Nodes_and_Edges_Format.md

Docs - Table of Contents

• Output formats
• Alignment Format
• Nodes and Edges Format
• Data
• Hand Alignments
• Training
• Config File
• Step by Step Training
• Evaluation
• Alignment Evaluation
• Parser Performance

---

Nodes and Edges Format

The parser and aligner both output an additional "nodes and edges" format. This format lists the nodes and edges of the AMR graphs in a tab-separated format.

Each sentence is separated by a newline. The output looks like this:

# ::snt The president spoke in November.
# ::tok The president spoke in November .
# ::alignments 0-1|0.1 4-5|0.1.0+0.1 2-3|0 1-2|0.0 ::annotator JAMR dev v0.2 ::date 2014-10-29T20:53:39.435
# ::node        0       speak-01        2-3
# ::node        0.0     president       1-2
# ::node        0.1     date-entity     4-5
# ::node        0.1.0   11      4-5
# ::root        0       speak-01
# ::edge        date-entity     month   11      0.1     0.1.0   
# ::edge        speak-01        ARG0    president       0       0.0
# ::edge        speak-01        time    date-entity     0       0.1
(s / speak-01 
      :ARG0 (p / president) 
      :time (d / date-entity 
            :month 11))

Lines beginning with # ::node list the nodes. The format is:

# ::node TAB uniq_id TAB concept TAB start_alignment-end_alignment

• uniq_id field is a unique id for each node.
• concept field is the node's concept.
• start_alignment-end_alignment field is the span of words in the # ::tok field. start_alignment is inclusive, but end_alignment is not. So 0-1 means the first token in the sentence. If the node is unaligned, then the start_alignment-end_alignment field is blank.

Lines beginning with # ::root list the root. The format is:

# ::root TAB uniq_id TAB concept

The uniq_id is the same as listed in the nodes. The concept field is the concept of the root node.

Lines beginning with # ::edge list the edges. The format is:

# ::edges TAB source_concept TAB relation TAB dest_concept TAB source_uniq_id TAB dest_uniq_id

The edge points from source_uniq_id to dest_uniq_id with label relation. source_concept and dest_concept are there just for reference.

Nodes are printed first, then the root, and then the edges.