From your files to a glossary ready for review.
Term Extractor speeds up the most time-consuming part of terminology work without replacing the final judgment of a professional linguist: the result is a solid base to review, not a final glossary ready for use.
Upload translation memories, PDFs and Word documents. Term Extractor analyzes every sentence, finds terminology candidates and returns a ranked list in JSON and Excel.
How it works.
Three steps.
No spreadsheet to build by hand.
Upload your project material, even in mixed formats, and let the NLP analysis
take care of the initial extraction.
Upload TMX, PDF and DOCX* files, even at the same time in the same session. The aligned translation of the examples is also extracted from TMX files, when present.
*Image-only PDF pages and Word files in the legacy .doc format are skipped and reported in the log.
The NLP engine (spaCy) parses every sentence, isolates noun phrases, discards noise (codes, measurements, numbers) and counts how many segments each extracted term appears in.
Download a glossary in JSON and Excel format, ranked by frequency, with up to three examples per term.
You decide what to extract.
From 1 to 5 words per term.
Minimum occurrences required, settable for each length.
A limit per length, or one overall limit.
Removes product codes, phone numbers, measurements and other superfluous segments.
Skips segments too short to be useful sentences.
Exclude specific frequent words that are not terms.
What happens to your files, step by step.
The tool identifies the supported files and checks the source language of TMX files. In addition, if a TMX contains several target languages, it lets you choose which one to use for extracting the examples.
Sentences are extracted from TMX, PDF and DOCX files and cleaned of superfluous segments according to the parameters you set.
The NLP model identifies noun phrases and extracts in-context examples for each term.
The tool counts how many segments each term appears in and how many times it occurs, then filters the results according to the minimum thresholds and the limit on the number of terms you set.
The output is the glossary in JSON and Excel format, plus a log file. The columns for translation, definition and notes are left empty, ready to be filled in during the human review phase.
Every row is a terminology candidate. The frequency figures and the examples
help the linguist decide whether to keep it. The data below is an illustrative example.
| source_term | target_term | term_type | ngram | seg_count | raw_freq | sources | definition | example_1 | target_example_1 | notes |
|---|---|---|---|---|---|---|---|---|---|---|
| instructions for use | — | Noun | 3 | 31 | 40 | docx, tmx | — | Read the instructions for use before inserting the device. | Leggere le istruzioni per l'uso prima di inserire il dispositivo. | — |
| valve | — | Noun | 1 | 18 | 27 | pdf, tmx | — | Close the valve after each use. | Chiudere la valvola dopo ogni utilizzo. | — |
| pressure sensor | — | Noun | 2 | 9 | 13 | tmx | — | The pressure sensor is mounted on the valve stem. | Il sensore di pressione è montato sullo stelo della valvola. | — |
| syringe barrel | — | Noun | 2 | 14 | 22 | docx | — | Hold the syringe barrel with your free hand. | — | — |
seg_count indicates the unique occurrences within segments, while raw_freq indicates the total number of occurrences. Target examples (target_example_1…3) appear when the term comes from a TMX file that contains them.
Catalan
Chinese
Croatian
Danish
Dutch
English
Romanian
Russian
Finnish
French
German
Greek
Italian
Japanese
Slovenian
Spanish
Korean
Lithuanian
Macedonian
Norwegian Bokmål
Polish
Portuguese
Swedish
Ukrainian
It comes from day-to-day work on project glossaries: the filters and extraction rules reflect the way a professional linguist decides what to keep and what to drop.
For each term, the tool counts occurrences directly in the uploaded files and extracts the in-context examples from them.
The human translations already in your TMX files are not lost: the target examples appear in the glossary next to the source examples, so the linguist can see right away how the term was translated in previous projects.
No account, no cost. Ideal for an initial extraction to be followed by a human review step.
FAQ.
It is a terminology extraction tool developed by Creative Words for translators, professional linguists, project managers and language teams who need to build a glossary from a project’s materials.
TMX, PDF and DOCX files are supported as input, even together in the same extraction session. PDFs must contain text: pages made only of images are skipped. Word files in the legacy .doc format are not read. There are 24 supported source languages.
No. The tool is free and does not require creating an account.
No, it extracts source-language terms. The target_term, definition and notes columns are left empty to be filled in during the human review of the glossary. If the input files are TMX, the corresponding translation will also appear next to the source example.
First by the number of distinct segments the term appears in, then by total frequency. In this way, the terms most widespread across the files you uploaded appear at the top, not those repeated many times in a single sentence.
No: it is a list of candidates to review. Automatic extraction can include entries that are not real terms or forms that need normalizing. Final cleanup and validation must be carried out by a professional linguist.
