Document Authoring the Bible for Minority Language Translation

What it is about

The paper describes The Bible Translator's Assistant (TBTA), a document-authoring and generation system built by the Summer Institute of Linguistics with a university group. A translation consultant writes the Bible text in a controlled English. The system then analyses it into a semantic representation, and a person corrects the analysis. The controlled language is strict: possessives are not allowed, and "of" must carry one of 22 semantic relations (ownership, kinship, made-of and so on); pronouns are used only where they are unambiguous; imperatives and questions are marked; every event is turned into a proposition; most figurative language is removed except where it has theological weight. Word sense is picked by the sense most often used earlier in the analysed corpus. Corrections are stored in the knowledge sources and used for later texts. Generation for a target language uses a lexicon that maps each word only to its basic concept, descriptive grammar rules for surface forms, and restructuring rules for the cases where a construction departs from the standard mapping. The system has been tested on English, Korean, Jula and Kewa, and several Old and New Testament books have been analysed. The paper reports use and a qualitative judgement, not measurements.

Ideas

• I1. A deterministic source rewrite before the reader: turn possessives into an explicit "of" relation and mark the proposition boundaries, then count how many reader errors disappear on Genesis. This is a rule set in tools/roundtrip or prag, not a model.
• I2. Restructuring rules as author deltas: TBTA keeps the standard mapping apart from the exceptions. Mova's delta has the same shape. Count how many overrides each author needs and how many tokens each one fixes, and compare with the three-token rule.
• I3. A generation trace: for each output token, record which profile context (concept+tag+neighbours, and so on) produced it, as TBTA logs which rules fired. Feed the trace into the residual by class (lexical, construction, order, function words).
• I4. The 22 relations of "of" are a closed inventory. A count of which relations the Bible uses for the genitive, by author, would show where the meta-world loses information when it drops the possessive form.

For Mova

TBTA is the nearest existing system to what tools/translate does at the meta level: a shared semantic representation, a per-language lexicon and grammar, and exceptions kept apart. Its main lesson for Mova is the trace (I3): the residual is only useful if each class of error can be traced to the rule that caused it. The controlled-language rewrite (I1) is a cheap test of how much of the reader's error comes from the source form rather than from the parser. The paper's own grammars are not for Mova.

Topics: Machine translation

Cite

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

@inproceedings{beale2005document,
  title = {{Document Authoring the Bible for Minority Language Translation}},
  author = {Beale, Stephen and Nirenburg, Sergei and McShane, Marjorie and Allman, Tod},
  booktitle = {Proceedings of Machine Translation Summit X: Papers},
  year = {2005},
  url = {https://aclanthology.org/2005.mtsummit-papers.9/}
}

References

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

1. B aker, Kathryn L., Alexander M. Franz, Pamela W. Jordan, Teruko Mitamura, Eric H. Ny berg. 1994. “Coping with Ambiguity in a Large-Scale Machine Tr anslation System,” In Proceedings of COLING-94, Kyoto, Japan. find
2. S tephen Beale, Sergei Nirenburg and Kavi Mahesh. 1995. Semantic analysis in the Mikrokosmos machine translation project. Proceedings of the 2nd Sym posium on Natural Language Processing. Bangkok, Thailand. find
3. B eale, Stephen and Sergei Nirenburg. 2003. “Just in time grammar.” In Proceedings 2003 International Multiconference in Computer Science and Computer Engineering, Las V egas, Nevada. find
4. M cShane, M., and Nirenburg, S. 2003. “Parameterizing and Eliciting Text Elements Across Languages for Use in Natural Language Processin g Systems,” In Machine Translation 18(2): 129-165. find
5. M cShane, M., Nirenburg, S., Cowie, J. and Zacharski, R. 2002. “Embedding Knowledge Elicitation and MT Systems within a Single Architecture.” I n Machine Translation 17(4): 271-305. find
6. M itamura, T., Nyberg, E. and Carbonell, J. 1991. “An Efficient Interlingua Translation System for Multilingual Document Production,” In Proceedin gs of MT Summit III, Washington, DC. find
7. Probst, Katharina, Lori Levin, Erik Petersen, Alon Lavie and Jamie Carbonell. 2003. “MT for Minority Languages Using Elicitation-Based Learning of Syntactic Transfer Rules.” In Machine Translation 17(4): 245-270. find
8. S ergei Nirenburg and Victor Raskin. 2004. Ontological Semantics. The MIT Press. 6 Only a small portion of this has been checked by consultants. find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/beale-2005-document-authoring-bible-minority-language/