SRL on top of UD - Githubissues

74 How much is its accuracy , I think , is still quite hard to determine at this time , indeed ,

─┮  
 │     ╭─╼ How ADV advmod 1 2  
 │   ╭─┶ much ADJ dep 2 3  
 │ ╭─┾ is VERB ccomp 3 13              <= ARG1-DSP
 │ │ │ ╭─╼ its PRON nmod:poss 4 5  
 │ │ ╰─┶ accuracy NOUN nsubj 5 3  
 │ ├─╼ , PUNCT punct 6 13  
 │ │ ╭─╼ I PRON nsubj 7 8  
 │ ├─┶ think VERB parataxis 8 13       <= PRED think.01
 │ ├─╼ , PUNCT punct 9 13  
 │ ├─╼ is AUX cop 10 13  
 │ ├─╼ still ADV advmod 11 13  
 │ ├─╼ quite ADV advmod 12 13  
 ╰─┾ hard ADJ root 13 0                <= ARG1-DSP
   │ ╭─╼ to PART mark 14 15  
   ├─┾ determine VERB xcomp 15 13  
   │ │ ╭─╼ at ADP case 16 18  
   │ │ ├─╼ this DET det 17 18  
   │ ├─┶ time NOUN obl 18 15  
   │ ├─╼ , PUNCT punct 19 15  
   │ ╰─╼ indeed ADV advmod 20 15  
   ╰─╼ , PUNCT punct 21 13

The cases of ARG?-DSP, when projected to dependencies may not be trivial.

Does the annotation above make sense? @MarthaSPalmer? Considering that token 3 is argument of token 13, can we only annotate token 13? The DSP would only mark that the argument is actually an ancestor of the token 8 (the predicate).

@nschneid, what about the syntactic annotation?

Yet another case to encourage some thoughts. Basically, here, the 'that' was annotated as WHADVP in the constituent analysis and in UD it is SCONJ/mark introducing the subordinate clause.

13> It was during this year that the Japanese army developed a strategy to rapidly force the Chinese people into submission by the end of 1940 .

─┮  
 │ ╭─╼ It PRON nsubj 1 5  
 │ ├─╼ was AUX cop 2 5  
 │ ├─╼ during ADP case 3 5  
 │ ├─╼ this DET det 4 5  
 ╰─┾ year NOUN root 5 0              <= ARGM-TMP
   │ ╭─╼ that ADP mark 6 10  
   │ │ ╭─╼ the DET det 7 9  
   │ │ ├─╼ Japanese ADJ amod 8 9  
   │ ├─┶ army NOUN nsubj 9 10  
   ├─┾ developed VERB acl 10 5       <= develop.02
   │ │ ╭─╼ a DET det 11 12  
   │ ╰─┾ strategy NOUN obj 12 10  
   │   │ ╭─╼ to PART mark 13 15  
   │   │ ├─╼ rapidly ADV advmod 14 15  
   │   ╰─┾ force VERB acl 15 12  
   │     │ ╭─╼ the DET det 16 18  
   │     │ ├─╼ Chinese ADJ amod 17 18  
   │     ├─┶ people NOUN obj 18 15  
   │     │ ╭─╼ into ADP case 19 20  
   │     ├─┶ submission NOUN obl 20 15  
   │     │ ╭─╼ by ADP case 21 23  
   │     │ ├─╼ the DET det 22 23  
   │     ╰─┾ end NOUN obl 23 15  
   │       │ ╭─╼ of ADP case 24 25  
   │       ╰─┶ 1940 NUM nmod 25 23  
   ╰─╼ . PUNCT punct 26 5

Previously

Tree:
-----
    (TOP (S-CLF (NP-SBJ (PRP It))
                (VP (VBD was)
                    (PP-TMP-PRD (IN during)
                                (NP (DT this)
                                    (NN year)))
                    (SBAR (WHADVP-2 (IN that))
                          (S (NP-SBJ (DT the)
                                     (JJ Japanese)
                                     (NN army))
                             (VP (VBD developed)
                                 (NP (NP (DT a)
                                         (NN strategy))
                                     (SBAR (WHNP-1 (-NONE- 0))
                                           (S (NP-SBJ (-NONE- *T*-1))
                                              (VP (TO to)
                                                  (ADVP-MNR (RB rapidly))
                                                  (VP (VB force)
                                                      (NP (DT the)
                                                          (JJ Chinese)
                                                          (NNS people))
                                                      (PP-CLR (IN into)
                                                              (NP (NN submission)))
                                                      (PP-TMP (IN by)
                                                              (NP (NP (DT the)
                                                                      (NN end))
                                                                  (PP (IN of)
                                                                      (NP (CD 1940))))))))))
                                 (ADVP-TMP (-NONE- *T*-2))))))
                (. .)))

Leaves:
-------
    0   It
    1   was
           sense: be-v.1
           prop:  be.01
            v          * -> 1:0,  was
            ARG2       * -> 2:1,  during this year
            ARG1       * -> 5:2,  that the Japanese army developed a strategy 0 *T*-1 to rapidly force the Chinese
                                  people into submission by the end of 1940 *T*-2
    2   during
    3   this
           coref: IDENT        247   3-4    this year
           name:  DATE               3-4    this year
    4   year
    5   that
    6   the
           coref: IDENT        211   6-8    the Japanese army
    7   Japanese
           name:  NORP               7-7    Japanese
    8   army
    9   developed
           sense: develop-v.1
           prop:  develop.02
            v          * -> 9:0,  developed
            ARG0       * -> 6:1,  the Japanese army
            ARG1       * -> 10:2, a strategy 0 *T*-1 to rapidly force the Chinese people into submission by the end of
                                  1940
            ARGM-TMP   * -> 27:0, *T*-2 -> 5:1, that
            LINK-SLC   * -> 27:0, *T*-2 -> 5:1, that
                       * -> 2:1,  during this year

74 How much is its accuracy , I think , is still quite hard to determine at this time , indeed ,

@nschneid, what about the syntactic annotation?

Well, the sentence isn't 100% grammatical.

This sentence exemplifies a tough-construction: [How much is its accuracy] + hard + [to determine]. I'm not entirely sure how to analyze it, but I suspect it should be csubj instead of ccomp for the attachment of the first clause. There is a long unresolved discussion at UniversalDependencies/docs#308.

This is way above my pay grade.

Katie, haven’t you done a lot of PB annotation on dependency parses? How would you have annotated the sentence below?

Nathan may well be right that it isn’t grammatical or the right parse, but PropBank would still try to annotate what is there.

Martha

Begin forwarded message:

From: Nathan Schneider @.**@.>> Subject: Re: [propbank/propbank-release] SRL on top of UD (Issue #17) Date: November 26, 2021 at 11:42:52 PM MST To: propbank/propbank-release @.**@.>> Cc: Martha Palmer @.**@.>>, Mention @.**@.>> Reply-To: propbank/propbank-release @.**@.>>

74 How much is its accuracy , I think , is still quite hard to determine at this time , indeed ,

@nschneidhttps://github.com/nschneid, what about the syntactic annotation?

Well, the sentence isn't 100% grammatical.

This sentence exemplifies a tough-construction: [How much is its accuracy] + hard + [to determine]. I'm not entirely sure how to analyze it, but I suspect it should be csubj instead of ccomp for the attachment of the first clause. There is a long unresolved discussion at UniversalDependencies/docs#308https://github.com/UniversalDependencies/docs/issues/308.

On Nov 24, 2021, at 1:25 PM, Alexandre Rademaker @.**@.>> wrote:

74 How much is its accuracy , I think , is still quite hard to determine at this time , indeed ,

─┮ │ ╭─╼ How ADV advmod 1 2 │ ╭─┶ much ADJ dep 2 3 │ ╭─┾ is VERB ccomp 3 13 <= ARG1-DSP │ │ │ ╭─╼ its PRON nmod:poss 4 5 │ │ ╰─┶ accuracy NOUN nsubj 5 3 │ ├─╼ , PUNCT punct 6 13 │ │ ╭─╼ I PRON nsubj 7 8 │ ├─┶ think VERB parataxis 8 13 <= PRED think.01 │ ├─╼ , PUNCT punct 9 13 │ ├─╼ is AUX cop 10 13 │ ├─╼ still ADV advmod 11 13 │ ├─╼ quite ADV advmod 12 13 ╰─┾ hard ADJ root 13 0 <= ARG1-DSP │ ╭─╼ to PART mark 14 15 ├─┾ determine VERB xcomp 15 13 │ │ ╭─╼ at ADP case 16 18 │ │ ├─╼ this DET det 17 18 │ ├─┶ time NOUN obl 18 15 │ ├─╼ , PUNCT punct 19 15 │ ╰─╼ indeed ADV advmod 20 15 ╰─╼ , PUNCT punct 21 13

The cases of ARG?-DSPhttps://github.com/propbank/propbank-release/blob/master/docs/conll-conversion-notes.md#argx-dsp-arguments, when projected to dependencies may not be trivial.

Does the annotation above make sense? @MarthaSPalmerhttps://github.com/MarthaSPalmer? Considering that token 3 is argument of token 13, can we only annotate token 13? The DSP would only mark that the argument is actually an ancestor of the token 8 (the predicate).

@nschneidhttps://github.com/nschneid, what about the syntactic annotation?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHubhttps://github.com/propbank/propbank-release/issues/17, or unsubscribehttps://github.com/notifications/unsubscribe-auth/ABB327USNK7XHCLNG4ZXVNDUNVC27ANCNFSM5IW56WYQ. Triage notifications on the go with GitHub Mobile for iOShttps://apps.apple.com/app/apple-store/id1477376905?ct=notification-email&mt=8&pt=524675 or Androidhttps://play.google.com/store/apps/details?id=com.github.android&referrer=utm_campaign%3Dnotification-email%26utm_medium%3Demail%26utm_source%3Dgithub.

propbank / propbank-release

SRL on top of UD #17