Proposal: change list-item values from concepts to string

[uhermjakob]

Current AMR Dict entry for :li http://www.isi.edu/~ulf/amr/lib/amr-dict.html#:li

Current policy

• :li (x / 1) :li (x2 / 2) :li (x3 / 3)
• :li (a / a) :li (b / b) :li (c / c)
• :li (i / i) :li (i2 / ii) :li (i3 / iii)

Proposed policy

• :li 1 :li 2 :li 3 ... :li "-1" (for "lastly", in analogy to ordinals)
• :li "a" :li "b" :li "c"
• :li "i" :li "ii" :li "iii"

Current reality

• inconsistent annotation
• 130+ instances conforming to current policy
• 30+ instances conforming to proposed policy

Motivation

• AMRs more legible.
• Proposed policy more consistent with rest of AMR.
  • Chapter 2: (c / chapter :mod 2)
  • Figure 5b: (f / figure :mod "5b")
• 2/b/ii etc. are not really concepts.
• Interference with AMR checking.
  • Should concept i really be shared for the i/I in "(i) Opening Remarks" and "I made some remarks."?
• list-items do not have complex structure (after all).
• Automatic conversion possible.
• AMR Editor could steer annotators to consistent new annotation.

[nschneid]

I like strings better than concepts. But I'm not sure I see the need to encode list item numbers at all—they are about the presentation of part of a sentence, not its meaning. Is the point to clarify that the order of items under and is not arbitrary? Wouldn't something like (o / ordered-list :op1 (x1 ...) :op2 (x2 ...) :op3 (x3 ...)) be simpler?

[uhermjakob]

Adopted by AMR meeting on Oct. 19, 2015 as proposed by Ulf.

Ulf updated AMR Dict at http://www.isi.edu/~ulf/amr/lib/amr-dict.html#list%20item as well as AMR Editor, AMR guidelines examples, AMR release corpus (automatically updated), AMR Checker; added have-li-91 as a reification for :li to be able to deal with a few cases of stranded list item markers.

@nschneid List item markers are typically spread over multiple sentences, so (o / ordered-list :op1 (x1 ...) :op2 (x2 ...) :op3 (x3 ...)) would not work as simple as that in our current annotation setup.