Generation in Machine Translation from Deep Syntactic Trees

What it is about

The authors generate English from Praguian tectogrammatical trees, which keep content words, functors and grammatemes and drop the function words. Generation has two statistical parts. First, classifiers predict the auxiliary words that the tree lacks (articles, prepositions and subordinating conjunctions, and modal verbs by a fixed map), each from the node's lemma, tag and functor and its governor's. When a classifier is unsure (probability below 0.5), a few fixed rules apply, such as to before infinitives and of before actors. Second, a bottom-up procedure orders each head's children, scoring pairs by tag and functor, and places the governor. On PCEDT 1.0 the article accuracy was about 89–90% on manual trees and the preposition and conjunction accuracy about 89–92%; reordering was scored by whether all children of a node come out in order. The BLEU of the full system was about 0.46 on manual trees.

Ideas

• I1. Insertion from frame roles. For each meta token, predict the function word (article, preposition) from its concept, tag and frame role (prag), with governor features, through a backoff chain of count tables in BTreeMap form. Measure the accuracy per class on dev, as the paper reports it.
• I2. Fallbacks from counts. Where the top count probability is below 0.5, let the encoder fall back to the most frequent insertion for the same role, instead of a hand list. Compare both at the same threshold.
• I3. A strict reorder metric. Score the linearizer (en/src/lin.rs) with the paper's rule that a node counts as right only when all its children are in order, a harder measure than pairwise order.
• I4. Article definiteness from the text. The paper inserts the when the noun appeared in the previous four sentences. A verse window of the same kind, with the source text's earlier mentions, is a feature for the encoder's article class.

For Mova

The paper is a generator from a deep tree with few words, which is close to the meta-world read from frames. Its strict reorder metric and its classifier-accuracy tables are measures Mova can copy, and the fallback to counts matches Mova's rule of learning from data rather than from hand lists.

Topics: Machine translation

Cite

The paper itself, not this card. BibTeX from the library's record of the work. DOI 10.3115/1626281.1626289

@inproceedings{hall2007generation,
  title = {{Generation in Machine Translation from Deep Syntactic Trees}},
  author = {Hall, Keith and Němec, Petr},
  booktitle = {Proceedings of SSST, NAACL-HLT 2007 / AMTA Workshop on Syntax and Structure in Statistical Translation},
  year = {2007},
  doi = {10.3115/1626281.1626289},
  url = {https://doi.org/10.3115/1626281.1626289}
}

References

15 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

1. Srinivas Bangalore and Owen Rambow. 2000. Exploiting a probabilistic hierarchical model for generation. In Proceedings of the 18th International Conference on Computational Linguistics (COLING 2000), Saarbrücken, Germany. find
2. Alena Böhmová, Jan Hajič, Eva Hajičová, and Barbora Vidová Hladká. 2002. The prague dependency treebank: Three-level annotation scenario. In Anne Abeille, editor, In Treebanks: Building and Using Syntactically Annotated Corpora. Dordrecht, Kluwer Academic Publishers, The Neter-lands. ↗
3. Joan Bresnan and Ronald M. Kaplan. 1982. Lexical-functional grammar: A formal system for grammatical representation. In The Mental Representation of Grammatical Relations. MIT Press. find
4. Eugene Charniak and Mark Johnson. 2005. Coarse-to-fine nbest parsing and MaxEnt discriminative reranking. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics. find
5. David Chiang. 2005. A hierarchical phrase-based model for statistical machine translation. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics, pages 263–270, Ann Arbor, MI. find our card
6. Michael Collins. 2003. Head-driven statistical models for natural language processing. Computational Linguistics, 29(4):589–637. find
7. Simon Corston-Oliver, Michael Gamon, Eric Ringger, and Robert Moore. 2002. An overview of Amalgam: A machine-learned generation module. In Proceedings of the International Natural Language Generation Conference, pages 33–40, New York, USA. find
8. Jan Hajič, Martin Čmejrek, Bonnie Dorr, Yuan Ding, Jason Eisner, Dan Gildea, Terry Koo, Kristen Parton, Dragomir Radev, and Owen Rambow. 2002. Natural language generation in the context of machine translation. Technical report, Center for Language and Speech Processing, Johns Hopkins University, Balitmore. Summer Workshop Final Report. find
9. Irene Langkilde and Kevin Knight. 1998. The practical value of n-grams in generation. In Proceedings of the International Natural Language Generation Workshop. find
10. Irene Langkilde-Geary. 2002. An empirical verification of coverage and correctness for a general-purpose sentence generator. In Proceedings of the International Natural Language Generation Conference. find
11. Carl Pollard and Ivan A. Sag. 1994. Head-Driven Phrase Structure Grammar. University of Chicago Press. find
12. Chris Quirk, Arul Menezes, and Colin Cherry. 2005. Dependency treelet translation: Syntactically informed phrasal SMT. In Proceedings of the 43rd Annual Meeting of the Association for Computational Linguistics (ACL’05), pages 271–279, Ann Arbor, Michigan, June. Association for Computational Linguistics. find our card
13. Petr Sgall, Eva Hajičová, and Jarmila Panevová. 1986. The Meaning of the Sentence in Its Semantic and Pragmatic Aspects. Kluwer Academic, Boston. find
14. Radu Soricut and Daniel Marcu. 2006. Stochastic language generation using WIDL–expressions and its application in machine translation and summarization. In Proceedings of the 44th Annual Meeting of the Association for Computational Linguistics. find
15. Dekai Wu. 1997. Stochastic inversion transduction grammars and bilingual parsing of parallel corpora. Computational Linguistics, 23(3):377–404. find our card

This page in Markdown → https://md.mova.elamur.ai/free/en/docs/papers/hall-2007-generation-machine-translation-deep-syntactic/