Elamur Research⌕
☰
home / papers / 2012 / The Parallel-TUT: a multilingual and multiformat treebank

The Parallel-TUT: a multilingual and multiformat treebank

Paper · 2012 · Cristina Bosco, Manuela Sanguinetti and Leonardo Lesmo

Summary

Parallel treebanks with alignment are valuable for translation and for projecting annotation to other languages, but they are expensive to build and alignment tools usually support only some levels and formats. The authors present ParTUT, an early-stage parallel treebank for Italian, English and French. Instead of building new machinery, they reused the Italian TUT pipeline: the rule-based TULE parser for native dependency annotation and the TUTtoPENNconverter for the Penn-style constituency version (TUT–Penn). The pilot data are the Universal Declaration of Human Rights, a sample of JRC-Acquis sentences and the Creative Commons licence. They export the data to Tiger–XML and CoNLL so that tools from other groups can use it. Sentence and word alignment were tried with existing tools, mostly by hand, to draft guidelines. The main contribution is a carefully annotated, multi-format gold sample, not a large training set. It is meant as a testbed for alignment based on predicate-argument structure and for comparing dependency-based and constituency-based alignment.

Key points

  • The source treebank, TUT, has more than 102,000 tokens (about 3,500 sentences) from newspaper, legal and Wikipedia texts.
  • The pilot ParTUT set contains 50 JRC-Acquis sentences, the full UDHR (about 100 sentences) and the Creative Commons licence, each in Italian, English and French.
  • Native TUT follows Word Grammar: determiners and prepositions act as heads of their nouns. Each label can have up to three layers, so PREP-RMOD-REASONCAUSE can be reduced to PREP-RMOD and then to RMOD.
  • TUT uses null elements, either co-indexed or free, for pro-drop, equi, ellipsis and gapping. This keeps the trees projective. TUT–Penn adds labels such as EXTPSBJ for post-verbal subjects.
  • Alignment tools: Omega Aligner for sentence alignment (TMX output) and WordAligner for word alignment, with links marked as exact or fuzzy, which correspond to sure and possible links.

Interesting ideas

  • I1. Dependency labels are built in layers (morphosyntactic, functional, semantic), so an annotator can stop at the level they are confident about, and the label set can be reduced when comparing with other resources.
  • I2. Explicit null nodes stand for dropped subjects and elided material, which makes missing arguments visible and keeps the trees projective.
  • I3. The same tools and annotation scheme are applied to every language in a parallel corpus, which makes cross-language comparison and later processing much simpler.
  • I4. One gold sample is kept in two paradigms, dependency and constituency, as a testbed for asking which representation supports alignment better.
  • I5. Word alignments are split into exact links (minimal translation units) and fuzzy links (valid but shifted translations), and are edited in a grid view rather than as lines.

For Mova

How these ideas could serve Mova, our small open language model — its files, stages and tests:

  • I1 (tree): Mova could attach an optional semantic suffix to UD relations, letting the parser keep plain UD labels while frames read the more specific layer when it is present.
  • I2 (frames): Mova could add implicit-argument nodes when building frames, for example the controlled subject of an infinitive or the predicate of a gapped clause, so that every role is either filled or explicitly marked as empty.
  • I3 (back): if translation through the meta-world uses one graph scheme for all languages, comparing graphs across languages becomes a direct structural diff.
  • I4 (learn): Mova could evaluate its witness parsers on the same sentences under several conversions, to check whether disagreements come from real ambiguity or only from the conventions of one format.
  • I5 (back): the text → graph → text compressor could score reconstruction with exact and fuzzy matches, so that a paraphrase counts as a partial match rather than a total failure.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

 Institutions

🎓University of TurinTurin · Italy27 works · 6,505 citations

📚 Shelves 4The field of science this work belongs to (OpenAlex fields).Computer Science · 9,085 worksArtificial Intelligence · 8,888 worksArts and Humanities · 9,296 worksLanguage and Linguistics · 9,206 works

🧭 Topics 4Research subjects and languages the work is about.English · 1,373 worksMultilingual & cross-lingual NLP · 465 worksTreebanks · 118 worksFrench · 64 works

Cite

Cite as Cristina Bosco, Manuela Sanguinetti and Leonardo Lesmo. 2012. The Parallel-TUT: a multilingual and multiformat treebank. Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12).

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

@inproceedings{bosco2012parallel,
  title = {{The Parallel-TUT: a multilingual and multiformat treebank}},
  author = {Cristina Bosco, Manuela Sanguinetti and Leonardo Lesmo},
  booktitle = {Proceedings of the Eighth International Conference on Language Resources and Evaluation (LREC'12)},
  year = {2012},
  doi = {10.63317/2zio4w62um52},
  url = {https://doi.org/10.63317/2zio4w62um52}
}

References

27 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. L. Ahrenberg, J. Tiedemann, and M. Volk, editors. 2010. Proceedings of the Workshop on Annotation and Exploitation of Parallel Corpora (AEPC) 2010. Tartu. find
  2. L. Ahrenberg. 2007. LinEs: an English-Swedish Parallel Treebank. In Proceedings of the 16th Nordic Conference on Computational Linguistics (NODALIDA ’07), Tartu.
  3. L. Ahrenberg. 2010. Clause restructuring in English-Swedish translation. In Proceedings of the Workshop on Annotation and Exploitation of Parallel Corpora (AEPC) 2010, Tartu. find
  4. A. Alicante, C. Bosco, A. Corazza, and A. Lavelli. 2012. A treebank-based study on the influence of italian word order on parsing performance. In Proceedings of the Language Resources and Evaluation Conference (LREC’12), Istanbul. find
  5. C. Bosco and A. Lavelli. 2010. Annotation schema-oriented validation for dependency parsing evaluation. In Proceedings of the 9th workshop on Treebanks and Linguistic Theories (TLT-9), Tartu. find
  6. C. Bosco. 2007. Multiple-step treebank conversion: from dependency to Penn format. In Proceedings of the Linguistic Annotation Workshop (LAW) 2007, Prague. find
  7. M. Buch-Kromann. 2007. Computing translation units and quantifying parallelism in parallel dependency treebanks. In Proceedings of the Linguistic Annotation Workshop (LAW) 2007, Prague. find
  8. T. Chung and D. Gildea. 2010. Effects of empty categories on machine translation. In Proceedings of Empirical Methods in Natural Language Processing - EMNLP’10, Boston. find
  9. J. Francom and M. Hulden. 2008. Parallel multi-theory annotation of syntactic structure. In Proceedings of the Language Resources and Evaluation Conference (LREC’08), Marrakech. find
  10. D. Gildea. 2004. Dependencies vs constituents for tree-based alignment. In Proceedings of Empirical Methods in Natural Language Processing - EMNLP’04, Barcelona. find
  11. J. Graça, J. P. Pardal, L. Coheur, and D. Caseiro. 2008. Building a golden collection of parallel multi-language word alignment. In Proceedings of the Language Resources and Evaluation Conference (LREC’08), Marrakech. find
  12. S. Grimes, X. Li, A. Bies, S. Kulick, X. Ma, and S. Strassel. 2010. Creating arabic-english parallel word-aligned treebank corpora at LDC. In Proceedings of Language Resources and Evaluation Conference (LREC’10), Malta. find
  13. J. Hajič and P. Zemánek. 2003. Prague Arabic Dependency Treebank: Development in data and tools. In Proceedings of the NEMLAR Conference on Arabic Language Resources and Tools, Cairo. find
  14. J. Hall and J. Nilsson. 2005. Converting Dependency Treebanks to MALT-XML. Technical report, Väjö University, School of Mathematics and Engeneering. find
  15. R. Hudson. 1984. Word grammar. Basil Blackwell, Oxford and New York. find
  16. P. Lambert, A. de Gispert, R. E. Banchs, and J. B. Mario. 2005. Guidelines for word alignment evaluation and manual alignment. Language Resources and Evaluation, 39(4). ↗
  17. L. Lesmo. 2007. The rule-based parser of the NLP group of the University of Torino. Intelligenza artificiale, 2(IV). find
  18. L. Lesmo. 2009. The Turin University Parser at Evalita 2009. In Proceedings of Evalita’09, Reggio Emilia. find
  19. B. Megyesi, B. Dahlqvist, E. Pettersson, and J. Nivre. 2008. Swedish-Turkish Parallel Treebank. In Proceedings of Language Resources and Evaluation Conference (LREC’08), Marrakech. find
  20. D. Melamed. 1998. Manual annotation of translational equivalence: The blinker project. Technical report, University of Pennsylvania. find
  21. H. Paulussen and L. Macken. 2010. Annotating the Dutch Parallel Corpus. In Proceedings of the Workshop on Annotation and Exploitation of Parallel Corpora (AEPC), Tartu. find
  22. A. Rios, A. Göhring, and M. Volk. 2009. A Quechua– Spanish parallel treebank. In Proceedings of 7th Workshop on Treebanks and Linguistic Theories (TLT-7), Groningen. find
  23. M. Sanguinetti and C. Bosco. 2011. Building the multilingual TUT parallel treebank. In Proceedings of the Workshop on Annotation and Exploitation of Parallel Corpora (AEPC) 2011, Hissar. find
  24. K. Simov, P. Osenova, L. Laskova, A. Savkov, and S. Kancheva. 2011. Bulgarian-english parallel treebank: Word and semantic level alignment. In Proceedings of Recent Advances in Natural Language Processing (RANLP), Hissar, Bulgaria. find
  25. R. Steinberger, B. Pouliquen, A. Widiger, C. Ignat, T. Erjavec, D. Tufiş, and D Varga. 2006. The JRC-Acquis: A multilingual aligned parallel corpus with 20+ languages. In Proceedings of Language Resources and Evaluation Conference (LREC’06), Genova.
  26. M. Čmejrek, J. Hajič, and V. Kuboň. 2004. Prague Czech-English Dependency Treebank: Syntactically Annotated Resources for Machine Translation. In Proceedings of EAMT 10th Annual Conference, Budapest. find
  27. M. Volk, A. Göhring, T. Marek, and Y. Samuelsson. 2010. SMULTRON (version 3.0) - The Stockholm MULtilingual parallel TReebank. An English-French-German-Spanish-Swedish parallel treebank with sub-sentential alignments. find

Updated

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/bosco-2012-parallel-tut-multilingual-multiformat-treebank/