Neural Machine Translation

Neural Machine Translation:
Revolution Before Our Eyes

A few decades ago, the word "translator" was associated exclusively with a human being. But with the advent of the computer age, machine translation began to develop: firstly in the interests of intelligence agencies, and then science, business and everyone with an laptop or a smartphone. Recently, the translation industry took an enormous leap forward with the emergence of a new machine translation technology called Neural Machine Translation.

The Evolution of Machine Translation

Since 1950

Rule-Based MT

Machine accent
in translation,
predictable
result

TRAINING WITH
DICTIONARIES

Since 2000

Statistical MT

Smoother translation
than in RBMT
but unpredictable
result

TRAINING WITH
PARALLEL DATA

Since 2015

Neural MT

More fluent
and readable
than ever
before

TRAINING WITH
PARALLEL DATA

The Evolution of Machine Translation

Rule-Based MT

Machine accent in translation, predictable result

TRAINING WITH DICTIONARIES

Since 1950

Statistical MT

Smoother translation than in RBMT but unpredictable result

TRAINING WITH PARALLEL DATA

Since 2000

Neural MT

More fluent and readable than ever before

TRAINING WITH PARALLEL DATA

Since 2015

Neural Machine Translation Drivers

Accelerated
Computing
Power

Digital
Content
Growth

Mature and
Cutting-edge
Algorithms

AI in business: breaking down language barriers

LOCALIZATION

product descriptions, online stores, user documentation, forums, websites

INTERNATIONAL COOPERATION

secure and confidential translation of internal and external documentation in real time

INTEGRATION

machine translation in corporate portals, electronic document management systems, BI systems, mobile applications

Why is natural language processing so tricky?

CONTENT DIVERSITY
Short and long texts, speech recognition and automated transcription, terminology from different branches of science and business, formal, informal style and slang are "mini-languages" in one big language.

LANGUAGE ASYMMETRY
Different languages describe the world differently and use different approaches in morphology, grammar, syntax, and semantics.

AVERAGE WORD LENGTH IN
DIFFERENT LANGUAGES
Increasing or decreasing the number of characters in the translation text compared to the original affects how the document is formatted during translation.

TRAINING DATA
Difficulty finding and selecting data for training and setting up systems for translation to specific business tasks.

PROMT Neural is a new word in AI

PROMT Neural is a brand-new hybrid technology that combines a neural network approach and rule-based machine translation (RBMT). The PROMT Neural algorithms pre-analyze the text and decide which technology is best suited for translating a particular piece of text.