What the Glossary Builder does
This glossary builder finds the candidate terms in a document - repeated two- to four-word phrases, capitalised names and acronyms - groups the different ways each one is written, and gives you a table to add definitions, target-language translations and forbidden variants. It exports the result as CSV, as TBX-style XML for translation tools, or as a formatted glossary page you can print.
It does not write definitions or translations for you: candidates come from counting the document, and every definition is yours. The document and glossary stay in your browser.
How to use it
- Paste the document or drop a .txt or .md file, then press Find candidate terms.
- Read the notes first. Terms written more than one way, such as e-mail and Email, are listed so you can pick a preferred form.
- Untick candidates that are not really terms. Single ordinary words start unticked, because they are the least likely to need a definition.
- Write a definition for each kept term. If you are preparing a document for translation, set the target language and add the approved translation, and list the spellings nobody should use as forbidden variants.
- Download CSV to keep working later (drop it back on the page to continue), TBX to load into a translation tool, or HTML for a printable glossary.
Reading the results
Uses counts every occurrence of the term, including its plural and hyphen or space variants. The forms under the term show how often each spelling appears.
Type says why the term was picked: Acronym (two or more capitals), Name (capitalised in the middle of a sentence), Phrase (a repeated run of two to four words) or Word (a single word used three or more times).
Candidates are ranked by uses x words in the term, with a small boost for names and acronyms, so a three-word name used three times ranks above a two-word phrase used four times.
A shorter phrase that only ever appears inside a longer one - "Desk Portal" inside "Help Desk Portal" - is dropped, and so is a single word used mostly inside longer terms.
Worked example: a support policy
Load the support policy, a 114-word document. The builder finds 9 candidates. Help Desk Portal appears 3 times, capitalised mid-sentence, so it is a Name with a score of 3 x 3 + 3 = 12 and ranks first. Support ticket appears 4 times - 3 as "support ticket" and 1 as "support tickets" - so it scores 4 x 2 = 8.
SLA (2 uses) is an Acronym, and Service Level Agreement (2 uses) is flagged for inconsistent capitals because the second use is lower case. On-call engineer, escalation path, priority level and maintenance window each appear twice.
The last candidate is the word Email, written once as "e-mail" and once as "Email". It is flagged as written more than one way. Keep it, choose email as the term and add e-mail as a forbidden variant: the page then highlights the one remaining "e-mail" in the document.
About the TBX export
TBX (TermBase eXchange, ISO 30042) is the XML format translation and terminology tools use to exchange termbases. The export follows the TBX-Basic dialect layout: one conceptEntry per term, a langSec for the source language holding the definition, the preferred term marked preferredTerm-admn-sts and each forbidden variant marked deprecatedTerm-admn-sts, and a second langSec for the target term.
It is written to that layout but not validated against the full schema, so test the import in your tool with a few terms first. CSV is the safer choice for spreadsheet-based workflows.
Limitations: what the result does not prove
- Extraction counts words; it does not understand them. A repeated phrase is not necessarily a term, and a term used only once is not found - add it by hand.
- Plurals are grouped with an English stemmer, so grouping works best in English. Other languages still get phrase counts, but plural and variant grouping will be weaker.
- Phrases never cross commas, brackets or sentence ends, so a term split by punctuation in the text is counted separately.
- The forbidden-variant check matches whole words and ignores capitals and accents, so it cannot tell when a forbidden spelling is correct in a quotation or a product name.
Privacy: where your data goes
Everything you paste, type or drop is processed in this browser tab. It is not uploaded, logged, stored or sent to analytics. Session recording and tag-manager scripts are switched off on this page.
Standards and sources
- ISO 30042:2019 - TermBase eXchange (TBX)
- TBX-Basic dialect (TBXInfo)
- Porter (1980) - An algorithm for suffix stripping
Frequently asked questions
How do I make a glossary from a document?
Paste the document, press Find candidate terms, untick anything that is not a term, then write a definition for each one. Download the result as CSV, TBX or a formatted HTML glossary. You can drop the CSV back on the page later to carry on.
What is a termbase and why would a translator want one?
A termbase is a glossary with approved translations and forbidden variants, used by translators and translation tools to keep terminology consistent. Sending one with a document means the translator does not have to guess how you want your product names and key terms translated.
Can I import the TBX file into a CAT tool?
The export uses the TBX-Basic layout that many translation tools accept, with definitions, preferred and deprecated terms and a target-language section. It is not validated against the full schema, so try a small import first; if your tool prefers CSV, use that.
What does a forbidden variant do?
It records a spelling or wording that should not be used for the term - for example e-mail when your style is email. The page highlights every forbidden variant still in the document, and the TBX export marks them as deprecated terms.
Does the glossary builder write definitions automatically?
No. It finds candidate terms by counting and shows the sentence where each first appears, but every definition and translation comes from you. That avoids confident-sounding definitions that are wrong for your context.
Why does the list skip common single words?
Single words appear in any document and are rarely glossary terms, so ordinary ones are left out, and those used mostly inside a longer term are dropped. You can include single words in the options, and add any term by hand.
Last reviewed by the A2Z.Tools team against the sources listed above.
