search

amrisi / amr-guidelines

250 stars 87 forks source link

Numbers as concepts? #169

Open nschneid opened 8 years ago

nschneid commented 8 years ago

@mdtux89 points out that there are cases like:

[2] cjconsensus bolt12_10510_9303.3 (snt. 1215 in workset bolt12-tune-master-1, last updated on Fri Nov 8, 2013) Tens of thousands or even millions of those displaced by the disaster have gradually had a home to shelter from the wind and rain!

(h / have-03 
  :ARG0 (p / person 
          :quant (m5 / multiple 
                   :op1 (o2 / or 
                          :op1 10000 
                          :op2 (x / 1000000 
                                 :mod (e / even)))) 
          :mod (t / that) 
          :ARG1-of (d / displace-01 
                     :ARG0 (d2 / disaster))) 
  :ARG1 (h2 / home 
          :purpose (s / shelter-01 
                     :ARG1 p 
                     :ARG3 (a / and 
                             :op1 (w / wind) 
                             :op2 (r / rain)))) 
  :manner (g / gradual))

Is (x / 1000000 :mod (e / even)) correct? If so, what are the conditions under which numeric concepts should be used? I couldn't find any examples in the AMR Dictionary or quantities documentation.

uhermjakob commented 8 years ago

We've tolerated numbers as concepts for a few years, but we have been cracking down on them lately. The AMR Checker has been flagging numbers as concepts as an error for several weeks now, including the AMR above. AMR annotation and QC teams are encouraged to de-conceptualize numbers, and I have been making a good number of corrections in this regard as well and will continue to do so for the upcoming release.

I keep adding entries to the AMR Dictionary, and this issue is certainly a worthy additional topic.

timjogorman commented 8 years ago

I've noticed the use of "numerical-quantity"; it might to either make that official or ban it. It definitely makes it easier to check over these rare cases where we don't know what we are quantifying:

225K is nowhere near a million. Ron Pauls crotchety meter was turned to 11

Perhaps more questionably, this has also been used for unknown quantified where you'd otherwise use "thing":

"I remember being thrilled at how many small donations he was receiving" "but if anyone earns x dollars, they are paying a certain percentage." "pay X percentage of taxes"

nschneid commented 8 years ago

numerical-quantity sounds reasonable to me (but I haven't thought about it much).

uhermjakob commented 8 years ago

Thanks, Tim, for pointing this out.

The use of numerical-quantity is described at http://www.isi.edu/~ulf/amr/lib/popup/quantity-types.html with a fairly limited scope, for which I added the following example to the consensus guidelines corpus:

1 mole of water weighs 18g.
(w / weigh-01
  :ARG1 (w2 / water
          :quant (n / numerical-quantity :quant 1
                   :unit (m / mole)))
  :ARG3 (m2 / mass-quantity :quant 18
          :unit (g / gram)))

I found that all uses of numerical-quantity in non-guidelines AMRs were improper and corrected them. The only case were I had at least some sympathy was one use of numerical-quantity in nominal relative clauses with how many, but the more common annotation used just number (or thing) in those cases.

I generally found the treatment of nominal relative clauses somewhat inconsistent, so I donated a new entry to the AMR Dict at http://www.isi.edu/~ulf/amr/lib/amr-dict.html#nominal%20relative%20clause, and made some corrections to improve annotation consistency.

The other tricky case was Ron Pauls crotchety meter was turned to 11 where I took the liberty to change numerical-quantity to score-on-scale-91.

-- Ulf