Abstract Meaning Representation (AMR; link) is a semantic representation language. An AMR is a single-rooted, directed graph representing the meaning of a sentence.
This is a demo for AMREager, an AMR parser that works by processing a given sentence left-to-right, similarly to transition-based dependency parsers.
We thank our funders: Bloomberg and EU H2020 (grant agreement: 688139, SUMMA).
Graph visualization for this demo is done using AMRICA.
AMREager’s source code can be found on github. The multilingual parser can be found on github as well.
We also have developed a set of evaluation metrics for AMR.
The parser is described in detail in the paper that can be downloaded here.
@inproceedings{damonte-17,
title={An Incremental Parser for Abstract Meaning Representation},
author={Marco Damonte and Shay B. Cohen and Giorgio Satta},
booktitle={Proceedings of {EACL}},
year={2017}
}
The multilingual parser is described in detail in the following paper that can be found here.
@inproceedings{damonte-18,
title={Cross-lingual Abstract Meaning Representation Parsing},
author={Marco Damonte and Shay B. Cohen},
booktitle={Proceedings of {NAACL}},
year={2018}
}
Our parser also gives alignments in a comment below the sentence. Alignments should be interpreted as follows. For example,
1-2|0.0+0.1
means that the second token (x-y refers to the tokens that come between words x and y exclusive of y) aligned with the first child of the root node (0.0) and the second child of the root node (0.1). The address 0.x.y… refers to the path from the root to its xth child, then yth child, until the end of the path.
A full alignment would be an output such as
# ::alignments 1-2|0.0+0.1.1 3-4|0 5-6|0.1 6-7|0.1.0 8-9|0.1.0.0
appearing after the sentence line.