Fftb error with "Is a constituent"

Dear all, and especially @sweaglesw some Treebankers here at UW are running into an error when they have the “is a constituent” filter on. Specifically, here is the reported behavior and workaround:

" When trying to save after applying “is a constituent” filter (which shows up as “span = [x] to [y]”), the terminal responds to prompt /save?{number of parse}&accepted=1 with segmentation fault (core dumped) (errored) as opposed to writing tsdb relation 'decision' with [#] tuples (in case of successfully accepting the tree). It will not allow me to go back or exit through the browser, giving error unexpected http code: 0= , hinting that the treebanker itself has stopped running and needs to be restarted. After re-running the treebanker, I see that progresses on that particular sentence was not saved, and other filters will have to be reselected for the sentence again. The filter does narrow down the interpretations as aid, but must be removed or otherwise sentence cannot be saved. I have been using it in this way as well, helping to narrow, but removed before saving."

And a couple of screen shots:

image (8)

If this isn’t a known bug, we definitely wanted to bring it to your attention. If there’s something we can do on our end (better than the described workaround) that would be good to know, too!

It has long been known that things sometimes crash when “is a constituent” constraints are used. The helpful details in this bug report enabled me to track what was going on (when I got there it sort of felt like I had discovered this before, so I must have failed to communicate it).

The problem is that there is no officially-blessed way to store an “is a constituent” constraint in the TSDB decision schema. Stephan Oepen allocated decision type 7 for the “exact unary chain” constraints when I originally got FFTB going, but for some reason lost to history did not allocate a type code for “is a constituent.”

In terms of engineering there is no particular challenge to saving these decisions and fitting the required information into the storage fields that are available, but Stephan will have to bless a type number, and confirm that doing so will not mess up other software that reads decision too badly. In the meantime, I suppose these “is a constituent” constraints will have to remain what they have always been – an incomplete feature.

1 Like

Thank you, Woodley!