Example:
# ::id nw.wsj_0003.5 ::amr-annotator ISI-AMR-01 ::preferred
# ::tok A Lorillard spokewoman said , `` This is an old story .
# ::alignments 1-1.1.1.1.2.1 2-1.1.1.2 3-1 6-1.2.1 7-1.2.1.r 9-1.2.2 10-1.2
(s / say-01~e.3
:ARG0 (p / person
:ARG0-of (h / have-org-role-91
:ARG1 (c / company :wiki "Lorillard_Tobacco_Company"
:name (l / name :op1 "Lorillard"~e.1))
:ARG2 (s2 / spokeswoman~e.2)))
:ARG1 (s3 / story~e.10
:domain~e.7 (t / this~e.6)
:mod (o2 / old~e.9)))
Explanation:
The alignment files show alignment in two (redundant) formats.
(1) Alignment suffixes "~e.n" attached to concepts, strings, roles, or
(reentrant) variables, where "n" is the nth token of the sentences
(starting at 0).
Example: say-01~e.3 means that the concept say-01 aligns to token 3 ("said").
(2) The line starting with "# ::alignments" is a list of alignments of
form n-1.x.y.z[.r] where "n" is the nth token of the sentences (starting at 0)
and where "1" designates the root node of the amr (say-01), "1.x" is
the x-th sub-amr of that root node (starting at 1), "1.x.y" is the y-th
sub-amr of "1.x" etc.
The suffix ".r" designates the role of a sub-amr.
Example: 10-1.2 means that there is an alignment between token 10 ("story")
and the second sub-tree of the amr root ("(s3 / story)").
7-1.2.1.r means that there is an alignment between token 7 ("is")
and the role of the first sub-tree of the second sub-tree of the
root amr (":domain").
Notes regarding XML tags in tokenized text
(1) xml tags count as a single token, even if they contain spaces.
Example:
# ::tok A CNN report.
Token 0: A
Token 1:
Token 2: CNN
Token 3:
Token 4: report
(2) If the sentence starts with an xml tag, it does not count as a token.
This is admittedly somewhat odd. We plan to fix this in future releases.
Example:
# ::tok CNN report
Token 0: CNN
Token 1:
Token 2: report