Skip to main content

How Taia counts words

Every quote for professional translation and every AI translation credit starts from one number: the word count of your source text. Taia counts words by GMX-V, a published standard for translation word counts, which builds on the Unicode rules for where words begin and end. Because the rules are public, any count Taia gives you can be checked against them, together with the few conversions described below.

Taia has counted words this way since 28 September 2026. A project keeps the count it was quoted on, unless files are later added to it or removed from it, which counts it again with the current rules.

What counts as a word​

GMX-V finds words with the Unicode rules for word boundaries, with a few additions for translation.

TextWordsWhy
The quick brown fox jumps over the lazy dog.9Punctuation is never a word
Store at 15-25 °C for 2 min7Numbers are words; ° is a symbol, C is a word
1,000.50 EUR, 1'000 CHF4A number keeps its separators
UDI-DI, state-of-the-art2A hyphen joins, it does not split
can't, don't2An apostrophe inside an English word keeps it whole
U.S.A.1A full stop between letters keeps the word whole
l'objectif (French)2The elided article is its own word
dell'apparecchio (Italian)2Same rule for Italian
aujourd'hui (French)1A compound, not an elision
Visit https://taia.io or write to info@taia.io4Web addresses and email addresses are not counted

Formatting is never counted. Bold, italics, links and other formatting codes in your document are invisible to the count, and so is whitespace.

Web and email addresses are not counted either. Nobody translates https://taia.io/pricing/ or info@taia.io, so they count as 0 words wherever they appear, including links written out in plain text or Markdown. The text of a link, such as [Pricing](https://taia.io/pricing/), is still counted. This rule is Taia's own, not part of GMX-V, and applies from 8 October 2026.

Chinese, Japanese, Korean and Thai​

GMX-V counts these scripts by characters rather than words: every character except spaces, punctuation and symbols, including any Latin letters and digits in the text. Taia turns that character count into words so that quotes and credits compare across languages:

Source languageWords per character
Chinese0.7
Japanese0.5
Korean0.27
Thai, Khmer, Lao, Myanmar0.25

For example, 1,000 Chinese characters in one segment count as 700 words. Each segment is rounded to the nearest word, and any segment with text in it counts as at least one word, so a file of very short strings (buttons, labels) counts a little higher. A segment of punctuation alone counts as nothing.

What the count is used for​

  • Professional translation quotes are priced per word on this count. Repeated segments and translation memory matches lower the price of the translation step; revision and proofreading are priced on the full count. In a very small order, each language is priced at a minimum of 250 words.
  • AI translation credits use this count too: one word of Standard AI translation into one language is one credit. See How credits are calculated.

The count comes from the source text only, so a 1,000-word document counts as 1,000 words whichever languages you translate it into (and each target language is charged separately).

Why another tool may show a different count​

  • Word processors such as Microsoft Word count everything between two spaces as one word. On most text that gives the same result as GMX-V, but Word may count a stand-alone dash or symbol as a word and counts l'objectif as one, so small differences in either direction are normal.
  • Chinese, Japanese, Korean and Thai are not comparable with a word processor's count at all: Taia converts characters to words as described above, while word processors count these scripts in their own way.
  • Other translation tools count by their own rules, or apply GMX-V with different settings, so small differences are normal.
  • Only the text you choose to translate is counted. If you translate highlighted text only or coloured text only, the rest of the document is not counted.