Translator counts every code point defined in Unicode as a character. Each translation counts as a separate translation, even if the request was made in a single API call translating to multiple languages. The length of the response doesn't matter and the number of requests, words, bytes, or sentences isn't relevant to character count.
Translator counts the following input:
Text passed to Translator in the body of a request.
Text when using the Translate, Transliterate, and Dictionary Lookup methods Text and Translation when using the Dictionary Examples method. All markup: HTML, XML tags, etc. within the request body text field. JSON notation used to build the request (for instance the key "Text:") is not counted.
An individual letter.
Punctuation.
A space, tab, markup, or any white-space character.
A repeated translation, even if you have previously translated the same text. Every character submitted to the translate function is counted even when the content is unchanged or the source and target language are the same.
For scripts based on graphic symbols, such as written Chinese and Japanese Kanji, the Translator service counts the number of Unicode code points. One character per symbol. Exception: Unicode surrogate pairs count as two characters.
Calls to the Detect and BreakSentence methods aren't counted in the character consumption. However, we do expect calls to the Detect and BreakSentence methods to be reasonably proportionate to the use of other counted functions. If the number of Detect or BreakSentence calls exceeds the number of other counted methods by 100 times, Microsoft reserves the right to restrict your use of the Detect and BreakSentence methods.
For detailed information regarding Azure AI Translator Service request limits, see Text translation request limits.
Note: Seeding question
