--- title: "A Statistical Approach to Language Translation — Brown 1988" description: "The paper proposes treating French-to-English machine translation as a statistical learning problem rather than one solved by hand-written grammars or an…" authors: ["P. Brown", "J. Cocke", "S. Della Pietra", "V. Della Pietra", "F. Jelinek", "R. Mercer", "P. Roossin"] year: 1988 venue: "Coling Budapest 1988 Volume 1: International Conference on Computational Linguistics" source: "https://aclanthology.org/C88-1016/" 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/en/papers/brown-1988-statistical-approach-language-translation/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/brown-1988-statistical-approach-language-translation/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [1988](https://md.research.elamur.ai/free/en/papers/1988/) / A Statistical Approach to Language Translation # A Statistical Approach to Language Translation **Authors:** P. Brown, J. Cocke, S. Della Pietra, V. Della Pietra, F. Jelinek, R. Mercer, P. Roossin · **Year:** 1988 · **Venue:** Coling Budapest 1988 Volume 1: International Conference on Computational Linguistics · **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/C88-1016/) · [PDF](https://aclanthology.org/C88-1016.pdf) · [OpenAlex](https://openalex.org/W2048390999) **Citations:** 275 citations (openalex-snapshot · 2026-10-14) · [64](#cited-by) in our library, of the 140673 papers whose bibliographies we read ## Summary The paper proposes treating French-to-English machine translation as a statistical learning problem rather than one solved by hand-written grammars or an interlingua. The IBM team relies on a large parallel corpus, the bilingual Hansard proceedings of the Canadian parliament (about 30 million words), with sentences aligned between the two languages. Translation is framed in three stages: split the source sentence into fixed locutions (single words or idiomatic, possibly discontinuous phrases), choose target-language counterparts from an automatically learned glossary using context, and then order the resulting target words. The authors describe ways to build the glossary. The first counts word co-occurrences across aligned sentences and ranks them by mutual information. The second is a crude generative model of how English words arise from French ones, with probabilities estimated from the most likely word correspondence pattern. The project had produced no full translations yet. However, an experiment showed that a trigram language model alone often recovers correct English word order. This early paper laid the groundwork for later statistical MT. ## Key points - The training data is the English/French Hansard corpus of Canadian parliamentary proceedings, roughly 30 million words. French and English were chosen partly because their word orders are similar. - In the first glossary approach, each French word spreads fractional counts (1/n) over all n words of the aligned English sentence. Candidate pairs are then normalized by overall English word frequency, which amounts to ranking them by mutual information. - The model-based glossary adds a NULL word and estimates three things: word translation probabilities, the probability that an English word spawns k extra words, and which secondary words appear. Counts come from the single best correspondence pattern in each sentence pair. - For reordering, sentences of 10 words or fewer were scrambled and every permutation was scored with an English trigram model. The original order won in 63% of cases, and the top-scoring order kept the meaning in 79% of cases. - Decision trees are suggested for choosing contextual guide words that disambiguate among translations, such as the many possible renderings of "affecting". ## Interesting ideas - **Frequency normalization against function words.** Raw co-occurrence counts favour words like "the" that appear in almost every sentence. Dividing by baseline frequency (mutual information) fixes this, and the same trick applies to any alignment or association mining between paired data, such as lexicon induction for low-resource languages. - **Greedy "explain away" elimination.** The method fixes the highest-scoring pair, removes both words from their sentences, and recomputes the scores. This cheap heuristic stops one target word from being credited to several source words, and it works well for quick bilingual lexicon bootstrapping. - **An oracle-style upper-bound experiment.** The authors assume the correct bag of target words is already known and test only whether word order can be recovered. Isolating one pipeline stage under optimistic conditions like this is a useful way to check whether a component is viable before building the whole system. - **An explicit NULL token and a fertility-like count in alignment.** Letting a source word generate zero, one, or several target words captures insertions, deletions, and one-to-many mappings. This is a reusable design choice for any sequence-to-sequence alignment model. - **Context-sensitive translation choice through learned guide words.** Decision trees pick the few context words that best predict which translation is right, which gives an interpretable form of word-sense disambiguation for building dictionaries or glossaries. *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/en/shelves/linguistics/) · [Machine translation](https://md.research.elamur.ai/free/en/shelves/machine-translation/) · [Natural language processing](https://md.research.elamur.ai/free/en/shelves/nlp/) ## Cited by Cited by 64 papers of our library cite it (of the 140673 whose bibliographies we read); 8 of them have a page here, 56 not yet. - [Automatic Processing of Large Corpora for the Resolution of Anaphora References](https://md.research.elamur.ai/free/en/papers/dagan-1990-automatic-processing-large-corpora-resolution/) - [Statistical Machine Translation with Scarce Resources Using Morpho-syntactic Information](https://md.research.elamur.ai/free/en/papers/en-2004-statistical-machine-translation-scarce-resources/) - [A Program for Aligning Sentences in Bilingual Corpora](https://md.research.elamur.ai/free/en/papers/gale-1993-program-aligning-sentences-bilingual-corpora/) - [Research methods and system designs in machine translation: a ten-year review, 1984-1994](https://md.research.elamur.ai/free/en/papers/hutchins-1994-research-methods-system-designs-machine/) - [Datasets: A Community Library for Natural Language Processing](https://md.research.elamur.ai/free/en/papers/lhoest-2021-datasets-community-library-natural-language/) - [Creating a massively parallel Bible corpus](https://md.research.elamur.ai/free/en/papers/mayer-2014-creating-massively-parallel-bible-corpus/) - [Retrieving Collocations from Text: Xtract](https://md.research.elamur.ai/free/en/papers/smadja-1993-retrieving-collocations-text-xtract/) - [A Syntax-based Statistical Translation Model](https://md.research.elamur.ai/free/en/papers/yamada-2001-syntax-based-statistical-translation-model/) ## Cite The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{brown1988statistical, title = {{A Statistical Approach to Language Translation}}, author = {Brown, P. and Cocke, J. and Pietra, S. Della and Pietra, V. Della and Jelinek, F. and Mercer, R. and Roossin, P.}, booktitle = {Coling Budapest 1988 Volume 1: International Conference on Computational Linguistics}, year = {1988}, doi = {10.3115/991635.991651}, url = {https://doi.org/10.3115/991635.991651} } ``` - [← previous 'Relaxed' compositionality in machine translation](https://md.research.elamur.ai/free/en/papers/arnold-1988-relaxed-compositionality-machine-translation/) - [next → An Architecture for Anaphora Resolution](https://md.research.elamur.ai/free/en/papers/rich-1988-architecture-anaphora-resolution/) - [↑ back to papers of 1988](https://md.research.elamur.ai/free/en/papers/1988/) ## References 6 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. 111 L.R. Bahl, F. Jclinek, and R.l,. Mercer: A maximum likelihood approach to contimlous speech recognition, IEEE Traosaclioos on Pattern Analysis and Machine Intelligence, PAM I-5 (2): 179-190, March 1983. 12] .I.K. Baker: Stochastic modeling for automatic speech tmdcrstanding. In R.A. P, eddy, edi… [find](https://scholar.google.com/scholar?q=111+L%2ER%2E+Bahl%2C+F%2E+Jclinek%2C+and+R%2El%2C%2E+Mercer%3A+A+maximum+likelihood+approach+to+contimlous+speech+recognition%2C+IEEE+Traosaclioos+on+Pattern+Analysis+and+Machine+Intelligence%2C+PAM+I%2D5+%282%29%3A+179%2D190%2C+March) 2. Princeton, New Jersey, IDA-CRD, Oct. 1980, pp. 8-15 14] J. Metl. Sinclair: "Lcxicogral~hic F.vidence" in, I)ielionarie,~, Lexicography and Langaage l,earniog (l!1+'F Doeoments: 1211), editol R. llson, NewYork: Pergamon Press, pp. 81-94, 1985. 151 P,.G. Garsidc, G.N. Lccch and (].P,. Sampson, The Com… [find](https://scholar.google.com/scholar?q=Princeton%2C+New+Jersey%2C+IDA%2DCRD%2C+Oct%2E+1980%2C+pp%2E+8%2D15+14%5D+J%2E+Metl%2E+Sinclair%3A+%22Lcxicogral%7Ehic+F%2Evidence%22+in%2C+I%29ielionarie%2C%7E%2C+Lexicography+and+Langaage+l%2Cearniog+%28l%211%2B%27F+Doeoments%3A+1211%29%2C+editol+R%2E+llson%2C) 3. [10] R.G. Gallager: Informalion Theory aad reliahle (ommuniealion, John Wiley and Sons, Ii1c,, New York, 1968. [find](https://scholar.google.com/scholar?q=%5B10%5D+R%2EG%2E+Gallager%3A+Informalion+Theory+aad+reliahle+%28ommuniealion%2C+John+Wiley+and+Sons%2C+Ii1c%2C%2C+New+York%2C+1968%2E) 4. [I I1 A.P. Dcmpstcr, N.M.l.aird, al/d It.B. ll.ubin: Maximum likelihood from ineolnpletc data via tile I"M algorithm, Journal of Ihe Royal S\|atistical Society, 3 9 ( B ) : 1-38, 1977. 1121 A.J, Viterbi: Error bounds Ior conw)httional codes and an asylntotically optimum decoding algorithm, 11,'1.;1,~… [find](https://scholar.google.com/scholar?q=%5BI+I1+A%2EP%2E+Dcmpstcr%2C+N%2EM%2El%2Eaird%2C+al%2Fd+It%2EB%2E+ll%2Eubin%3A+Maximum+likelihood+from+ineolnpletc+data+via+tile+I%22M+algorithm%2C+Journal+of+Ihe+Royal+S%5C%7Catistical+Society%2C+3+9+%28+B+%29+%3A+1%2D38%2C+1977%2E+1121+A%2EJ%2C+Viter) 5. [ 13] L.E. Bauln: All inequality and associated inaxilnization tcc]miquc in statistical estimatkm of probabilistic functions o1 a Maikov process, lneqoalities, 3:1-8, 1972. [find](https://scholar.google.com/scholar?q=%5B+13%5D+L%2EE%2E+Bauln%3A+All+inequality+and+associated+inaxilnization+tcc%5Dmiquc+in+statistical+estimatkm+of+probabilistic+functions+o1+a+Maikov+process%2C+lneqoalities%2C+3%3A1%2D8%2C+1972%2E) 6. [ 14] F. Jclinek: A fast sequential decoding algorithm using a stack, IBM T. a. Watson Research Development, vol. 13, pp. 6754~85, No\\. 19(¢) Mr. Speaker, I rise on a question of privilege Monsieur l'Orateur, je souleve la qoestion de privilege affecting the rights and prerogatives of pmliamentary c… [find](https://scholar.google.com/scholar?q=%5B+14%5D+F%2E+Jclinek%3A+A+fast+sequential+decoding+algorithm+using+a+stack%2C+IBM+T%2E+a%2E+Watson+Research+Development%2C+vol%2E+13%2C+pp%2E+6754%7E85%2C+No%5C%5C%2E+19%28%C2%A2%29+Mr%2E+Speaker%2C+I+rise+on+a+question+of+privilege+Monsieur+l)