How Amazon Translate Works
The Amazon Translate service is based on neural networks trained for language translation. This enables you to translate between a source language (the original language of the text being translated) and a target language (the language into which the text is being translated). It is not necessary to use English as either the source or the target language but not all language combinations are supported by Amazon Translate. For more information, see Supported Languages.
When working with Amazon Translate, you will provide source text and get output text:
-
Source text—The text that you want to translate. You provide the source text in UTF-8 format.
-
Output text—The text that Amazon Translate has translated into the target language. Output text is also in UTF-8 format. Depending on the source and target languages, there might be more characters in the output text than in the input text.
The translation model has two components, the encoder and the decoder. The encoder reads a source sentence one word at a time and constructs a semantic representation that captures its meaning. The decoder uses the semantic representation to generate a translation one word at a time in the target language.
Amazon Translate uses attention mechanisms to understand context. This helps it decide which words in the source text are most relevant for generating the next target word. Attention mechanisms enable the decoder to focus on the most relevant parts of a source sentence. This ensures that the decoder correctly translates ambiguous words or phrases.
The target word that the model generates becomes input to the decoder. The network continues generating words until it reaches the end of the sentence.
Supported Language Codes
To translate text, you call the TranslateText method and provide the source text and the target language, using one of the language codes listed in the following table:
| Language | Code |
|---|---|
| Arabic | ar |
| Chinese (Simplified) | zh |
| Chinese (Traditional) | zh-TW |
| Czech | cs |
| Danish | da |
| Dutch | nl |
| English | en |
| Finnish | fi |
| French | fr |
| German | de |
| Hebrew | he |
| Hindi | hi |
| Indonesian | id |
| Italian | it |
| Japanese | ja |
| Korean | ko |
| Malay | ms |
| Norwegian | no |
| Persian | fa |
| Polish | pl |
| Portuguese | pt |
| Russian | ru |
| Spanish | es |
| Swedish | sv |
| Turkish | tr |
Automatic Language Detection
Amazon Translate can automatically detect the language used in your source text. To
use automatic
language detection, specify auto as
the
source language. Amazon Translate calls Amazon Comprehend on your behalf to determine the
language used in the source text. By choosing automatic language detection, you agree
to
the service terms and agreements for Amazon Comprehend. For information about pricing
for Amazon Comprehend, see
Amazon Comprehend
Pricing.
Exception Handling
If you specify a source or target language that isn't supported, Amazon Translate returns the following exceptions:
-
UnsupportedLanguagePairException – Amazon Translate supports translation between the supported languages. This exception is returned if translation between the language pair is unsupported. For more information, see Supported Languages.
-
DetectedLanguageLowConfidenceException – If you use automatic language detection, and Amazon Translate has low confidence that it detected the correct source language, it returns this exception. If a low confidence level is acceptable, you can use the source language returned in the exception.
Next Steps
Now that you've learned how Amazon Translate works, you can explore the following sections to learn about creating a solution.
