Creating a Universal Networking Language Module within an Advanced NLP System

What it is about

The paper describes a deconverter that turns Universal Networking Language (UNL) expressions into Russian, built inside the ETAP-3 system. A UNL expression is a directed hypergraph: nodes are Universal Words (English-based concept labels, optionally restricted), arcs are semantic relations such as agent, object, place and possession, and attributes such as present tense, definiteness and modal meaning sit on the nodes. ETAP-3 is transfer-based and works on a normalised syntactic structure, which is a tree with word order. The authors list where the two structures agree (both sit between surface and semantics, both have lexical nodes and attributes) and where they differ: a UNL node can be a subgraph, a UNL graph can contain loops, and UNL has no word order. The paper reports the module and its construction, not accuracy figures.

Ideas

• I1. Allow a meta-world node to be a subgraph for a multi-word concept. Keep the inner relations of the phrase, as the UNL hypernode does, and attach the outer relation to the phrase head. Test it on the phrase concepts already listed in meta.rs.
• I2. Count the modal and aspect attributes in the residual. UNL keeps modal meaning as attributes, and Mova's meta tokens need the same information for the translator to choose a target word. Measure how many residual tokens are modal.
• I3. Keep word order out of the meta-world. The paper's split is clear: the interlingua has no order, and order lives in the target syntax. In Mova that is the job of en/src/lin.rs. Measure the order residual separately, as the exact mode already does.

For Mova

The most useful point is the comparison of a graph interlingua with a tree-shaped target structure, which is the same problem as the meta-world's handling of heads and relations. The code is tools/translate (meta.rs) and en/src/lin.rs, with tools/roundtrip for the trees. The paper gives no accuracy measures, so the counts in I1 to I3 come first.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/990820.990833

@inproceedings{boguslavsky2000creating,
  title = {{Creating a Universal Networking Language Module within an Advanced NLP System}},
  author = {Boguslavsky, Igor and Frid, Nadezhda and Iomdin, Leonid and Kreidlin, Leonid and Sagalova, Irina and Sizov, Victor},
  booktitle = {COLING 2000 Volume 1: The 18th International Conference on Computational Linguistics},
  year = {2000},
  doi = {10.3115/990820.990833},
  url = {https://doi.org/10.3115/990820.990833}
}

References

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

1. Apresjan Ju.D., I.M.Boguslavsky, L.L.Iomdin et al. (1992). ETAP-2: The Linguistics of a Machine Translation System. // META, Vol. 37, No 1, pp. 97-112. find
2. Boguslavsky I.(1995). A bi-directional Russian-to-English machine translation system (ETAP-3). // Proceedings of the Machine Translation Summit V. Luxembourg. find
3. Iomdin L.& O. Streiter. (1999). Learning from Parallel Corpora: Experiments in Machine Translation. // Dialogue'99: Computational Linguistics and its Applications International Workshop. find
4. Tarusa, Russia, June 1999. Vol.2, pp. 79-88. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/boguslavsky-2000-creating-universal-networking-language-module/