The current method of getting the text nodes related to an utterance uses a path made up of indices, e.g. [1, 0, 3]. it looks like i should be possible to replaced this by XPath-expressions, with all the benefits of using a standard implementation.
So lets do that.