Need help interpreting MRS compare output

I am investigating a regression test failure using the classic regression test system (not rtest, in this case).

MRS compare reports diffs like this:

I am confused:

  1. It is hard to read MRSs like this (and the unix diff utility isn’t being of much help because this is all one string), but from what I see there might be a difference on what information is on the noun and what is on the _exist_q_rel; I think we’ve established that this should not cause a diff but I am not seeing anything else. Does anyone see what the meaningful diff is?

  2. I don’t understand what “item 19” means; I only have 17 sentences in the txt-suite. What do these numbers correspond to in the logs? Perhaps each tree gets an item number, i.e. if the first sentence in the txt-suite gets assigned two trees, they will be item 1 and 2? Where do I look for these; the “result” file (I don’t think so)?

  3. When I try to compare the profiles in tsdb++, it doesn’t show me any diffs. Why would that be? I checked that I don’t have any TSQL condition checked…

(By the way. This test is passing with rtest. Usually it is the other way around; I deleted the repository on the VM and checked it out again, to make sure nothing is out of sync. But, I guess, for now let’s assume that something must be out of sync.)

Those item numbers should be the i-id field in the tsdb profile. The first field of the “item” file.

And yes, it is hard to see differences between similar MRS text printouts…

That is puzzling because there is no item 19 in the corresponding item file:
Screen Shot 2020-01-21 at 6.36.09 PM

In an email to you I previously offered a script that colors character-diffs of MRSs (or anything). Maybe it’s worth putting here so others can see it, too:

The wdiff utility is not MRS aware, so it will see any changes as diffs (e.g., re-ordered EPs or properties), but maybe it’s good enough. For convenience I copy the scripts as text below:



wdiff -n -w $'\033[30;41m' -x $'\033[0m' \
         -y $'\033[30;42m' -z $'\033[0m' \

getmrs (maybe adjust to your needs, particularly the path to the virtual environment activation script, if you’re using one):


source ~/repos/pydelphin/py3/bin/activate

delphin convert --color=never --select "mrs where i-id=$1" "$2" --indent 


For your extra-items problem, it could be that there’s unescaped newlines in your profile which make it look like additional items, but I would expect a more serious error message in that case. Can you attach the gold and test profiles?

1 Like

I think I can’t attach anything here but I could send them via email.

That’s fine, or I probably still have access to patas if you have it there somewhere