Phrase-based translation

Closed weights University of Southern California 9.2M parameters May 2003

No estimate

No hardware requirements for this model

The weights for this model have not been published, so it cannot be downloaded or run on your own hardware at any size. It is reachable only through its provider, and no graphics card changes that.

On record

Full specification

Everything on record for this model. Most of it describes how it was trained rather than how it runs — useful context for judging how much work went into it, and how it compares with models built at a different scale.

Origin

Who built this model, where, and when it was published.

Organisation
University of Southern California
Organisation type
Academia
Country
United States of America
Published
1 May 2003
Authors
Philipp Koehn, Franz Josef Och, Daniel Marcu

What it does

The problem areas the model was built for. A model can carry several of each.

Domain
Language
Task
Translation

Size

How large the model is and how much data it was trained on. Parameters are the figure that decides whether it fits on a given graphics card.

Parameters
9.2M

There are various components to the system: - Translation probability model phi - The distortion probability distribution d - A langage model p_LM - A length factor w Several translation probability models are considered. The most performant one is the AP word alignment model. The sentence length preferred by the authors is 3 words maximum. In the biggest corpus considered (320k phrase pairs) it produces a phrase translation probability table of 1996k entries. The distortion probability model…

Training data
20,000,000 tokens

[WORDS] "We used the freely available Europarl corpus to carry out experiments. This corpus contains over 20 million words in each of the eleven official languages of the European Union, covering the proceedings of the European Parliament 1996-2001. 1755 sentences of length 5-15 were reserved for testing." "These results are consistent over training corpus sizes from 10,000 sentence pairs to 320,000 sentence pairs. " So 20 million words or 320k sentence pairs.

How it is classified

Labels the source dataset applies when tracking notable models, and how confident it is in the entry.

Citations
4,270

Sources

Where this record came from and when it was last checked.

Reference
Statistical Phrase-Based Translation
Last updated
28 November 2025

What the numbers mean

Background

Phrase-based translation was published by University of Southern California, in United States of America, in May 2003. academia is the category the publisher falls under.

It works in Language, and is recorded as doing translation.

This is a closed model: the trained values stayed with whoever produced them, and there is no local version to run.

How it was trained

Around 20,000,000 tokens went into training it.

Answers

Phrase-based translation — common questions

01

When was Phrase-based translation released?

Phrase-based translation was published in May 2003. Capability per parameter has improved considerably since, so a newer model of the same size is often the better use of the same hardware.

02

What is Phrase-based translation used for?

Phrase-based translation works in Language, and is recorded as handling translation. Models frequently carry more than one of each, and the tags describe purpose rather than capability limits.

03

What GPU do I need to run Phrase-based translation?

None. Phrase-based translation is a closed model — its weights were never published, so it cannot be downloaded or run on your own hardware at any price. It is reachable only through its provider.

04

Is Phrase-based translation open source?

The licensing for Phrase-based translation was never recorded in our source data. We treat unstated licensing as closed, because an unrecorded licence is not one to rely on.

05

How many parameters does Phrase-based translation have?

Phrase-based translation has 9.2M parameters. There are various components to the system: - Translation probability model phi - The distortion probability distribution d - A langage model p_LM - A length factor w Several translation probability models are considered. The most performant one is the AP word alignment model. The sentence length preferred by the authors is 3 words maximum. In the biggest corpus considered (320k phrase pairs) it produces a phrase translation probability table of 1996k entries. The distortion probability model d is taken from (Marcu and Wong, 2002). The distortion probability model must have ~10 parameters at most The language model p_LM is a back off trigram model from (Seymore and Rosenfeld,1997). AFAIK the cutoff used is not specified. Based on the example on section 4.3 of (Seymore and Rosefeld, 1997), a trigram probability model has about 3866964 + 2674322 + 641604 parameters. "For each possible phrase translation anywhere in the sentence (we call it a translation option), we multiply its phrase translation probability with the language model probability for the generated English phrase. As language model probability we use the unigram probability for the first word, the bigram probability for the second, and the trigram probability for all following words" The length factor w is an additional single parameter. "In order to calibrate the output length, we introduce a factor w for each generated English word in addition to the trigram language model " In summary, the parameter count seems to be dominated by the trigram language model and the word alignment phrase translation model. That figure is the total, and it is what decides how much memory the model needs — roughly half a gigabyte per billion at the compression most people use.

06

Who created Phrase-based translation?

Phrase-based translation was published by University of Southern California, based in United States of America, categorised as academia.

Source

Original publication

Record last updated 28 November 2025

The other direction

Looking at it from the other side?

This page starts from the model. If you already own a card and want to know everything it will run, start from the hardware instead.