Treebank of Chinese Bible Translations

What it is about

The paper describes a treebank of eight modern Chinese translations of the Bible, about 7.7 million characters in all, built by a parser rather than by hand. The trees follow the Penn Chinese Treebank guidelines, with two changes: the tree is organised around the verb (a VP for every verbal unit, without CP and IP nodes), and word-internal structure is kept, so that one unit can be read as one word or as two. Each verse gets its own tree, because verses are not always sentences, and the trees are "maximal parses" that leave clause relations open. The key idea is dynamic treebanking: when a tree is wrong, the author does not edit the tree; he changes the grammar rule, the dictionary entry or the tree ranking, and regenerates. The parser is an HPSG grammar with a domain dictionary of about 110,000 words. Segmentation matches the gold standard in 98.13 per cent of the 31,151 verses with no error, and the other versions gain from the fixes made on the one used for development. The paper gives no parsing gold standard.

Ideas

• I1. Dynamic repair for the Hebrew reader: when MACULA or tools/hbo attaches a word wrongly, fix the rule or the lexicon entry, regenerate, and count the verses that change. Open problem 1 (reader attachment noise) is then answered with a count of changed verses per fix, not with hand edits.
• I2. One tree per verse with maximal, underspecified clauses. The meta-reader already aligns verses, so test whether reading each verse as its own maximal parse, with clause links left open, lowers the reader's residual on the Bible pairs.
• I3. Keep alternative cuts of multi-word units with a type (compound, verb-result, noun-suffix). This matches open problem 3: the encoder should keep each chunk's inner relations, and the type tells which cut to use per author.
• I4. Store frequent phrases with their correct analysis, looked up like dictionary entries. For Mova, this is the phrase level of the lexicon (e.g. unto_thou), which should carry a fixed parse, not a merged concept with no structure.

For Mova

The paper's method is the one Mova wants for its reader: rules and lexicon are the place of a fix, and the tree is regenerated. The code to read is tools/hbo and tools/macula (the trees), tools/roundtrip (the stack) and tools/translate/hebrew.rs. The Chinese grammar is not for Mova; the process is. Measure per fix: verses changed, bits of residual changed, and whether the change holds on the second run.

Topics: Treebanks · Machine translation

Cite

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

@inproceedings{wu2010treebank,
  title = {{Treebank of Chinese Bible Translations}},
  author = {Wu, Andi},
  booktitle = {CIPS-SIGHAN Joint Conference on Chinese Language Processing},
  year = {2010},
  url = {https://aclanthology.org/W10-4118/}
}

References

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

1. Sproat, Richard and Thomas Emerson. 2003. The First International Chinese Segmentation Bakeoff. In Proceedings of the Second SIGHAN Workshop on Chinese Language Processing, July 11-12, Sapporo, Japan. find
2. Wu, Andi, J. and Z. Jiang, 1998. Word segmentation in sentence analysis, in Proceedings of 1998 International Conference on Chinese Information Processing, pp. 46--51. 169--180, Beijing, China. find
3. Wu, Andi. 2003. Customizable Segmentation of Morphological Derived Words in Chinese. In International Journal of Computational Linguistics and Chinese Language Processing, 8(1):1-27. find
4. Wu, And, Arron Ma, Dong Wang. 2009. Fidelity and Readablity – a quanatative comparison of Chinese translations of the New Testament. Proceedings of the Conference on “Interpretation of Biblical Texts in Chinese Contexts”, Sichuan Univeristy, December 2009. find
5. Xia, Fei. 2000. Segmentation Guidelines for the Penn Chinese Treebank (3.0). Technical Report, University of Pennsylvania. find
6. Xia, Fei. 2000. The Part-Of-Speech Tagging Guidelines for the Penn Chinese Treebank (3.0). Technical Report, University of Pennsylvania. find
7. Xue, Nianwen and Fei Xia. 2000. The Bracketing Guidelines for the Penn Chinese Treebank (3.0). Technical Report, University of Pennsylvania. find
8. Oepen, Stephan, Dan Flickinger, Kristina Toutanova, Christoper D. Manning. 2002. LinGO Redwoods: A Rich and Dynamic Treebank for HPSG In Proceedings of The First Workshop on Treebanks and Linguistic Theories (TLT2002), Sozopol, Bulgaria. find our card

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/wu-2010-treebank-chinese-bible-translations/