Although it appears fairly innocuous, the last commit to tawny-owl seems momentus to me. While I still need to go through the spec line-by-line, and the code needs some clean up, this commit essentially represents the completion of the tawny.owl namespace; the addition of data properties and data types was the last part of the spec that I have to fulfil.

When I started off the tawny-owl library in October (http://www.russet.org.uk/blog/2214) I was most interested in getting a test environment, and the ability to use a normal editor. Subsequently, and particularly in the course of writing up my first paper on this library (http://www.russet.org.uk/blog/2366), it became obvious to me that I needed to support all of OWL2. I think I have achieved my original design motivations and some more besides. I have also learned a lot about OWL, the OWL API and Manchester syntax. It is also a strange project, because it is the first time I have fulfilled a specification in quite this way. I cannot recall the last time I could reasonably be said to have finished something, as research is generally open-ended.

I did not, however, start with a regular syntax in mind. In general, the conversion to lisp has worked reasonably well: the object side of OWL in particular falls into a prefix, lisp syntax very naturally; the individual side less so. The data side of OWL had another surprise in store for me: it looks very similar to the object side; so I wanted to share syntax. However, all the Java method calls are named differently and take different types and parameter number.

In the end, I have supported this through a multi-method and some heuristics to guess which call is wanted. For instance, with these two calls from the pizza ontology:

(owlsome hasTopping CheeseTopping)
(owlsome hasCalorifiContentValue (span =< 400))

we generate quite different types of OWL object. The owlsome method defers to either object-some or data-some respectively, which can also be used directly. In this case, the difference is obvious; however, tawny also takes strings in most of these places; in this case, we convert to an IRI and check whether it exists in the ontology or any ontology we know about first. I suspect that these heuristics will work in most cases, but fail in some; only time and experience will tell me about these.

Before the next release, 0.12, I will finish both the inline, function documentation and update the tutorial. After this I plan to sit on the API a while, think about the functions and the syntax to make sure I am happy; the release after should be 1.0 and as is the way of these things, I will be stuck with the apperance of the API for quite a while. This also allows me to avoid a 0.13 release without accusation of superstition.

There are still many parts of tawny that I wish to improve; in particular, I need to extend the repl facilities with doc and apropos features — my attempt to hijack the Clojure native facilities have failed despite extensive efforts. And explanation code needs to go in; currently, waiting for protege to reason and produce these results in a soul destroying experience; I want me continuous integration tests to automatically dump explanations whenever inconsistencies happen.

But new features are for the future; for this iteration, tawny-owl is finished and now will be polished.

Bibliography

Leave a Reply