Search Descriptions

Main Topics

Search Publications


author

title

other

year

Word-Based Models

Being the initial models for statistical machine translation, word based models are tied to the translation of individual words.

Word Based Models and its 13 sub-topics are the main subject of 396 publications.

Publications

The initial approach to statistical machine translation led to the development of the IBM Models (Brown et al., 1988; Brown et al., 1990; Brown et al., 1993). A popular implementation of the training of these models is GIZA++ (Och and Ney, 2000) which is still used for word alignment as a initial training step of more complex models.

Benchmarks

Discussion

None of the currently competitive machine translation systems are word based models, but nevertheless the principles such as generative modelling and the use of the expectation maximimization algorithm are still core methods today. Moreover, word alignment based on word based models is more often than not the first step in training more complex models.

New Publications

Actions

Download

Contribute