nsaphra / jamr

JAMR - AMR Parser

JAMR is a semantic parser for the Abstract Meaning Representation.

Building

JAMR depends on

• Scala,
• the Illinois NER system,
• the tokenization scripts in cdec,
• WordNet (for the aligner).

Install these dependencies, and then change the relevant environment variables in scripts/config.sh. Source the config script so that other bash scripts have access to these environment variables (or add them to ~/.bashrc to set them permanently):

. scripts/config.sh

Run ./compile to build an uberjar, which will be output to target/scala-{scala_version}/jamr-assembly-{jamr_version}.jar (If you get out of memory errors during this step, you may need to edit the Java memory options in the script sbt and build.sbt.)

Preprocessing

Download LDC2013E117.tgz from the LDC Catalog (requires an LDC subscription). Extract the file deft-amr-release-r3-proxy.txt into data/LDC-2013-Sep/ and rename it amr-release-proxy.txt.

cd scripts/preprocessing

Run ./PREPROCESS.sh.

Training

(To skip this step, download and extract model weights current.tgz into the directory experiments/current.)

cd scripts/training

Extract concept table:

./cmd.conceptTable.train

Concept identification (stage1) training:

./cmd.stage1-weights

Relation identification (stage2) training:

./cmd.stage2-weights

(Search for 'Performance on Dev' in stage2-weights.err for early stopping)

Evaluating

Decode test set:

./cmd.test.decode.allstages

or

./cmd.test.decode.stage2only

Evaluate the predictions using smatch:

${JAMR_HOME}/scripts/smatch_v1_0/smatch_modified.py --pr -f ${JAMR_HOME}/experiments/current/test.decode.allstages ${JAMR_HOME}/data/LDC-2013-Sep/amr-release-proxy.test
${JAMR_HOME}/scripts/smatch_v1_0/smatch_modified.py --pr -f ${JAMR_HOME}/experiments/current/test.decode.stage2only ${JAMR_HOME}/data/LDC-2013-Sep/amr-release-proxy.test