A server to test parsing text to Abstract Meaning Representation. All of these are for experimentation purpose. I would like to develop a small meaning representation corpus for incident report in the future.
- Use AMRLib for AMR parsing, with a Sentence to Graph model from amrlib-models (AMR is a graph, rather than a tree)
- Use Uvicorn and FastAPI for web server and API framework
- Download source
git clone https://github.com/bact/incidentamr-server.git
- Install required libraries
cd incidentamr-server pip install -r requirements.txt
- Install a model
- Download any Sentence to Graph model from amrlib-models/releases.
- For example,
model_parse_xfm_bart_large-v0_1_0.tar.gz
.
- For example,
- Extract the tar.gz file, you will get a directory containing .json and .bin files. Rename that directory to
stog
. - Put the
stog
directory insideincidentamr-server/incidentamr_server/models
directory.
- Download any Sentence to Graph model from amrlib-models/releases.
-
Start the server. From inside
incidentamr_server
directory, run:uvicorn main:app --reload
-
Then, from within a web browser, open http://127.0.0.1:8000.
The web interface will look like this:
The main paper for Abstract Meaning Representatio (AMR) is
- Banarescu, Laura, Claire Bonial, Shu Cai, Madalina Georgescu, Kira Griffitt, Ulf Hermjakob, Kevin Knight, Philipp Koehn, Martha Palmer, and Nathan Schneider. 2013. ‘Abstract Meaning Representation for Sembanking’. In Proceedings of the 7th Linguistic Annotation Workshop and Interoperability with Discourse, 9. Sofia, Bulgaria: Association for Computational Linguistics. https://aclanthology.org/W13-2322.pdf.
More information about AMR can be found at AMR Bank website.
Another interesting development is Uniform Meaning Representation (UMR). Their main paper is
- Van Gysel, Jens E. L., Meagan Vigus, Jayeol Chun, Kenneth Lai, Sarah Moeller, Jiarui Yao, Tim O’Gorman, et al. 2021. ‘Designing a Uniform Meaning Representation for Natural Language Processing’. KI - Künstliche Intelligenz 35 (3): 343–60. https://doi.org/10.1007/s13218-021-00722-w.
More information about UMR can be found at UMR Project.
*U in UMR in this meaning representation area can be many things: Unified, Uniform, Universal, etc.