Automatic or machine translation is perhaps one of the most challenging artificial intelligence tasks given the fluidity of human language. Classically, rule-based systems were in a popular use for this task, but statistical methods replaced them in the 1990s. More recently, deep neural network models achieve state-of-the-art results in a field that aptly uses the name of neural machine translation.
What is Machine Translation?
Machine translation is automatically converting source text in one language to the text in another language.
The fact is that accurate translation requires background knowledge to resolve ambiguity and establish the content of the sentence.
Classical machine translation methods often involve rules for converting text in the source language to the target language. Linguists develop the rules and they may operate at the lexical, syntactic, or semantic level.
The key limitations of the classical machine translation approaches are both the expertise required for developing the rules, and the vast number of rules and exceptions required.
What is Statistical Machine Translation?
Statistical machine translation, or SMT for short, is the use of statistical models that learn to translate text from a source language to a target language gives a large corpus of examples.
This approach does not need a complex ontology of interlingua concepts, nor does it need handcrafted grammars of the source and target languages, nor a hand-labeled treebank. All it needs is data—sample translations from which an expert can learn a translation model.
Quickly, the statistical approach to machine translation outperformed the classical rule-based methods to become the de facto standard set of techniques.
The most popular models for statistical machine translation have been sequence-based. In these models, the basic units of translation are words or sequences of words. These kinds of models are simple and effective, and they work well for man language pairs.
The most widely used techniques werephrase-based and focus on translating sub-sequences of the source text piecewise.
Statistical Machine Translation (SMT) has been the dominant translation paradigm for decades. Practical implementations of SMT are phrase-based systems (PBMT) which translate sequences of words or phrases where the lengths may differ.
Although effective, statistical machine translation methods suffered from a narrow focus on the phrases being translated, losing the broader nature of the target text.
The hard focus on data-driven approaches also meant that methods may have ignored important syntax distinctions known by linguists. Finally, the statistical approaches required careful tuning of each module in the translation pipeline.
What is Neural Machine Translation?
Individuals have a plethora of platforms that allow them to access consumers all over the globe and work with other companies in faraway places – if only they could speak the same language.
In an ironic twist, language has turned from something that first facilitated human cooperation and growth, to something that impedes our ability to work together.
Technology may finally be ready to abolish that barrier forever. Remarkably, in 2018, over 20 years after widespread use of the Internet began, we still rely almost only on humans to translate language in commercial formats.
But translation bears all the earmarks of those functions that artificial intelligence ought to replicate, and a technology called Neural Machine Translation (NMT) does just that.
The key benefit to the approach is that a single system can apply directly on the source and target text, no longer requiring the pipeline of specialized systems used in statistical machine learning.
Unlike the traditional phrase-based translation systems which comprise many small sub-components that operate separately, neural machine translation efforts to build and train a single, large neural network that reads a sentence and outputs a correct translation.
Contextual translation ability
By leveraging its contextual translation ability alongside its deep learning functions, NMT has achieved historic results in the journey to a post-language economy.
In a side-by-side comparison with human translators, in a technical domain translation for English-Korean, translators preferred SYSTRAN’s NMT translations 41 percent of the time.
That success can come from advancing language translation beyond rule-based translation methods.
NMT is a deep learning technology that translates within the context, not just one word at a time.
Encoder-Decoder Model
Multilayer Perception neural network models are good for machine translation although there are several factors that limit models such as a fixed-length input sequence where the output must be the same length.
These early models have improved upon recently through the use of recurrent neural networks organized into an encoder-decoder architecture that allows for the variable length input and output sequences.
An encoder neural network reads and encodes a source sentence into a fixed-length vector. A decoder then outputs a translation from the encoded vector.
The whole encoder-decoder system, which comprises the encoder and the decoder for a language pair, is in a great usage to maximize the probability of a correct translation given a source sentence.
Key to the encoder-decoder architecture is the ability of the model to encode the source text into an internal fixed-length representation called the context vector.
Interestingly, once encoded, different decoding systems are in use, in principle, to translate the context into different languages.
Encoder-Decoders with Attention
Although effective, the Encoder-Decoder architecture has problems with long sequences of text which demand translation.
The problem stems from the fixed-length internal representation that demand decoding each word in the output sequence.
The solution is to use of an attention mechanism. It allows the model to learn where to place attention on the input sequence as each word of the output sequence is decoded.
Using a fixed-sized representation to capture all the semantic details of a very long sentence is very difficult. A more efficient approach, however, is to read the whole sentence or paragraph. The next step is to produce the translated words one at a time, each time focusing on a different part of the input sentence to gather the semantic details required to produce the next output word.
You would also be interested Obstacles on the Way to the Perfect Translation
References:
