A Fast and Portable Realizer for Text Generation Systems

What it is about

A technical note on REALPRO, a C++ surface realiser built on Meaning-Text Theory. Its input is a deep-syntactic dependency tree of lexemes only: meaning-bearing words, labelled syntactic arcs, no function words, and no lexical choice, so the input fixes the output and the system is deterministic. Internally it runs a chain of modules, each driven by an ASCII rule base read at run time: a checker and default-feature step, a deep-syntactic component that inserts auxiliaries and governed prepositions, a surface-syntactic component that linearises by arc-label precedence rules, a morphology component that inflects, and a graphical component for punctuation and formatting, with formatters for ASCII, HTML and RTF. Default word order, topicalisation by features, full English pronoun morphology and commas around relative clauses are covered. Clefts and long-distance dependencies are not. The paper argues for speed and portability and gives no accuracy figures.

Ideas

• I1. The layered design (lexeme tree, then function words, then linearisation, then inflection) is a clean split for the translator's encoder. Make the meta-world tree the input, insert function words with a rule table, and keep a separate residual for function-word insertion, which exact mode can then report on its own.
• I2. Feed the encoder a tree with no function words, as REALPRO requires. The meta-world already drops spelling and preferred constructions; dropping function words too is a test. Strip them from the KJV meta stream and count how many bits the encoder then spends on function words, against the current figure.
• I3. A rule list for linear precedence per relation (subject before verb group, indirect before direct object) is a small deterministic baseline next to the Markov order in en/src/lin.rs. Compare the exact order match on dev verses, run twice.
• I4. Keep every rule base as an ASCII data file read at run time, so an author delta can be swapped without a rebuild. Check that output stays byte-identical across two runs with two different delta files loaded in turn.

For Mova

The code is en/src/lin.rs (linearisation) and tools/translate/src/profile.rs (the standard encoder). The layering touches open problem 4 (function words as features) and problem 3 (the encoder should write trees). The paper's coverage gaps (clefts, long-distance dependencies) are a reminder to list what the translator cannot write before claiming a rule set is complete.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/974557.974596

@inproceedings{lavoie1997fast,
  title = {{A Fast and Portable Realizer for Text Generation Systems}},
  author = {Lavoie, Benoit and Rainbow, Owen},
  booktitle = {Fifth Conference on Applied Natural Language Processing},
  year = {1997},
  doi = {10.3115/974557.974596},
  url = {https://doi.org/10.3115/974557.974596}
}

References

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

1. Bateman, J. A. (1996). KPML development environment. Technical report, Institut ffir Integrierte Publikations- und Informationssysteme (IPSI), GMD. find
2. Coch, J. (1996). Overview of AlethGen. In Proceedings of the Eighth International Natural Language Generation Workshop (INLG '96) (Demonstrations Volume), Herstmonceux Castle, Sussex. find
3. Elhadad, M. and Robin, J. (1996). An overview of SURGE: a reusable comprehensive syntactic realization component. In Proceedings of the Eighth International Natural Language Generation Workshop (INLG '96) (Demonstrations Volume), Herstmonceux Castle, Sussex. find our card
4. Iordanskaja, L., Kim, M., Kittredge, R., Lavoie, B., and Polgu~re, A. (1992). Generation of extended bilingual statistical reports. In Proceedings of the l~th International Conference on Computational Linguistics (COLXNG'9~). find
5. Iordanskaja, L., Kittredge, R., and Polgu~re, A. (1988). Implementing the Meaning-Text Model for language generation. Paper presented at COLING-88. find
6. Lavoie, B., Rainbow, O., and Reiter, E. (1997). Customizable descriptions of object-oriented models. In Proceedings of the Conference on Applied Natural Language Processing (ANLP'97), Washington, DC. find
7. Mel'~uk, I. A. (1988). Dependency Syntax: Theory and Practice. State University of New York Press, New York. find
8. Rainbow, O. and Korelsky, T. (1992). Applied text generation. In Third Conference on Applied Natural Language Processing, pages 40-47, Trento, Italy. find
9. Reiter, E. (1994). Has a consensus NL generation architecture appeared, and is it psycholinguistically plausible? In Proceedings of the 7th International Workshop on Natural Language Generation, pages 163-170, Maine. find
10. White, M. (1995). Presenting punctuation. In Proceedings of the FiSh European Workshop on Natural Language Generation (EWNLG5). find

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/lavoie-1997-fast-portable-realizer-text-generation/