Combining UML, XML and relational database technologies - the best of all
worlds for robust linguistic databases
Larry Hayashi and John Hatton
SIL International

This paper describes aspects of the data modeling, data storage, and
retrieval techniques we are using as we develop the FieldWorks suite of
applications for linguistic and anthropological research. While the data is
stored in an off-the-shelf relational database, the data modeling is done in the
Unified Modeling Language (UML) and data can be input and output using XML. Thus
Fieldworks benefits from both the maturity of its database engine and the
productivity of these formats.

The hierarchical nature of linguistic data lends itself to object-oriented
analysis (OOA) techniques. Fieldworks uses object-oriented data models
represented in UML. Over the last 2 decades some of the major OOA notation
methods have converged into UML. As a standard notation it allows for greater
communication between groups that want to share research. The XML Metadata
Interchange standard (XMI) - a document type definition for storing UML models
in XML format - allows groups to not only communicate using a common standard,
but to share data models with minimal re-implementation effort.

We create our linguistic data models in a UML tool which stores the model as
XMI. From the XMI file, we use standard XML transformations to generate the
following products:

* HTML documentation of the models,
* a database schema which is run through a C++ program to create a SQL
    Server relational database,
* an XML schema definition for validating XML data files to be imported
    into the generated database and
* an XDR schema for SQL Server that enables it to output relational data
    as hierarchical XML.

This XDR
schema has annotations describing the mapping between the XML elements and the
relational tables. After submitting this schema to the SQL Server, we can query
the database using either SQL or XPath
and get XML documents. We then use XSLT
style sheets to produce other data formats, including HTML for display and
Standard Format for input into our legacy parsing engine.

Our use of the mapping schema is still in experimental stages. Without it, we
have found it difficult to rapidly construct applications on top of the
FieldWorks system, in part because the developer must think and work at both the
relational and object-oriented levels simultaneously. Using the mapping schema
we can use Microsoft's SQL Server XML functionality to query the database
using XPath (rather than SQL) and generate hierarchical XML documents. The
advantage is that the developer does less translating between object-oriented
data and relational representation. The implication is, hopefully, faster
development time. Another further implication is the potential for an increased
interoperability between tools of different developers. A mapping schema could
be made for the FieldWorks linguistic tools that creates XML data usable by
other tools. Ideally, XML standards will be developed for lexicons and annotated
texts. Data could then be shared among different tools - much in the same way
that XMI allows UML data to be used in different modeling tools.

References:

Feldman, B. Unknown. UML FAQS. http://www.uml-zone.com/umlfaq.asp

Simons, Gary F. 1994. Conceptual modeling versus visual modeling: a
technological key to building consensus. SIL International.
http://www.sil.org/cellar/ach94/ach94.html

Simons, Gary F. 1995. Multilingual data processing in the CELLAR environment.
SIL International. http://www.sil.org/cellar/mlingdp/mlingdp.html

Web Resources:

Choosing a UML Tool. http://www.objectsbydesign.com/tools/modeling_tools.html

SQL Server and XML Support.
http://msdn.microsoft.com/library/default.asp?url=/library/en-us/architec/8_ar_cs_9oj8.asp

UML Specification.
http://www.omg.org/technology/documents/formal/omg_modeling_specifications.htm

XMI Specification.
http://www.omg.org/technology/documents/formal/omg_modeling_specifications.htm