Mapping AMR to UMR: Resources for Adapting Existing Corpora for Cross-Lingual Compatibility

What it is about

A reference paper that lists how AMR structures map onto UMR, the typologically motivated successor to AMR. It goes through the general roles, rolesets and abstract concepts. Some AMR roles are kept, some are renamed or narrowed, and some are split. For example, the AMR destination role is split into goal and recipient by animacy, and the source role into three roles for start point, material and source. Every role in UMR has an inverse and a reifying roleset. Abstract rolesets such as have-rel-role-91 and include-91 are mapped to UMR counterparts. The paper gives illustrated figures and examples from Czech and Chinese corpora. It argues that much of the conversion of an AMR corpus into UMR can be done deterministically, and that the hard cases are the new UMR coverage for tense, aspect, modality and scope. There are no experiments.

Ideas

• I1. Role splits as rules. Write the role splits, such as destination into goal or recipient by animacy, as deterministic rules over an enum of roles. Count how often each split fires on the Bible pairs before anything is learned.
• I2. Abstract predicates as enums. Keep abstract concepts such as identity-91 as enum values in the meta-world, so the encoder knows they are not words to output. This makes the function-word question in open problem 4 concrete.
• I3. Derived inverses. Every UMR role has an inverse. Store only the forward edge in the meta-world and derive the inverse, which may cut the size of the delta tables.

For Mova

The relevant part is the role inventory and how it is split. In Mova, these are the relation labels of the meta-world, in tools/translate/src/meta.rs. Roles are enums, and the rules are compared as integers, as the project's conventions require. The split and inverse rules bear on open problem 4 (function words and boundaries as features) and on the coarse relation set in open problem 1.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work.

@inproceedings{bonn2023mapping,
  title = {{Mapping AMR to UMR: Resources for Adapting Existing Corpora for Cross-Lingual Compatibility}},
  author = {Bonn, Julia and Myers, Skatje and Gysel, Jens E. L. Van and Denk, Lukas and Vigus, Meagan and Zhao, Jin and Cowell, Andrew and Croft, William and Hajič, Jan and Martin, James H. and Palmer, Alexis and Palmer, Martha and Pustejovsky, James and Urešová, Zdenka and Vallejos, Rosa and Xue, Nianwen},
  booktitle = {Proceedings of the 21st International Workshop on Treebanks and Linguistic Theories (TLT, GURT/SyntaxFest 2023)},
  year = {2023},
  url = {https://aclanthology.org/2023.tlt-1.8/}
}

References

1 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Rafael Anchiêta and Thiago Pardo. 2018. Towards AMR-BR: A SemBank for Brazilian Portuguese language. In Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018), Miyazaki, Japan. European Language Resources Association (ELRA). find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/bonn-2023-mapping-amr-umr-resources-adapting/