A user mentioned that it would be useful to, in addition to text spans, to be able to have the actual text which triggered extraction. This has a number of uses, and I'll integrate it into the Python version for the next release.