A python library that makes AMR parsing, generation and visualization simple.

For the latest documentation, see ReadTheDocs.

!! Note: The models must be downloaded and installed separately. See the Installation Instructions.

amrlib is a python module designed to make processing for Abstract Meaning Representation (AMR) simple by providing the following functions

• Sentence to Graph (StoG) parsing to create AMR graphs from English sentences.
• Graph to Sentence (GtoS) generation for turning AMR graphs into English sentences.
• A QT based GUI to facilitate conversion of sentences to graphs and back to sentences
• Methods to plot AMR graphs in both the GUI and as library functions
• Training and test code for both the StoG and GtoS models.
• A SpaCy extension that allows direct conversion of SpaCy Docs and Spans to AMR graphs.
• Sentence to Graph alignment routines
  • FAA_Aligner (Fast_Align Algorithm), based on the ISI aligner code detailed in this paper.
  • RBW_Aligner (Rule Based Word) for simple, single token to single node alignment
• An evaluation metric API including including...
  • Smatch (multiprocessed with enhanced/detailed scores) for graph parsing
  • BLEU for sentence generation
  • Alignment scoring metrics detailing precision/recall
• There is also a related co-referencing project/model at amr_coref.

The system includes different neural-network models for parsing and for generation. !! Note: Models must be downloaded and installed separately. See amrlib-models for all parse and generate model download links.

• Parse (StoG) model_parse_xfm_bart_large gives an 83.7 SMATCH score with LDC2020T02.

• Generation (GtoS) generate_t5wtense gives a 54 BLEU with tense tags or 44 BLEU with un-tagged LDC2020T02.

• CoReference resolution at amr_coref achieves a 0.548 CoNLL-2012 average score.

The GUI allows for simple viewing, conversion and plotting of AMR Graphs.

The project was built and tested under Python 3 and Ubuntu but should run on any Linux, Windows, Mac, etc.. system.

See Installation Instructions for details on setup.

To convert sentences to graphs

import amrlib
stog = amrlib.load_stog_model()
graphs = stog.parse_sents(['This is a test of the system.', 'This is a second sentence.'])
for graph in graphs:
    print(graph)

To convert graphs to sentences

import amrlib
gtos = amrlib.load_gtos_model()
sents, _ = gtos.generate(graphs)
for sent in sents:
    print(sent)

For a detailed description see the Model API.

To use as an extension, you need spaCy version 2.0 or later. To setup the extension and use it do the following

import amrlib
import spacy
amrlib.setup_spacy_extension()
nlp = spacy.load('en_core_web_sm')
doc = nlp('This is a test of the SpaCy extension. The test has multiple sentences.')
graphs = doc._.to_amr()
for graph in graphs:
    print(graph)

For a detailed description see the Spacy API.

For an example of how to use the library to do paraphrasing, see the Paraphrasing section in the docs.

If you find a bug, please report it on the GitHub issues list. Additionally, if you have feature requests or questions, feel free to post there as well. I'm happy to consider suggestions and Pull Requests to enhance the functionality and usability of the module.