PROMT RBMT

PROMT SMT

PROMT DeepHybrid

Key Components

PROMT Dictionaries

PROMT Translation algorithms

Translation Model

Language Model

PROMT Baseline Rule-based Engine

Language Model

Statistical post-editing

Additional dictionaries

Benefits

Full control over terminology and translation style

More accurate syntax and morphology

Predictable and deterministic

Profiling (multiple profiles can be easily created in one engine)

Fast and fully automated engine training (in most cases, language independent)

More fluent and “human-like” MT output

More customizable and predictable than pure SMT

More fluent and “human-like” MT output than pure RBMT

Engine training is faster than pure RBMT

Profiling (multiple profiles can be easily created in one engine)

Limitations

Language-dependent (algorithms depend on source/target languages)

High customization effort

Requires large and clean parallel corpora for training

Domain-specific (usually trained on/for specific texts)

Hard to customize the translation of a particular word/construction

Requires a parallel corpora for training (but less than pure SMT)

Domain-specific (usually trained on/for specific texts)

Available Languages

English, Russian, German, French, Spanish, Italian, Portuguese, Chinese (Simplified and Traditional), Ukrainian, Kazakh, Turkish, Bulgarian, Latvian, and Polish

Any language pair, for which there are enough training data

English, Russian, German, French, Spanish, Italian, Portuguese

Available Products

Desktop and Server solution

Server-based solutions only

Server-based solutions only

Prerequisites for Engine Training

Source texts (at least, 100,000 words / 10,000 translation units)

Parallel corpora (at least, 5,000,000 words / 500,000 translation units)

Baseline rule-based profile (if there is no custom rule-based profile, PROMT uses an out-of-the-box profile)

Parallel corpora (at least, 500,000 words / 50,000 translation units)