Self-edges!

[nschneid]

I didn't think these were allowed, but here is an example where "X when and if X takes place" is annotated with X :time X :condition X

A lot of no-votes guys and here they are voting on something like this and here they're going to be voting on it again when and if that vote ever takes place.

(s2 / say-01 
      :ARG1 (a / and 
            :op1 (p / person 
                  :ARG0-of (v / vote-01 
                        :ARG1 (h / have-polarity-91 
                              :ARG2 -) 
                        :quant (l / lot))) 
            :op2 (v2 / vote-01 
                  :ARG0 p 
                  :ARG1 (s / something 
                        :ARG1-of (r / resemble-01 
                              :ARG2 (t / this))) 
                  :location (h2 / here)) 
            :op3 (v3 / vote-01 
                  :ARG0 p 
                  :ARG1 s 
                  :location h2 
                  :mod (a2 / again) 
                  :time v3 
                  :condition v3 
                  :time (e / ever))) 
      :ARG2 (g / guy))

[goodmami]

But these "magically disappear upon reification", just as the cycle does, yes?

(s2 / say-01 
      :ARG1 (a / and 
            :op1 (p / person 
                  :ARG0-of (v / vote-01 
                        :ARG1 (h / have-polarity-91 
                              :ARG2 -) 
                        :quant (l / lot))) 
            :op2 (v2 / vote-01 
                  :ARG0 p 
                  :ARG1 (s / something 
                        :ARG1-of (r / resemble-01 
                              :ARG2 (t / this))) 
                  :location (h2 / here)) 
            :op3 (v3 / vote-01 
                  :ARG0 p 
                  :ARG1 s 
                  :location h2 
                  :mod (a2 / again) 
                  :ARG1-of (t2 / be-temporally-at-91 :ARG2 v3) 
                  :ARG1-of (c / have-condition-91 :ARG2 v3) 
                  :time (e / ever))) 
      :ARG2 (g / guy))

Although in the above, it doesn't seem to matter if you attach the reified nodes via :ARG1-of or :ARG2-of.

[nschneid]

Sure, they disappear on reification because reification turns the edges into nodes.

But I don't think I agree with the linguistic analysis in the first place. It seems there are two aspects of voting being referred to: (i) having the opportunity to cast a vote, and (ii) casting a vote for or against a proposal. The only sense I can make of "they're going to be voting on it again when and if that vote ever takes place" is that "they're going to be voting on it again" refers to (ii) while "when and if that vote ever takes place" refers to (i)—roughly, they're going to vote (cast ballots) when there is a vote (a call to the legislative body to cast ballots). (ii) has an ARG0, (i) doesn't. So I think I'd use two different vote-01 frames instead of the self-edge.

[goodmami]

Sorry, I didn't make my point very clear. From the guidelines regarding whether to use reification or not:

The resolution: we consider “AMR with reification” to be “real AMR”, with non-reified relations as semantic sugar. Therefore, if you are translating English into AMR, the rule is “whenever you feel like it”, because your AMRs will be normalized into reified form behind the scenes.

According to this statement, the self-edges you're seeing are just the "semantic sugar" and not the real graph. I'm curious how accurate the statement is because, for instance, smatch does not give 1.0 for two AMRs that differ only in reification.

Regarding the analysis... there's a lot not to love about that AMR (where does (s2 / say-01 ... :ARG2 (g / guy)) come from? Maybe I'm missing context.), and the original sentence isn't terribly clear. I think you're right about the (i) vs (ii) interpretations of "vote" here.

[nschneid]

I'm curious how accurate the statement is because, for instance, smatch does not give 1.0 for two AMRs that differ only in reification.

Yeah, my impression was always that the policy was "as annotators let's feel free to reify without guilt", but smatch does not actually handle the equivalence in practice.

where does (s2 / say-01 ... :ARG2 (g / guy)) come from?

I think "A lot of no votes guys" has been interpreted as vocative. say-01 is sometimes introduced for speech acts such as vocatives and quotations. (In retrospect, we probably should have used a different frame for those.)

[goodmami]

I see. So it's like "[There are] a lot of no-votes, guys, ..." or "Hey guys, there's a lot of no-votes ...". The focusing of (p / person) as the operand of (a / and) threw me off into thinking it was "[There are] a lot of guys-who-vote-no" or something (although the :quant relation is on the votes and not the person).

Anyway I don't mean to go off-topic. Thanks for explaining.

[nschneid]

Oh yeah, that is strange. Right now it looks like "there is/are a person/people who vote no a lot", which can't be right. Either vote-01 should be focused ("there is a lot of no-voting") or the :quant should be under person ("there are a lot of no-voters").