone representation of any overlapping
structures that the XML cannot represent. ECLIX can be trivially (and
generically) translated into CLIX, and thence into an XML "textbase"
(called xLMNL) which uses standoff structures to represent the ranges
marked over the text. This format in turn provides for further
processing, including rendition back into XML, presentation (for
example, diagrams in SVG) or heuristic analysis. Processing for these
transformations can be deployed on any XML pipelining platform, such as
Apache Cocoon or an XProc implementation.

Strengths, difficulties and limitations of this approach will be discussed.

A (static) copy of a demonstration may be seen at
http://piez.org/wendell/dh2010/clix-sonnets/index.html.

More information about LMNL is at
http://www.lmnl.org/wiki/index.php/Main_Page.