LinES: An English-Swedish Parallel Treebank

What it is about

The paper presents LinES, an English-Swedish parallel treebank that was still under development. It is built on the Linköping Translation Corpus and has dependency syntax, part-of-speech and morphology for both languages, with word alignments that were checked by hand. The aim is to study translation shifts, that is, where a translation changes structure. Two sub-corpora were ready: Access (software help texts) and Bellow (a novel), with 600 sentence pairs each. The paper describes the file format, including null links for level shifts, and a link-based query interface. Tree-based search was still being built.

Ideas

I1. en/src/convert.rs: the paper's rule that level shifts are coded as null links gives a shift class that can be counted. Count English tokens that have no counterpart in a translation, per sub-corpus, as a test set for the converter's dialect checks. Measure on the test split only; the Swedish side is not used in the build.
I2. en convert-check: LinES uses structure-oriented labels, for example prepositional objects and particles split off from adverbials. List the English label pairs where the paper's scheme and UD differ, and check that the converter maps each one the same way on every run.
I3. Word order probe for en lin-eval: take the English side of the link pairs where the word order changes and check whether the generator from trees reproduces the English order. Report the share of matches on the test split.

For Mova

The chapters are languages/en/29-ellipsis-word-order-information-structure.md (word order) and languages/en/37-register-and-variation.md. The code is en/src/lin.rs and en/src/convert.rs. The data is Swedish-English and NC: use it for tests and measurements only, never in what is built. Measure first how many English tokens have no counterpart in translation.

Library summary

The Research library's own summary of the paper, in our words — not the paper's text.

Written by Mova (claude-opus-5-5) from the paper text, in our own words: the license "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" does not allow redistributing the text itself. Original — aclanthology.org/W07-2441.pdf

Summary

The paper describes LinES (Linköping English-Swedish Parallel Treebank), a parallel English–Swedish treebank that was still under development at the time of publication. The resource was created to study how typical syntactic constructions change in translation from English into Swedish. Particular attention is paid to structural translation shifts (in Catford's sense). It is based on the LTC (Linköping Translation Corpus). The sentences were first parsed with Connexor's Machinese Syntax parsers, and the parser output was then converted into the project's own dependency scheme. This scheme is projective, shared by both languages, and oriented towards structure rather than semantics. Word-level alignments were checked manually in the I*Link tool. The author proposed a web interface for searching by links between aligned nodes. The author also proposed formal notions for future subtree search: inclusive and single-level subtrees, and isomorphism between a subtree and its counterpart. The resource is useful for translation studies, for evaluating and training parsers for several languages, and for machine translation, since it is precisely the shifts that make translation difficult.

Key points

The current version has two subcorpora of 600 sentence pairs each: Access (MS Access help) and Bellow (Saul Bellow's novel "Jerusalem and Back"). There are plans to add 1–2 more genres, with about 1000 pairs from each text.
Each word has the attributes base, pos, msd, func and fa. Each segment has a single root, and the analysis is projective, so it can easily be converted into a flat constituent structure.
Word alignments are stored either in XML or in a compact five-number format. Empty correspondences (null links) are marked with the pair (-1,-1). The alignment guidelines prefer many small links, so level shifts are encoded through null links.
The web interface already supports search by links, with constraints on base, POS, morphology and dependency for one or both languages. An example query is an English apposition that became a subject in the Swedish text. Tree-based search is still under development.
The author compares LinES with FuSe, CroCo and SMULTRON. Unlike them, LinES aims at full alignment of segment pairs and deliberately marks non-equivalent correspondences.

Topics: Treebanks

Cite

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

@inproceedings{ahrenberg2007lines,
  title = {{LinES: An English-Swedish Parallel Treebank}},
  author = {Ahrenberg, Lars},
  booktitle = {Proceedings of the 16th Nordic Conference of Computational Linguistics (NODALIDA 2007)},
  year = {2007},
  url = {https://aclanthology.org/W07-2441/}
}

References

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

J. C. Catford 1965. A Linguistic Theory of Translation: An Essay in Applied Linguistics, London, Oxford University Press Lea Cyrus 2006. Building a resource for studying translation shifts. Proceedings of LREC 2006, Genoa, May 24–26, 2006 1240-1245. find
Silvia Hansen-Schirra, Stella Neumann, and Mihaela Vela. 2006. Multi-dimensional Annotation and Alignment in an English-German Translation Corpus. Proceedings of the EACL Workshop on Multidimensional Markup in Natural Language Processing (NLPXML-2006), Trento, Italien, 4. April 2006, 35-42. find
Magnus Merkel 1999. Understanding and enhancing translation by parallel text processing. Linköping Studies in Science and Technology, Dissertation No. 607. find
Magnus Merkel, Michael Petterstedt, and Lars Ahrenberg. Interactive Word Alignment for Corpus Linguistics. Proceedings of Corpus Linguistics 28-31 March, 2003, Lancaster. UK. find
Joakim Nivre, Koenraad de Smedt, and Martin Volk 2005. Treebanking in Northern Europe: A White Paper. find
Holmboe, Henrik (ed.) Nordisk Sprogteknologi 2004: Årbog for Nordisk Sprogteknologisk Forskningsprogram 2000-2004, 97-112. København: Museum Tusculanums Forlag. find
Yvonne Samuelsson and Martin Volk. 2006. Phrase Alignment in Parallel Treebanks Jan Hajiĉ and Joakim Nivre (eds.) Proceedings of the Treebank in Linguistic Theory Workshop, Prague, Czech Republic, December 2006. 91-102. find
Douglas T. L. Rhode 2004. TGrep2 - the next-generation search engine for parse trees. http://tedlab.mit.edu/˜dr/Tgrep2/. ↗
Martin Volk, Sofia Gustafson-Capková, Joakim Lundborg, Torsten Marek, Yvonne Samuelsson, and F. Tid-ström. 2006. XML-Based Phrase Alignment in Parallel Treebanks. Proceedings of EACL Workshop on Multi-dimensional Markup in Natural Language Processing, Trento, Italy, April 2006. find