machine translation

How did machine translation evolve?

The idea of using computers to automatically translate natural language texts first emerged in the early 1950s. However, at the time, the complexity of translation was much higher than scientists had anticipated. At the time, computers lacked the processing power to handle and store data to be able to use machine tчranslation technology.

It was only in the early 2000s that computer software, storage technology and hardware started to meet the requirements of machine translation. At the early stages of development, statistical language databases were used to teach computers how to translate text. This required a lot of manual work and time. For each new language they had to develop a new database. Since then, machine translation has become faster and more accurate, and several different machine translation strategies have emerged.

Rule-based machine translation

For this approach, linguistic experts have developed built-in linguistic rules and bilingual dictionaries for specific sectors or topics. Rule-based machine translation uses these dictionaries to accurately translate specific content.

What are the approaches to machine translation?

In machine translation, the source text or language is called the source language, and the language being translated is called the target language.

The process of machine translation is done in two steps:

decoding the meaning of the source text into the target language;

transferring the meaning to the target language.

We will discuss some common approaches to implement machine translation.

This process consists of the following steps:

The machine translation software analyses the source text and creates an intermediate version of its translation; the result is converted into text in the target language using the grammar rules and dictionaries as reference.

Advantages and disadvantages

Machine translation based on rules can be customised to the subject area and adapted to the needs of the specific industry. It provides predictable results and an acceptable translation quality. However, if the source text contains errors or uses words not found in the built-in dictionaries, the translation quality can suffer. The only way to fix this is to update the dictionaries manually.

Statistical machine translation

Instead of linguistic rules, statistical machine translation relies on machine learning to translate text. Machine learning algorithms analyse a large volume of existing human translations and look for statistical patterns. The software then produces the most likely match when translating new texts. This approach is based on finding the most likely translation of words or phrases using statistical data extracted from a bilingual population of texts.

Machine translation based on syntactic rules

Machine translation based on syntactic rules is a sub-category of statistical machine translation. It uses grammatical rules to translate syntactic units. It analyses sentences to incorporate the syntax rules into statistical translation models.

Advantages and disadvantages

The result of statistical machine translation depends on the number of language pairs, and the accuracy of their correspondence. However, if enough data is available, machine translations are generated with high accuracy.

Neural Machine Translation

Neural network

Comparison of neural machine translation with other methods

Neural machine translation uses artificial intelligence technologies to train languages and continuously improve this knowledge using neural networks, a special machine learning technique. It is often used in conjunction with statistical translation techniques.

A neural network is a set of interconnected nodes resembling the human brain. It is an information system in which input data passes through several interconnected nodes to create output data. Neural machine translation software uses neural networks to handle huge data sets. Each node performs a single attribute change to the source text to convert it to the target text until the output node produces the final result.

Neural networks take full account of the source sentence at each step in creating the output sentence. Other machine translation models break down the input sentence into sets of words and phrases, matching them with a word or sentence of the target language. Neural machine translation systems can address many of the limitations of other methods, and often provide higher quality translations.