
Downloaded from ....
http://www.isi.edu/natural-language/mt/dev-gold.txt
http://www.isi.edu/natural-language/mt/test-gold.txt

Note that these alignments correspond to LDC2014T12 (AMR 1.0).  LDC2020T02 incorporates all thata data and
thus has the same id's in it, however there's no guarenee that the sentences or alignments haven't been 
modified.  In addition, there is no tokenization supplied which means that any variation will drastically
impact the results.

From the paper "Supervised Syntax-based Alignment between English Sentences and AMR Graphs.pdf" section 5.1

The data used in our experiments was the Linguistic Data Consortium AMR corpus release 1.0 (LDC2014T12), 
consisting of 13,050 AMR/English sentence pairs. The statistics of the original split in the AMR 
corpus for training, development, and testing AMR parsers are shown in upper part of Table 1. Among 
which, 100 development 8 and 100 testing 9 pairs were manually annotated with gold alignments
[Pourdamghani et al., 2014].

