Cornell AMR Semantic Parser

Quick Start

You can get started and parse a few simple sentences in under five minutes using our AMI image (ami-bdc8d8d7). Launch an EC2 instance (recommended r3.2xlarge at least) with our AMI image. The code in the image is fully complied and ready to run. The ~/amr/do.sh command will quickly parse a few example sentences. The AMI is based on Ubuntu and the user name is ubuntu. The code repository is available here.

Requirements

Java 8.

Preparing the Repository

Get all required resources: ./getres.sh (form the root of the repository)
Compile: ant dist

Pre-trained Models

A pre-trained model is available to download here.

Parsing

Given a file sentences.txt, which contains a sentence on each line, and a model file amr.sp, both located in the root of the repository:

java -Xmx8g -jar dist/amr-1.0.jar parse rootDir=`pwd` modelFile=`pwd`/amr.sp sentences=`pwd`/sentences.txt

The output files will be in experiments/parse/logs. To see the full set of options (including increasing the logging level), run:

java -jar dist/amr-1.0.jar

Preparing the data (required only for training and testing)

To re-create our experiments, obtain the AMR Bank release 1.0 (LDC2014T12) form LDC. Extract the corpus to the directory corpus/amr_anno_1.0 and place this folder in your git root directory.

Then run the following:

Compile the code: ant dist
Alias python command to ensure shell commands are executed using Python 2.7: alias python=python2.7 source ~/.bashrc python --version
Prepare the environment: ./config.sh
Prepare the data: ./prepdata-ldc.sh
Disable alias in current shell: unalias python

Training/Testing

Training and testing experiments are written according to the conventions established by the Cornell Semantic Parsing Framework's experiments platform (Explat). I suggest you read through the documentation for SPF and look through the .exp files in this repo's experiments folder in order to get a sense of how to create a new experiment. Training on even a small amount of data demands quite a bit of memory (hopefully memory usage can be reduced once the bottleneck is identified). Check your system monitor to determine the amount of RAM and swap memory available to you and ensure you are setting your JVM heap size accordingly (the Xmx parameter specifies the maximum memory allocation pool for a JVM, while Xms specifies the initial memory allocation pool). If you recieve an OutOfMemory error, reduce your batch size or try using a cloud service with more available memory to train your model. I required 256Gb of RAM to train a model on all of the AMR Bank release 1.0 data in reasonable time, although you could definately use less memory.

Attribution

@InProceedings{artzi-lee-zettlemoyer:2015:EMNLP,
  author    = {Artzi, Yoav  and  Lee, Kenton  and  Zettlemoyer, Luke},
  title     = {Broad-coverage CCG Semantic Parsing with AMR},
  booktitle = {Proceedings of the 2015 Conference on Empirical Methods in Natural Language Processing},
  month     = {September},
  year      = {2015},
  address   = {Lisbon, Portugal},
  publisher = {Association for Computational Linguistics},
  pages     = {1699--1710},
  url       = {http://aclweb.org/anthology/D15-1198}
}

HoltSpalding/amr