I'd only like to suggest that a bit more "design" be put into the Relex API regarding Hobbs. Because somebody else might get caught in this too. RelexExtractor as an API seems to be pretty straightforward: configure it and then call process for each sentence. And it is bound to the rules defined in the text file. Until...oops, you realize something else is going on that eats up memory! I would put Hobbs in a separate class that decorates RelexExtractor with anaphora resolution (that'll make the anaphora algorithm pluggable). Or at least make it optional for RelexExtractor which is false by default and documentation on what happens if it's turned on :) At the very minimum, I'd put a check that monitors the number of sentences processed without calling 'clear' and throw an exception to warn the programmer.
I'm planning on redoing how the anaphora resolution thing works in ..
maybe a few weeks or month. I'm working on word-sense disambig
in opencog, and will want to change how hobbs works to be a bit
more consistent with that.