Bad mappings in tab file for PWN31

[goodmami]

There are some ILI links with skos:closeMatch in the Turtle file for the PWN31 mapping but these are showing up in the corresponding tab file simply as ILI links. These should be removed or fixed somehow.

$ grep closeMatch ../cili/ili-map-wn31.ttl 
ili:i10601 skos:closeMatch pwn31:301940473-s . # objective, documentary
ili:i10765 skos:closeMatch pwn31:301972355-s . # unasked, unsolicited
ili:i13487 skos:closeMatch pwn31:302445314-s . # impossible, insufferable, unacceptable, unsufferable
ili:i14242 skos:closeMatch pwn31:302532138-s . # unintentional, unwilled
ili:i14242 skos:closeMatch pwn31:301342529-s . # unintentional, unwilled
ili:i34549 skos:closeMatch pwn31:202578494-v . # victimize, swindle, rook, goldbrick, nobble, diddle, bunco, defraud, scam, mulct, gyp, gip, hornswoggle, short-change, con
ili:i39442 skos:closeMatch pwn31:100767587-n . # crime, offense, criminal_offense, criminal_offence, offence, law-breaking
ili:i39502 skos:closeMatch pwn31:100781071-n . # bunco, bunco_game, bunko, bunko_game, con, confidence_trick, confidence_game, con_game, gyp, hustle, sting, flimflam
ili:i40396 skos:closeMatch pwn31:100951435-n . # technology, engineering
ili:i40396 skos:closeMatch pwn31:100951878-n . # technology, engineering
ili:i49794 skos:closeMatch pwn31:102675726-n . # accordion, piano_accordion, squeeze_box
ili:i50032 skos:closeMatch pwn31:102716628-n . # annulet
ili:i50032 skos:closeMatch pwn31:102716929-n . # annulet
ili:i50034 skos:closeMatch pwn31:102716929-n . # annulet, roundel
ili:i50034 skos:closeMatch pwn31:102716785-n . # annulet, roundel
ili:i50490 skos:closeMatch pwn31:102799187-n . # barrel_organ, grind_organ, hand_organ, hurdy_gurdy, hurdy-gurdy, street_organ
ili:i56414 skos:closeMatch pwn31:104455013-n . # mousse, hair_mousse, hair_gel
ili:i63228 skos:closeMatch pwn31:107059027-n . # forte, fortissimo
ili:i63228 skos:closeMatch pwn31:107059160-n . # forte, fortissimo
ili:i67468 skos:closeMatch pwn31:105828731-n . # technicality, trifle, triviality
ili:i72354 skos:closeMatch pwn31:106836790-n . # umlaut, dieresis, diaeresis
ili:i73625 skos:closeMatch pwn31:107048857-n . # plainsong, plainchant, Gregorian_chant
ili:i89444 skos:closeMatch pwn31:110020122-n . # deist, freethinker
ili:i89454 skos:closeMatch pwn31:110021663-n . # democrat, populist
ili:i89516 skos:closeMatch pwn31:110032289-n . # dietician, dietitian, nutritionist
ili:i90722 skos:closeMatch pwn31:110230422-n . # interior_designer, designer, interior_decorator, house_decorator, room_decorator, decorator
ili:i102993 skos:closeMatch pwn31:112599160-n . # mung, mung_bean, green_gram, golden_gram, Vigna_radiata, Phaseolus_aureus

It looks like the skos:closeMatch ones usually have the ILI in in the OEWN:

$ grep 'id="oewn-00767587-n"' english-wordnet-2021.xml 
    <Synset id="oewn-00767587-n" ili="in" members="oewn-offense-n oewn-offence-n" partOfSpeech="n" dc:subject="noun.act">

I tried to determine if this is always the case:

$ # get offset-pos for ILI mappings with "closeMatch" (remove superfluous leading digit on IDs)
$ grep 'closeMatch' ../cili/ili-map-wn31.ttl | sed 's/.*pwn31:[0-9]\([^ ]*\) .*/\1/' | sort -n > closematches
$ # get offset-pos for synsets with ili="in"
$ grep 'ili="in"' english-wordnet-2021.xml | sed 's/.*id="oewn-\([^"]*\)".*/\1/' | sort -n > ins
$ # OEWN has many introduced ILIs not part of this set, so show those only in ILI mapping
$ comm ins closematches -13
01342529-s
02532138-s
02716628-n
02716929-n
$ # Further inspect those synsets in CILI
$ grep '01342529-s\|02532138-s\|02716628-n\|02716929-n' ../cili/ili-map-wn31.ttl 
ili:i7301 owl:sameAs pwn31:301342529-s . # unintentional, unplanned, unwitting
ili:i13987 owl:sameAs pwn31:302532138-s . # unwilled
ili:i14242 skos:closeMatch pwn31:302532138-s . # unintentional, unwilled
ili:i14242 skos:closeMatch pwn31:301342529-s . # unintentional, unwilled
ili:i50032 skos:closeMatch pwn31:102716628-n . # annulet
ili:i50032 skos:closeMatch pwn31:102716929-n . # annulet
ili:i50033 owl:sameAs pwn31:102716628-n . # annulet, bandelet, bandelette, bandlet, square_and_rabbet
ili:i50034 skos:closeMatch pwn31:102716929-n . # annulet, roundel

So it looks like 3 of 4 closeMatch ILI links are parallel to a sameAs link to the same synset. The other one only had closeMatch links and no sameAs to the same synset for two ILIs. That synset has ili="in" (it wasn't excluded before because there were two ILIs (i50032 and i50034) which pointed to the same synset, and comm did not exclude the second instance). Digging into that one further:

$ # Only closeMatch links for i50032 and i50034 for PWN31
$ grep i5003[24] ../cili/ili-map-wn31.ttl 
ili:i50032 skos:closeMatch pwn31:102716628-n . # annulet
ili:i50032 skos:closeMatch pwn31:102716929-n . # annulet
ili:i50034 skos:closeMatch pwn31:102716929-n . # annulet, roundel
ili:i50034 skos:closeMatch pwn31:102716785-n . # annulet, roundel
$ # Only sameAs links for both in PWN30
$ grep i5003[24] ../cili/ili-map-wn30.ttl 
<i50032>    owl:sameAs  pwn30:02713769-n . # annulet
<i50034>    owl:sameAs  pwn30:02713992-n . # annulet, roundel

We don't have a precedent or a good way to add introduced ILIs to the mapping. Assigning ili="in" is done for a wordnet project, and here we should generate new IDs. So I suppose all of these closeMatch cases should simply be dropped from the .tab files?

[fcbond]

OK, I rebuilt the mapping (ili-map-pwn31.tab) as follows:

grep sameAs ../cili/ili-map-wn31.ttl | cut -d ' ' -f1,3 --output-delimiter=$' ' | sed s/pwn31:[12345]// | sed s/ili://| sort -nk 1.2 > ili-map-pwn31.tab

This keeps only the sameAs links.

[goodmami]

Somehow the file had null (\0) delimiters instead of tab delimiters between the fields. The following command produces tabs:

$ sed -rn '/owl:sameAs/{s/ili:([^ ]*) owl:sameAs pwn31:[1-5]([^ ]*) .*/\1\t\2/;p}' ili-map-wn31.ttl | sort -nk 1.2 > ili-map-pwn31.tab

The sort command is unnecessary; the same results are obtained without it as the Turtle file is already sorted in this order. But it's also good to be explicit.

I'll check in the new file.

[fcbond]

I checked with ediff-buffers, don't know how that crept in.

Thanks for the fix.

On Thu, Nov 4, 2021 at 7:45 AM Michael Wayne Goodman < @.***> wrote:

Somehow the file had null (\0) delimiters instead of tab delimiters between the fields. The following command produces tabs:

$ sed -rn '/owl:sameAs/{s/ili:([^ ]) owl:sameAs pwn31:[1-5]([^ ]) .*/\1\t\2/;p}' ili-map-wn31.ttl | sort -nk 1.2 > ili-map-pwn31.tab

The sort command is unnecessary; the same results are obtained without it as the Turtle file is already sorted in this order. But it's also good to be explicit.

I'll check in the new file.

— You are receiving this because you modified the open/close state. Reply to this email directly, view it on GitHub https://github.com/globalwordnet/cili/issues/13#issuecomment-960299325, or unsubscribe https://github.com/notifications/unsubscribe-auth/AAIPZRUVZQ2I7M7UOTD2RGTUKHJTXANCNFSM5HHFKPKA . Triage notifications on the go with GitHub Mobile for iOS https://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Android https://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

-- Francis Bond http://www3.ntu.edu.sg/home/fcbond/ Division of Linguistics and Multilingual Studies Nanyang Technological University

[goodmami]

Yeah, not sure. I tried out your commands and it seemed to work. Strange.