Learning Terms is wxrks's AI-assisted glossary feature: instead of a linguist manually spotting and typing in every new term, wxrks watches your source content and your translations, and adds terminology to your glossary on its own as work happens.
💡 Who is this for? This guide is for Account Admins and Project Managers who need to turn on AI-assisted term extraction so wxrks automatically learns and adds new terminology to a glossary during translation.
What Learning Terms actually does
Learning Terms works in two stages, and it's worth knowing both because they're configured in two completely different places.
1. Candidate extraction, when a file is parsed. As soon as a file is uploaded to a project, wxrks can scan its source content and pull out a list of candidate terms — words and short phrases that repeat often enough, or stand out enough, to plausibly be terminology. This step is statistical (based on frequency and word count), not AI-based, and just produces a shortlist for the next stage.
2. AI-based learning, during translation. As segments get translated and confirmed, wxrks's AI reviews each one in context — using the candidate list from step 1 as a hint — and decides which source/target pairs are genuinely worth keeping as terminology. Anything it keeps is added to your glossary automatically, tagged as a learned term.
This is different from the other two ways terms get into a glossary: adding a concept manually, and requesting one through the Terminology Board's review workflow. Learning Terms is the only one of the three that needs no human action at all once it's turned on — see All about Glossaries for how glossaries and concepts work in general.
Before you start
A glossary must already be linked to the Organizational Unit you want to use Learning Terms in.
wxrks's AI features (what the app calls Augmented Actions) must be enabled for the Account and for that Organizational Unit — Learning Terms runs on the same underlying AI, so without it enabled at both levels, turning the glossary's Learning Terms switch on has no effect.
1. Enable Learning Terms for a glossary
Learning Terms is turned on per glossary, at the Organizational Unit level. Open the Organizational Unit, go to the Context tab, and find the glossary in the Glossary table.
Switch on Enable Learning Terms for that glossary's row. From this point on, any project whose translations use this glossary is eligible to have terms learned into it automatically. Turning it off at any time stops new terms from being learned — it does not remove terms that were already learned.
2. Configure candidate term extraction for a file
The candidate-extraction step (stage 1 above) is configured per file, when it's uploaded. When you create a project and get to the Upload Files step, click a file to expand its parsing options, then look at the Term Extraction panel on the right.
The panel's switch and fields:
Extract Terms from Source Content: turns candidate extraction on or off for this file. When it's off, this file contributes no candidate terms — the AI-based learning in step 2 above can still run on its segments, just without the extra hint.
Keep Case: treats differently-capitalized versions of the same word as separate candidates (useful for acronyms or proper nouns that shouldn't be folded into a lowercase match).
Max Words Per Term: the longest a candidate phrase is allowed to be. Keep this low (6 words or fewer) — a higher ceiling tends to pull in whole sentence fragments instead of real terms.
Min Occurrences: how many times a word or phrase has to repeat in the file before it's considered a candidate at all.
Min Words Per Terms: the shortest a candidate phrase is allowed to be.
Remove Sub Terms: when a longer candidate contains a shorter one (e.g. "translation memory" contains "memory"), this discards the shorter one so it isn't double-counted.
Sort by Occurrence: ranks candidates by how often they appear, most frequent first.
Top Terms Limit: caps how many candidates this file can contribute in total.
These defaults work well for most files — there's rarely a reason to change them unless a specific file is producing candidates that are clearly too short, too long, or too noisy.
3. Recognize and review learned terms
A term that Learning Terms added on its own is marked with a small graduation-cap icon, wherever it shows up — in the glossary's own term table, and on the Terminology Board if it was routed there. This is the one visual cue that tells you a term came from the AI rather than from a person.
[SCREENSHOT PLACEHOLDER]
By default, a learned term is written straight into the glossary. An Account Admin can instead route learned terms through the Terminology Board for review before they're published, using the Enable Sending Learned Terms to Terminology Board setting under the Organizational Unit's LLM tab (this also needs to be enabled at the Account level to take effect). This is worth turning on if you'd rather have someone approve AI-suggested terminology before it becomes official — see Terminology for how the Board's review workflow works.
Good to know
Learning Terms never overwrites a term you (or anyone) added manually. If a matching term already exists as a manual entry, it's left alone; only terms that were themselves learned get updated on a later pass.
Turning off Enable Learning Terms for a glossary stops new terms from being learned going forward — it doesn't remove terms already added.
If nothing is being learned even with the toggle on, check the prerequisites above first: AI features are enabled for both the Account and the Organizational Unit, and the project's translations are actually using the glossary you enabled it on.
Related articles