Compatibility with current semistructured models is achieved by assuming that the string labels in those models default to name properties. Hence our framework can represent any existing semistructured database by modeling it as a database in which every label contains exactly one name property.
Using the same default, retrieval queries also remain unchanged. In existing semistructured databases all paths are valid. In our framework, if every label consists of a single name property, then all paths are also valid (names are collapsed using string concatenation, which never results in an undefined value). In existing semistructured databases, the labels are matched using string comparison, just like in our framework, so path regular expressions match exactly the same paths in both models.
Finally, we observe that our framework seamlessly supports the mixing of data from existing semistructures with data that has richer meta-data since properties can vary from label to label. Hence as much or as little data as desired can be migrated to use the new type of labels.