--- title: "Using cognates to align sentences in bilingual — Simard 1992" description: "Tools such as computer-assisted translation revision need sentence alignments of parallel texts that are nearly complete and correct." authors: ["Michel Simard", "George F. Foster", "Pierre Isabelle"] year: 1992 venue: "Proceedings of the Fourth Conference on Theoretical and Methodological Issues in Machine Translation of Natural Languages" source: "https://aclanthology.org/1992.tmi-1.7/" licence: "CC BY-NC-SA 3.0 (ACL Anthology, before 2016)" licence_url: "https://creativecommons.org/licenses/by-nc-sa/3.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/papers/simard-1992-cognates-align-sentences-bilingual-corpora/" --- [Elamur Research](https://md.research.elamur.ai/) · [this page in HTML](https://research.elamur.ai/free/papers/simard-1992-cognates-align-sentences-bilingual-corpora/) [home](https://md.research.elamur.ai/) / [papers](https://md.research.elamur.ai/free/papers/) / [1992](https://md.research.elamur.ai/free/papers/1992/) / Using cognates to align sentences in bilingual corpora # Using cognates to align sentences in bilingual corpora **Authors:** [Michel Simard](https://md.research.elamur.ai/free/authors/michel-simard/), George F. Foster, [Pierre Isabelle](https://md.research.elamur.ai/free/authors/pierre-isabelle/) · **Year:** 1992 · **Venue:** Proceedings of the Fourth Conference on Theoretical and Methodological Issues in Machine Translation of Natural Languages · **Licence:** [CC BY-NC-SA 3.0 (ACL Anthology, before 2016)](https://creativecommons.org/licenses/by-nc-sa/3.0/) **Links:** [Publisher's page](https://aclanthology.org/1992.tmi-1.7/) · [PDF](https://aclanthology.org/1992.tmi-1.7.pdf) · [OpenAlex](https://openalex.org/W1543107604) **Citations:** 289 citations (openalex-snapshot · 2026-10-14) · [107](#cited-by) in our library, of the 140673 papers whose bibliographies we read ## Summary Tools such as computer-assisted translation revision need sentence alignments of parallel texts that are nearly complete and correct. The length-only aligner of Gale and Church recovers about 96% of correct couples. However, it struggles when paired paragraphs have different sentence counts: it places contractions or expansions in the wrong spot and then misaligns the sentences around them. Simard, Foster and Isabelle add a cheap source of linguistic evidence, cognates. They define "cognateness" as the proportion of tokens in two segments that can be matched as cognates, a score that does not depend on segment length. Cognates are found automatically with crude rules covering numbers, punctuation and the first four letters of longer words. On a hand-aligned Canadian Hansards extract, true translation pairs score about three times higher than random pairs. The final aligner works in two passes. Length prunes unlikely candidates, and cognates then select the best alignment among the survivors. The gain over length alone is modest, but errors stay more local. The design also lets costlier knowledge sources join alignment without hurting efficiency. ## Key points - Test data for the cognate correlation: 13 English–French paragraph pairs from the Canadian Hansards (102 English and 94 French sentences). - Average cognateness was 0.21 for true translations versus 0.06 for random pairs with hand-identified cognates, and 0.30 versus 0.09 with the automatic definition. - Automatic cognate candidates fall into three groups: - alphanumeric tokens containing a digit, which must match exactly; - purely alphabetic words of at least four letters, which count as cognates if their first four characters agree; - single punctuation marks, which must match exactly. Case and accents are ignored. - On the error types that differ between methods (missed or misplaced contractions/expansions and miscellaneous errors), the two-pass program made three times fewer errors than the length-based one. The length-based program in turn made half as many as the cognate-only program. - All three programs failed on the same hard cases: all 5 omissions, 5 identical decomposition errors caused by periods that do not end a sentence, and unusual 3-to-1 and 4-to-1 patterns. Average missing couples per error region were 2.58 (length), 2.17 (two-pass) and 1.92 (cognate). ## Interesting ideas - **Operational cognates instead of a lexicon.** A simple shared-prefix rule plus exact matching of numbers and punctuation gives a usable cross-lingual similarity signal without any bilingual dictionary. This could work for quick bootstrapping of alignment between related languages that share an alphabet, such as Ukrainian–Russian, or for digit-heavy texts like trade documents. - **Cheap filter, expensive scorer.** First prune the search space with a fast, reliable criterion, then apply a richer but costlier criterion only to the remaining candidates. This cascade pattern applies to many matching or retrieval tasks. - **Combining methods with different failure modes.** The authors explain the two-pass gain by noting that length-based and cognate-based methods rarely make the same mistakes. Combining weak but uncorrelated signals can beat each one alone. - **Measuring error size, not just error count.** Counting missing couples per error region shows whether a method fails locally or derails over long stretches. This evaluation idea is useful for any sequence alignment or segmentation task. - **Score gaps as an error detector.** The authors found that first-pass scoring differences reliably locate difficult regions. Such gaps could flag doubtful sentence pairs for manual review or remove them when cleaning a parallel corpus. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* Fields: [Linguistics & languages](https://md.research.elamur.ai/free/shelves/linguistics/) · [Machine translation](https://md.research.elamur.ai/free/shelves/machine-translation/) ## Cited by Cited by 107 papers of our library cite it (of the 140673 whose bibliographies we read); 1 of them have a page here, 106 not yet. - [EtymoLink: A Structured English Etymology Dataset](https://md.research.elamur.ai/free/papers/lex-2024-gao-etymolink-english/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{simard1992using, title = {{Using cognates to align sentences in bilingual corpora}}, author = {Simard, Michel and Foster, George F. and Isabelle, Pierre}, booktitle = {Proceedings of the Fourth Conference on Theoretical and Methodological Issues in Machine Translation of Natural Languages}, year = {1992}, url = {https://aclanthology.org/1992.tmi-1.7/} } ``` - [← previous Tokenization as the Initial Phase in NLP](https://md.research.elamur.ai/free/papers/aclbulk-1992-webster-tokenization-as-initial-phase-nlp/) - [↑ back to papers of 1992](https://md.research.elamur.ai/free/papers/1992/) ## References 5 references, 1 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. Brown, Peter F., Jennifer C. Lai and Robert L. Mercer, 1991, "Aligning Sentences in Parallel Corpora", in COLING 91, pp. 169-176. [find](https://scholar.google.com/scholar?q=Brown%2C+Peter+F%2E%2C+Jennifer+C%2E+Lai+and+Robert+L%2E+Mercer%2C+1991%2C+%22Aligning+Sentences+in+Parallel+Corpora%22%2C+in+COLING+91%2C+pp%2E+169%2D176%2E) 2. Brown, Peter F., John Cocke, Stephen A. DellaPietra, Vincent J. DellaPietra, Frederick Jelinek, John D. Rafferty, Robert L. Mercer and Paul S. Roossin, 1990, "A Statistical Approach to Machine Translation", Computational Linguistics, 16(2), pp. 79-85, June 1990. [↗](https://doi.org/10.5555/92858.92860) 3. Gale, William A. and Kenneth W. Church, 1991, "A Program for Aligning Sentences in Bilingual Corpora", in COLING 91, pp. 177-184. [find](https://scholar.google.com/scholar?q=Gale%2C+William+A%2E+and+Kenneth+W%2E+Church%2C+1991%2C+%22A+Program+for+Aligning+Sentences+in+Bilingual+Corpora%22%2C+in+COLING+91%2C+pp%2E+177%2D184%2E) [our card](https://md.research.elamur.ai/free/papers/gale-1993-program-aligning-sentences-bilingual-corpora/) 4. Harris B., 1988, "Bi-Text, a New Concept in Translation Theory", in Language Monthly, #54, pp. 8-10. [find](https://scholar.google.com/scholar?q=Harris+B%2E%2C+1988%2C+%22Bi%2DText%2C+a+New+Concept+in+Translation+Theory%22%2C+in+Language+Monthly%2C+%2354%2C+pp%2E+8%2D10%2E) 5. Isabelle, Pierre, 1991, "Une nouvelle génération d'aides à la traduction et la terminologie", paper presented at the Terminologie et documentation colloquium, Hull, October 1991. Acknowledgments Many thanks go to Marc Dymetman for his constructive comments on the more technical aspects of the paper. Many thanks also to Peter Brown for pointing out important omissions in the original draft. Finally, we want to thank Marie-Louise Hannan, Jean-Marc Jutras, Elliott Macklovitch, François Perrault and Xiaobo Ren for their contribution in producing a 600 000-word hand-aligned corpus of bitext. [find](https://scholar.google.com/scholar?q=Isabelle%2C+Pierre%2C+1991%2C+%22Une+nouvelle+g%C3%A9n%C3%A9ration+d%27aides+%C3%A0+la+traduction+et+la+terminologie%22%2C+paper+presented+at+the+Terminologie+et+documentation+colloquium%2C+Hull%2C+October+1991%2E+Acknowledgments+Many+)