Aligning FrameNet and English Wiktionary

[Andrew Gargett]

hi

I have recently begun using the UBY resources, and have found them to be very useful on the whole, and I have made an initial exploration of these, thanks in large part to some very kind help from Silvana Hartmann. However, I have run into issues to do with aligning various components.

I would like to extract enough information to allow me to align the English FrameNet and English Wiktionary; specifically, given a WK entry I would like to find the matching FN lexical unit.

I have downloaded the UBY SQL dump, and I have discovered that the tables "SenseAxis" and "sense", have fields "senseTwoId" and "senseId", respectively, both of which yield values like "WktEN_sense_...". So I have reasoned that by aligning these tables using these fields, I would be able to then extract English FN lexical units matching the actual written forms from the English WK.

I have written the following SQL queries to carry this out:

(1)
SELECT mer.externalReference AS FNLU, sa.senseTwoId AS WKSENSE 
FROM    SenseAxis AS sa
               JOIN MonolingualExternalRef AS mer ON (mer.senseId=sa.senseOneId)
WHERE  substring(sa.senseOneId,1,2)="FN" AND
               substring(sa.senseTwoId,1,4)="WktE" AND
               mer.externalSystem="FrameNet_1.5_eng_lexicalUnit";

(2)
SELECT  s.senseId AS WKSENSE,fr_l.writtenForm AS ENGWRITTENFORM
FROM    sense as s
               JOIN senserelation as sr ON (s.senseId=sr.senseId)
               JOIN formRepresentation_lemma AS fr_l ON (sr.formRepresentationId=fr_l.formRepresentationId)
               JOIN lexicalEntry as le ON (le.lemmaId=fr_l.lemmaId)
               JOIN lexicon as l ON (l.lexiconId=le.lexiconId)
WHERE  l.languageIdentifier='eng' AND
               l.lexiconName='FrameNet' AND
               fr_l.languageIdentifier='eng';

However, after aligning "SenseAxis.senseTwoId" and "sense.senseId", the FN lexical units that are returned are not in fact an appropriate match for the returned WK written form. For example, "activity" is matched to FN lexical unit with id "7900", which is in fact "book" (as in "Arrrest"). So either something has gone wrong with my SQL queries, and/or I have approached this problem from the wrong angle, and/or what I am trying to do is not in fact possible in UBY.

Either way, any help with this would be very much appreciated.

Many thanks
Andrew Gargett

[Silvana Hartmann]

Dear Andrew,

the alignment of English FrameNet and English Wiktionary is contained in the SenseAxis. From the aligned senses you can get the corresponding (aligned) lexical entries and  lemmas, and FrameNet lexical Unit Ids.

Your first query works just fine, but your second query goes wrong. You're extracting the writtenForm for a word related to your given Sense via SenseRelation, not for the given Sense.

To get the writtenForm from Sense, you need to go via its LexicalEntry to FormRepresentation_Lemma. To get the writtenForm for a given MonolingualExternalRef, you need to go via Sense, LexicalEntry, and so forth:

select S.senseId, F.writtenForm
from MonolingualExternalRef M
join Sense S on M.senseId=S.senseId join LexicalEntry E on E.lexicalEntryId=S.lexicalEntryId join FormRepresentation_Lemma F on F.lemmaId=E.lemmaId where M.externalReference="13972" and M.externalSystem="FrameNet_1.5_eng_lexicalUnit";

The result of this query on uby_medium_0_6_0 is:

+---------------+-------------+
| senseId       | writtenForm |
+---------------+-------------+
| FN_Sense_8732 | stay        |
+---------------+-------------+

If you combine this kind of query with your first query, you will get the correct lemmas for the Wkt and the FN sense.

Does that help?

Best wishes,

Silvana

 

--
You received this message because you are subscribed to the Google Groups "UBY Users" group.
To unsubscribe from this group and stop receiving emails from it, send an email to uby-users+...@googlegroups.com.
For more options, visit https://groups.google.com/d/optout.

[Andrew Gargett]

Dear Silvana

Many thanks for your help with this -- this is exactly what I was looking for!

best
Andrew