Tree Annotation Format for RRG Syntactic Structures

Since the standard data structure for treebank annotations is trees, specifically, a single tree per sentence, we adopt a notational variant for the RRG structures (exemplified in the right half of the picture below).

It departs from the usual representation of RRG structures (see the left half of the figure) in several ways:

• Since the spine of the operator projection mirrors the spine of the constituent projection, we identify the corresponding nodes (for example, CLAUSE, CORE, NUC and V) and attach operators in the same tree as other constituents.
• We represent arrows as ordinary edges (and eliminate PERIPHERY nodes). The roots of operators, peripheries and clause linkage markers become daughters of the noes they attach to (see the TNS, CLM and AP nodes in the example).
• In order to still distinguish operators and peripheries, we decorate the labels of their roots with -OP and -PERI, respectively. Clause linkage markers are already distinguished by the root label CLM.