Atlas record: https://www.aitimeline.dev/en/events/evt-0168/
Research · June 1993
The IBM translation models
The paper set out five word-alignment models in detail and a way to estimate their parameters from an unannotated bilingual corpus.
Why it matters
Machine translation acquired a reproducible recipe that held until neural translation arrived in 2014.
Models one through five grow steadily more complex: from a simple word correspondence to accounting for order and fertility, that is how many words one word maps to. Parameters are estimated by expectation maximisation or an approximation of it (Och and Ney 2003). Word alignments from these models remained a basic step in the phrase-based systems of the 2000s and in training sets for neural systems.
Event record
- Event date
- June 1993
- Timeline date
- Event date
- Verification
- Sources gathered automatically · September 17, 2026
- Lines
- ID
- evt-0168
Computational Linguistics volume 19, issue 2, June 1993.
Related events
Earlier
Later
- Builds on Maximum entropy for language processing 1996links to this record
All of the paper's examples are parts of IBM's Candide translator, and it opens with a review of the statistical translation on whose models Candide rests.
- Builds on Object recognition as translation 2002links to this record
The lexicon between image regions and words is fitted with "model 2 of Brown et al.", the second of the IBM models for statistical translation, by EM.