Supervertaler for memoQ: Docs for the memoQ plugin only # Supervertaler for memoQ Supervertaler for memoQ brings AI translation and your own terminology into **memoQ 12**, as two add-ins that work together. Unlike the Trados plugin, which docks its own panels into the editor, memoQ gives an add-in no window of its own. Supervertaler therefore works through memoQ’s existing surfaces – the machine-translation engine and the terminology pane – rather than adding new ones. In practice that turns out to suit it: the results appear exactly where you already look for them. ### What it does **AI translation.** An LLM machine-translation engine using Anthropic, OpenAI or Google, with your own API key and your own instructions. It works segment by segment as you translate, and in bulk through **Pre-translate**. Inline tags survive the round trip. **It learns as you work.** Every segment you confirm is remembered, and the most relevant ones are shown to the model when it translates later segments in the same document. Settle on a term once and the rest of the document follows it – no configuration, no retraining, just your own approved choices fed forward. See [Self-learning translation](/memoq/self-learning/). **Your terminology, twice over.** A glossary you point Supervertaler at appears as a memoQ terminology provider – matched terms highlighted in the source, entries listed in Translation results – *and* is sent to the model as required or forbidden terminology. Forbidden terms are enforced, not merely displayed. See [Terminology](/memoq/terminology/). **Translate with Claude Desktop.** Through the [Supervertaler MCP Server](/memoq/mcp-server/), Claude reads the document you are translating, your confirmed segments and your glossary, and stages translations that flow into the grid when you press Pre-translate. Tokens are billed to your Claude subscription rather than an API key, and every write into your document goes through your own hands. See [MCP Server](/memoq/mcp-server/). **A prompt library, shared with Trados.** Translation instructions come from the same library the Trados plugin uses, chosen from a dropdown and edited in a small companion [editor](/memoq/prompt-editor/). Claude can draft prompts into it too. ### What it does not do memoQ does not let a plugin read its own term bases or translation memories, so terms defined in a memoQ term base are not visible to the AI. Supervertaler reads its own glossary file instead, which it can also display alongside memoQ’s own term base hits. There is no chat panel, no document-wide search and no cursor control inside memoQ: a plugin can answer when asked for a translation, and that is all. The chat lives in Claude Desktop; the prompt library lives in its own editor; and Claude’s translations reach the grid only when you Pre-translate. [MCP Server → What it can and cannot do](/memoq/mcp-server/#what-it-can-and-cannot-do) has the full comparison with the Trados plugin. ### Where to start * [Installation](/memoq/installation/) – putting the add-ins in place * [Getting started](/memoq/getting-started/) – a first translation * [Terminology](/memoq/terminology/) – using a glossary * [MCP Server](/memoq/mcp-server/) – translating with Claude Desktop * [Prompt Library & Editor](/memoq/prompt-editor/) – choosing and writing instructions # Context layers > The layers of context Supervertaler for memoQ puts in front of the AI, what each one adds, where memoQ lets it come from, and how to control them. A translation engine that sees only the sentence in front of it will translate that sentence well and the document badly. Supervertaler’s design is the opposite: every time memoQ asks it for a translation, it assembles a fresh snapshot of your work and hands the whole thing to the AI. That snapshot is built in **layers**. Some come from memoQ, some from your own recorded knowledge, and two of them come from parts of the document that no CAT tool normally shows you at all. They stack, and each one that is present removes a class of mistake the AI would otherwise make. This page is the single place that lists every layer. Supervertaler for Trados has [its own version of this page](/trados/context-layers/); the ideas are the same and the sources differ, because memoQ hands a plugin a different set of things. ## The layers | # | Layer | Where it comes from | Needs work from you | | -- | ------------------------------------ | --------------------------------------- | ------------------------ | | 1 | Project and job information | memoQ, with each request | no | | 2 | The segment being translated | memoQ | no | | 3 | Segments you have confirmed | your own confirmed rows | one setting, once | | 4 | The closest translation memory match | your TMs, routed by you | one setting, once | | 5 | Whether the row was rejected | memoQ | no | | 6 | Glossary terms and forbidden terms | your glossary | a glossary | | 7 | SuperMemory memory banks | what you have recorded about the client | a bank per client | | 8 | List numbering | the original Word file | the live document link | | 9 | Figure descriptions | the images in your documents | two clicks in FigureLens | | 10 | The whole document | the live document link | for AutoPrompt only | ### 1. Project and job information The client, domain and subject recorded in memoQ’s project, the language pair, and which document the segment is in. memoQ sends these with every translation request, so they need no setting – but they are only as good as what the project manager filled in. An empty *Client* field in memoQ is an empty line in the prompt. **Toggle:** Translation settings → *Send surrounding segments and project metadata to the model*. ### 2. The segment being translated The source text, with its inline tags preserved so they can be put back in the right places. Always included; there is nothing to configure. ### 3. Segments you have confirmed Every segment you confirm is recorded, and the most lexically similar ones are shown to the model when it translates later segments of the same document. Confirm *electric module* once and the rest of the document follows. This is memoQ’s answer to a layer Trados gets for free. A Trados plugin is handed the segments surrounding the one it is translating; a memoQ MT plugin is not – memoQ reserves that channel for its own AGT engine – so Supervertaler builds the equivalent out of your confirmed work instead. It is arguably the better trade: every example is one you approved, rather than merely one that happens to be nearby. memoQ only sends confirmations to an engine selected under **Self-learning MT**, so that box must be ticked as well as the engine being chosen for translation. Until it is, nothing is captured and the [Activity window](/memoq/prompt-editor/#the-activity-window) says so. See [Self-learning](/memoq/self-learning/). **Toggle:** Translation settings → *Send surrounding segments and project metadata to the model*. ### 4. The closest translation memory match memoQ can forward the best fuzzy match for a segment to an MT engine, and when you route it to Supervertaler that match goes into the prompt ahead of everything else – presented as the thing to adapt rather than as background reading, because a human wrote and approved it for a nearly identical source. Set it in memoQ under **Edit machine translation settings → Send best fuzzy TM match to → Supervertaler**. It is deliberately *not* governed by the document-context toggle: a match you went out of your way to route here should not disappear because you turned off surrounding context. Two things to know about its shape. It is **one** match per segment – memoQ forwards the single best one, which is what its own setting says – rather than a set to choose among. And memoQ hands over the two segments without a match rate, so the prompt cannot tell the model *how* close the match is; it is described as the closest approved rendering, and the model judges the difference from the text itself. The match is forwarded for every segment memoQ asks about, batches included. ### 5. Whether the row was rejected memoQ tells the plugin each row’s translation state. When you have rejected a previous translation of a segment, the prompt says so and instructs the model to reconsider the terminology, structure and register rather than paraphrase what you refused. No setting; it simply happens on a row you marked rejected. ### 6. Glossary terms and forbidden terms Terms matched in the segment, with their approved renderings, and any terms marked forbidden. Forbidden terms are enforced rather than merely displayed. Worth being precise about the source, because the setting’s wording is optimistic: these come from **Supervertaler’s own glossary** – the tab-separated file the [terminology plugin](/memoq/terminology/) reads – not from memoQ’s term bases. memoQ passes termbase hits to an MT plugin only through the rich lookup channel it reserves for its own engine, so a third-party plugin never receives them. Your memoQ term bases still work normally in the grid; they just do not reach the model. [Export glossary](/memoq/prompt-editor/#export-glossary-the-prompts-terms-as-the-project-glossary) is the bridge: it turns a drafted prompt’s locked terms into a glossary Supervertaler does read. While an AutoPrompt-drafted prompt is selected, its own locked-terms table is the authority and the glossary’s preferred renderings are held back – forbidden terms always travel. **Toggle:** Translation settings → *Send memoQ’s termbase hits and forbidden terms to the model*. ### 7. SuperMemory memory banks A bank of Markdown articles – `brief.md`, `terminology.md`, `style.md`, and any others you add – goes out with every request, up to about 32,000 tokens, and to AutoPrompt up to 40,000. Where a glossary gives the model flat pairs of terms, a memory bank gives it the **reasoning**: the decisions, the caveats, the client-specific overrides. The `_shared` bank travels alongside as house defaults. Banks are remembered per memoQ project, and a project you have never chosen one for uses none rather than inheriting the last – a bank carries one client’s terminology, and the wrong one is worse than none. See [Memory banks](/memoq/mcp-server/#memory-banks). ### 8. List numbering Word numbers claims, letters steps and bullets lists as paragraph properties, not as text, so memoQ’s grid never contains the `a)` or the `9.` and neither did anything the model received. Shown six unlettered steps and then *“steps a. to f.”*, a model will flag the reference as a possible defect in the source – a note that would have reached the client. Supervertaler reads the numbering out of the original `.docx`, counted over the whole document exactly as Word renders it, and sends each paragraph’s marker in front of its first segment as `[#e)]`, declared as structure to use and never to reproduce. Anything echoed back is stripped before it reaches the document. On by default, with no switch – but it needs the [live document link](/memoq/mcp-server/#the-live-document-link) connected, because that is what names the file memoQ imported. On a project checked out from a server that file is on the project manager’s machine, not yours; locate it once in [FigureLens](/memoq/prompt-editor/#where-the-documents-come-from) and the numbering is read from your copy. See [List numbering](/memoq/prompt-editor/#list-numbering-reaches-the-model-as-structure). ### 9. Figure descriptions The AI reads your text and cannot see your pictures. [**FigureLens**](/memoq/prompt-editor/#figurelens-what-the-figures-show) closes that gap: it takes the images out of your documents into the memory bank’s `figures\` folder, shows each one to the AI together with what the document says about it, and saves a description – what it shows, which figure it is, which reference signs appear on it – as `figures.md` in the bank. From then on it rides along with layer 7 and is read with every request, so the model knows that the *valve (12)* in the sentence is the thing at the top right of Figure 3. Vector drawings – EMF and WMF, which is what a drawing placed from CAD usually is – are rendered to PNG on the way out, because no AI can read a metafile. Reference signs the model reads in a drawing that appear nowhere in the text are listed for you, because that is a defect worth raising with the client before filing. Two clicks per project, then it is automatic. One AI request per image; the panel says how many and to which provider before you click, and **Describe from the text only** is a free alternative that uses what the document itself says about each figure. ### 10. The whole document Not a translation layer: memoQ hands an MT plugin about ten segments at a time and never the document, so a per-segment prompt cannot contain it. What does get the whole document is [**AutoPrompt**](/memoq/prompt-editor/#autoprompt-drafting-a-prompt-for-the-open-project), which reads it through the live document link to classify the job and draft a prompt for it. That prompt then carries the document’s character into every subsequent request – which is the point: the document is read once, expensively, and its conclusions ride along cheaply thereafter. Where memoQ shows several files merged into one view, AutoPrompt offers **All N documents memoQ is showing** so the prompt is drafted from all of them rather than whichever file you happened to pick. ## What memoQ does not give a plugin Worth stating plainly, because the Trados page lists them and the difference is not a fault in either product: * **Surrounding segments.** memoQ passes neighbouring segments only through the rich lookup channel reserved for its own AGT engine. Layer 3 exists because of this. * **Term base hits.** The same channel, the same result. See layer 6. * **Attached files.** memoQ has no equivalent; there is nothing to attach a file to. None of these can be unlocked by a setting on your side. They are consequences of the plugin model, measured rather than assumed – and where one bites, the layer above it says what stands in for it. ## Stacking the layers The default composition – everything above except the ones that need a click – is a strong baseline and the one to start from. More context is not automatically better. The context window is finite, and a mature memory bank plus a rich glossary can push a prompt into the tens of thousands of tokens. The cost is smaller than it looks, because the stable half of every request is identical from batch to batch and is cached: on one 370-segment run, caching turned roughly $10 into roughly $4. But token count is not the only cost – three overlapping sources describing the same term can contradict each other, and the model then has to reconcile them on the fly. The layers that describe the **document** rather than the **client** – 5, 8 and 9 – are cheap, never contradict each other, and there is no case yet found where switching one off improved a translation. ## Seeing what is being sent The [Activity window](/memoq/prompt-editor/#the-activity-window) in the prompt editor logs every request with its token counts – regular, cached, written, output – so a layer that is not arriving shows up as a number that does not move. It also says, once per document, whether list numbering was available and why not when it was not, and which memory bank is in force. AutoPrompt’s **Preview context…** button shows exactly what would be sent before anything is sent, and makes no AI call. ## See also * [Prompt library and editor](/memoq/prompt-editor/) – AutoPrompt, FigureLens, list numbering, the settings * [Self-learning](/memoq/self-learning/) – how confirmed segments are captured and fed back * [Terminology](/memoq/terminology/) – the glossary the model reads * [MCP server and the live document link](/memoq/mcp-server/) – memory banks, and the channel layers 8 to 10 depend on * [Context layers in Supervertaler for Trados](/trados/context-layers/) – the same idea, a different host # Getting Started This walks through a first translation, assuming the add-ins are [installed](/memoq/installation/). ### 1. Set up the translation engine Open the **Resource console** → **MT settings**. Create or edit an MT settings resource, and on the **Services** tab tick **Supervertaler**. Click **Configure plugin** and fill in: | | | | ------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | **Provider** | Anthropic, OpenAI or Google | | **Model** | a short list of the models worth recommending for that provider, with a line on each saying what it is for. **Fetch list** asks the provider for its full catalogue and **Show all models** adds it underneath – that is where to look for a model released after your copy of Supervertaler was built. The box also takes anything typed, for a gateway or a local model | | **API key** | your own key for that provider. All Supervertaler products share one key file at `C:\Users\\Supervertaler\settings\api-keys.json`, so a key you have already set in Supervertaler for Trados or Sidekick is picked up here and you can leave this empty – see [Prompt Library & Editor](/memoq/prompt-editor/#api-keys-live-in-one-file) | | **Endpoint** | leave blank unless you are using a local model or a gateway | | **Segments per request** | how many segments go into one request during Pre-translate (memoQ hands the plugin about 10 at a time, so values above 10 make no difference) | | **Prompt** | a prompt from the shared library, or *(instructions below)* to type your own – see [Prompt Library & Editor](/memoq/prompt-editor/) | | **Pre-translate via Claude Desktop (MCP)** | leave **unticked** unless you translate through Claude Desktop – see [MCP Server](/memoq/mcp-server/). Also in the prompt editor under Settings | Press **Test connection**. It translates a short sentence for real, so it exercises the key, the model name and the endpoint together – a green result means everything works. ### 2. Turn on learning Still in the MT settings resource, go to the **Settings** tab and set **Self-learning MT** to **Supervertaler**. This is what makes memoQ hand Supervertaler each segment as you confirm it. Without it the engine still translates, but it will not learn from your work. See [Self-learning translation](/memoq/self-learning/). Caution Changing this takes effect when memoQ next builds a translation engine. Restart memoQ after setting it. ### 3. Translate Open a document. With **Translation results** set to *Always*, landing on a segment fetches a translation automatically; it appears in the Translation results pane, labelled with the provider and model it came from. To translate in bulk, use **Preparation → Pre-translate** with *Use machine translation* enabled. ### 4. Confirm as you go Confirm segments as you normally would. Each confirmation is remembered, and later segments that resemble it are translated with your wording in front of the model. The effect is most visible on a document with recurring phrasing: settle a term in segment 3 and segment 40 will use it. ### Next * [Terminology](/memoq/terminology/) – add a glossary, including forbidden terms * [Self-learning translation](/memoq/self-learning/) – what is remembered, and for how long # Glossary Format A plain text file, tab-separated, one term per line. ```plaintext elektrische module electric module elektrische module electrical module forbidden koppelmechanisme coupling mechanism ``` | Column | | | ------ | ------------------------------------------------------------- | | 1 | source term | | 2 | target term | | 3 | *optional* – `forbidden` marks a target that must not be used | Blank lines are ignored, and so is any line starting with `#`, so the file can carry comments. Tabs, not spaces Columns must be separated by actual tab characters. This is the commonest reason a glossary loads with no terms. The options dialog reports how many terms it parsed – check that number after choosing a file. ### Matching Matching is case-insensitive and respects word boundaries, so `wire` does not match inside `wireless`. Where two entries could both match, the longer wins and the shorter one inside it is suppressed: `electric module` beats a bare `module`. ### Editing while you work The file is re-read whenever you save it. Keep it open in a text editor beside memoQ, add a term, save, and the next segment sees it – no restart, no reloading the project. ### Size A real term base export works fine. A 9,000-term glossary loads in well under a tenth of a second and adds under a millisecond to each lookup. Because matching is case-insensitive, very short entries are worth avoiding: a two-letter term fires on almost every segment and crowds out terms that matter. ### Converting an existing term base The plugin repository includes a converter for Supervertaler Workbench term base exports: ```bash python tools/convert_termbase.py BEIJER.tsv BEIJER-glossary.txt ``` It handles the quoting and pipe-separated variants of the export format, drops entries that translate to themselves, and reports what it skipped. # Installation ### Requirements * **memoQ 12** (translator pro or project manager) * An API key for Anthropic, OpenAI or Google * Administrator rights on the machine, once, to place the files ### Installing Supervertaler for memoQ ships as two files: ```plaintext Supervertaler.MemoQ.dll the AI translation engine Supervertaler.MemoQ.Terms.dll the terminology provider ``` Both belong in memoQ’s `Addins` folder, inside the memoQ program directory – typically: ```plaintext C:\Program Files\memoQ\memoQ-12\Addins\ ``` Copy both files there and restart memoQ. ### The unsigned plugin warning The first time memoQ starts after installation it will say it has detected one or more unsigned plugins, and ask whether to load them. **Answer Yes.** The default button is *No*, so a stray Enter keypress will decline it. Supervertaler is not yet signed by memoQ. Signing is a review process memoQ runs for plugins with proven demand; until then, this prompt appears whenever the files change. ### Where the program directory is version-stamped memoQ’s install folder carries its version number (`memoQ-12`, `memoQ-13`, …). A memoQ major upgrade creates a **new folder**, and the add-ins are not carried across – they will need to be copied again. If Supervertaler disappears after a memoQ update, this is almost always why. ### Uninstalling Close memoQ, delete the two DLLs from the `Addins` folder, and restart. To remove what Supervertaler has stored on your computer as well, use **Forget stored context** in the plugin’s options dialog before uninstalling – see [Self-learning translation](/memoq/self-learning/#what-is-stored-and-where). # MCP Server (Claude Desktop) Supervertaler for memoQ can connect **Claude Desktop** – or any AI app that runs a local MCP server – to your live memoQ project. You chat in Claude’s window; Claude reads the document you are translating, your confirmed segments and your glossary, and translates for you. The tokens are billed to your Claude subscription, not to an API key. It is the same [Supervertaler MCP Server](/trados/mcp-server/) the Trados plugin uses. What differs is what memoQ lets a plugin do – which is a good deal less than Trados – so read [What it can and cannot do](#what-it-can-and-cannot-do) before you expect Trados behaviour. > **Which AI apps work?** The same answer as for Trados: any app that runs a **local (STDIO) MCP server on your own machine** – Claude Desktop, ChatGPT’s desktop app, Claude Code. Cloud-hosted clients (the claude.ai and chatgpt.com websites) have no route to a bridge that lives on your PC. ## The one thing to understand first **memoQ never lets a plugin write into the grid.** A plugin cannot move your cursor, edit a segment or confirm anything. It can only *answer when memoQ asks it for a translation*. So Claude does not write translations into memoQ. It **stages** them. They wait inside the plugin until you run **Pre-translate** (or land on the segment), at which point memoQ asks Supervertaler for a translation and receives Claude’s. Every write into your document goes through your own hands – which is not a limitation so much as a built-in review step. The other half: a plugin only *sees* what memoQ sends it. Claude cannot read your document until Supervertaler has been shown it. One Pre-translate pass does that. ## The workflow With everything set up (below), a chat-driven job looks like this: 1. **Open the project** in memoQ and tick **Pre-translate via Claude Desktop (MCP)** in Supervertaler’s settings (see [The checkbox](#the-checkbox)). 2. **Pre-translate** with Supervertaler as the MT engine. It is instant and free: the grid stays empty, but Supervertaler now holds every source segment. 3. **In Claude Desktop:** *“Read my memoQ project and translate it into Dutch.”* Claude reads the segments, checks your glossary, and stages translations. Nothing has changed in memoQ yet. 4. **Pre-translate again.** The grid fills. Each row is marked `Claude (staged via Supervertaler MCP)` in Translation results. 5. **Confirm as you go.** If [Self-learning](/memoq/self-learning/) is on, each confirmation is visible to Claude too – *“what have I confirmed so far?”* – so a mid-job conversation about terminology is grounded in your actual choices. Rows Claude has not staged still get a live suggestion from the model as you land on them, exactly as before. The checkbox only changes what **Pre-translate** does. ## The checkbox Either in the [prompt editor](/memoq/prompt-editor/), on the **Settings** menu, which needs no project open and no memoQ running, or in memoQ under **Resource console → MT settings → Supervertaler → Configure plugin**: > ☐ **Pre-translate via Claude Desktop (MCP) instead of the API key above** *Pre-translate then only hands the segments to the chat and inserts the translations it sends back; nothing is charged to the API key. Suggestions as you move through segments still use the API key.* Both paths call an AI model. The checkbox decides **which one pays and who drives**: unticked, this plugin translates through the API key you entered; ticked, Pre-translate leaves the translating to the chat app, billed to that subscription. | | Pre-translate | Landing on a segment | | ---------------------- | --------------------------------------------------------------------------------------------------------------- | -------------------------------------------------------- | | **Unticked** (default) | This plugin translates every segment through your API key. Staged translations are used first where they exist. | Live suggestion from the model; staged first if present. | | **Ticked** | Hands the segments to the chat and inserts what it sends back. Nothing charged to the API key. | Unchanged – live suggestion from the model. | Two consequences worth knowing: * **You never have to toggle it mid-job.** Ticked is right for the whole of a chat-driven job: the capture pass is free, the delivery pass is free, and walking the document afterwards still gives you live suggestions for anything Claude did not cover. * **It is one switch for the whole installation, not a project setting.** It says how you are working at this moment, so the editor and memoQ show the same state and either can change it. * **Staged translations come through in either state.** Ticking the box is never a way to lose Claude’s work; unticking it is never a way to block it. It only decides whether *Pre-translate* spends API money on rows nothing was staged for. Leave it unticked if you use Supervertaler as an ordinary MT engine with no chat involved. That is the default, and it is what most memoQ users will want. ## Setting it up **Claude Desktop:** install the extension. 1. Download `Supervertaler-for-memoQ-MCP-Server.mcpb` (it ships with the plugin). 2. In Claude Desktop, open **Settings → Extensions → Advanced settings** and click **Install extension…** (double-clicking the file also works if `.mcpb` is associated with Claude; drag-and-drop does not). 3. In memoQ, open a project and click into any segment with Supervertaler selected as the MT engine. That creates the engine, which starts the bridge. 4. In Claude: *“What’s in my memoQ project?”* If it answers with your language pair and segment count, you are connected. If you also use the Trados plugin, both extensions coexist – Claude shows them as two servers – and they are in fact the same server exe: the memoQ one carries a single setting, `SUPERVERTALER_HOST=memoq`, which tells it to look for memoQ’s connection instead of Trados’s. **Other MCP clients** (ChatGPT desktop, Claude Code, anything that runs a local STDIO server): unzip the server exe somewhere permanent and register it with that one environment variable set: ```json "supervertaler-memoq": { "command": "C:\\path\\to\\SupervertalerMcpServer.exe", "args": [], "env": { "SUPERVERTALER_HOST": "memoq" } } ``` The exe finds memoQ’s connection file in your Supervertaler data folder (`C:\Users\\Supervertaler\memoq\runtime\bridge.json`, or wherever you moved that folder). If you ever need to point it somewhere else, `SUPERVERTALER_BRIDGE_FILE` with a full path overrides it. ## The live document link memoQ’s MT plugin interface never shows a plugin the target text, the row you are on, or even the document’s name. memoQ’s **Preview SDK** – the interface its own PDF and video preview tools use – shows all three, live. So Supervertaler ships a small preview tool, `Supervertaler.MemoQ.Preview.exe`, which registers with memoQ exactly as the PDF preview does and forwards what memoQ sends it to the plugin. With it running, Claude sees your document as it actually is: every row’s current target, memoQ’s own row order, the document’s real name, and the row your cursor is on – and it can ask memoQ to **jump to a segment**. **Setting it up (once):** 1. Run `C:\Users\\Supervertaler\memoq\preview\Supervertaler.MemoQ.Preview.exe` – inside your Supervertaler data folder, where the plugin’s deploy puts it. A tray icon appears. 2. In memoQ, accept the **Preview tool connection request** for *Supervertaler*, leaving *Auto-start with memoQ* ticked. From then on memoQ starts the tool itself. 3. The tray icon reads *memoQ: connected · plugin: connected* once you click into a segment (that is what starts the plugin’s bridge). It appears under **Options → External preview tools** alongside any other preview tools; it can be disabled there like any of them. It draws nothing on screen – it is a link, not a preview. One thing to know: memoQ’s Preview SDK works in **paragraphs**, not segments. A paragraph that memoQ splits into three grid rows arrives as one unit with the whole paragraph’s source and target. The active-segment tool still reports the exact sentence your cursor is on, and jumps can target a sentence within a paragraph. Without the tool running, the tools below fall back to what the plugin captured from translation requests, and the two cursor tools say so rather than guessing. ## What it can and cannot do Everything the Trados server can do that memoQ *cannot* comes down to one fact: memoQ has no project API and no editor API for plugins. The live document link recovers the reading half of that; writing into the document still goes through you. The table is the honest map. | Tool | memoQ | Notes | | ---------------------------------------------------------------- | :---: | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | `help` | ✓ | A menu of what you can ask, memoQ edition | | `get_project` | ✓ | Language pair, client/domain/subject, captured and live documents, what is staged | | `get_segments` | ✓ | With the live link: rows in memoQ’s order with source, **target**, and the active row marked. Without: the source segments captured from translation requests | | `get_active_segment` | ✓ | The row your cursor is on, with what is selected – needs the live link | | `go_to_segment` | ✓ | Asks memoQ to select a row – needs the live link | | `get_confirmed_pairs` | ✓ | Segments you have confirmed, via [Self-learning](/memoq/self-learning/) | | `lookup_term` / `add_term` | ✓ | Your Supervertaler [glossary](/memoq/terminology/), not memoQ’s term bases | | `stage_translations` | ✓ | **The write channel.** Translations wait until you Pre-translate | | `get_staged` / `clear_staged` | ✓ | Inspect and reset the staging area | | `list_prompts` / `get_prompt` / `save_prompt` | ✓ | The shared [prompt library](/memoq/prompt-editor/). A prompt Claude saves is recorded as drafted by the chat and marked for memoQ, so it does not appear in Trados’s list and the runtime treats its terminology the way it treats [AutoPrompt’s](/memoq/prompt-editor/#a-drafted-prompt-is-the-only-source-of-terminology) | | `list_supermemory_banks` | ✓ | Your [memory banks](#memory-banks), and how many articles each holds | | `get_supermemory_context` | ✓ | One bank’s brief, terminology and style, formatted for the model | | `search_supermemory` | ✓ | Full-text search inside a bank | | `update_segments`, `insert_into_active_segment` | ✗ | No write access to the editor – use `stage_translations` + Pre-translate | | `search_tm`, `search_studio_tm`, `compare_document_to_tm` | ✗ | memoQ’s TMs are not readable by plugins; confirmed pairs are the substitute | | `check_numbers`, `check_tags`, `check_nbsp`, `check_terminology` | ✓ | QA over the live document, paragraph by paragraph – needs the live link. `check_terminology` runs against the active Supervertaler glossary, so give it a project one: [Export glossary](/memoq/prompt-editor/#export-glossary-the-prompts-terms-as-the-project-glossary) from an AutoPrompt draft | | `find_inconsistencies` | ✓ | Repeated source paragraphs translated differently – needs the live link | | `run_verification` | ✗ | memoQ’s own QA cannot be run by a plugin; use memoQ’s Run QA | | `get_files`, `get_project_statistics`, `export_target` | ✗ | No project or file API | | `pretranslate` | ✗ | You press Pre-translate; that is the design | **Reading is complete with the live link; writing goes through you.** What you give up compared with Trados is Claude editing rows in place – and in memoQ the alternative, staging plus one Pre-translate, is a review step rather than a loss. ## Two channels for seeing the document Supervertaler captures segments in two ways, and it helps to know which is which when Claude reports what it can see: * **Translation requests** – every segment memoQ sends to Supervertaler as the MT engine. One Pre-translate captures the whole document, with its identity and metadata. This is the normal route. * **Terminology lookups** – every row your cursor lands on, through the [terminology plugin](/memoq/terminology/), *regardless of which MT engine is selected*. So a document you pre-translated with Google or from TM alone still becomes visible to Claude one visited row at a time. memoQ does not tell the terminology plugin which document a row belongs to, so these land in a per-language-pair bucket rather than under the document. `get_project` labels each captured document with its origin. ## Memory banks If you keep [SuperMemory](/trados/ai-assistant/super-memory/) banks – a folder per client, holding the brief, the terminology and the style rules you have settled on with them – Claude can read them here too. They live in one place for every Supervertaler product: ```plaintext C:\Users\\Supervertaler\memory-banks\ ``` Ask for what you want and Claude picks the tool: *“list my memory banks”*, *“read the Acme bank before you translate this”*, *“search the Acme bank for how we render ‘Vorrichtung’”*. Four things behave differently from the Trados plugin, and they are worth knowing before you rely on this: * **Claude is not told which bank is active.** You choose one in the [editor’s context bar](/memoq/prompt-editor/#what-memoq-is-using), and it is remembered per project – but that is what the plugin sends with its own translation requests. Over MCP the bank is named in the request instead, so say which one you mean. Claude will ask, or list them and let you choose. * **A name that does not exist is an error**, not a fall back to something else. Falling back would look exactly like success while feeding the model another client’s terminology, and nothing in the answer would say so. * **`_shared` is always underneath, and travels alone.** It is not a bank you select: whatever you do select is layered over it and wins wherever the two disagree – and when you select no client bank, `_shared` still goes on its own. Choosing “no client bank” is not the same as sending nothing. * **The answer is trimmed to about 6,000 tokens**, and whatever did not fit is listed under `trimmed` in the reply rather than dropped in silence. A tool result stays in the conversation and is re-sent on every following turn, so it is kept deliberately small – ask for a larger budget, or for one article by name, when you need the rest. Reading is all this does. Nothing writes into a bank from memoQ; you edit the files yourself, in Obsidian or any text editor. ## Troubleshooting **“Handshake file not found.”** memoQ has not created a Supervertaler engine yet in this session. Open a project and click into a segment with Supervertaler selected as the MT engine. **Claude says the project is empty.** Nothing has been captured yet. Run Pre-translate once, or visit some segments. **Staged translations do not appear after Pre-translate.** They are matched by exact source text. If you edited a source segment after Claude read it, the match fails – ask Claude to re-read and re-stage that segment. Also check that Supervertaler is the selected MT engine for the Pre-translate run. **`get_confirmed_pairs` is always empty.** Self-learning is not on. See [Self-learning translation](/memoq/self-learning/). # Prompt Library & Editor Supervertaler for memoQ translates with the instructions you give it. Those instructions can be typed straight into the settings dialog, or chosen from the **shared Supervertaler prompt library** – the same folder of prompts the Trados plugin uses, so a prompt tuned in one tool is available in the other. memoQ gives an add-in no window of its own, so the library cannot be a panel inside memoQ. Instead there is a small **Prompt Library editor**, opened from the settings dialog, that runs alongside memoQ. ## Choosing a prompt In **Resource console → MT settings → Supervertaler → Configure plugin**, the **Prompt** dropdown lists every translation prompt in the library, grouped by folder. Pick one and its text appears (read-only) in the **Instructions** box below. Choose **(instructions below)** instead to type your own; the box becomes editable. The dropdown stores *which* prompt you chose, not its text. Edit the prompt anywhere – in the editor, in the Trados plugin, in a text editor – and memoQ uses the new version on the next segment. Only prompts in the library’s **Translate** folder are offered. Proofreading and QuickLauncher prompts exist for other tasks and would produce commentary where a translation belongs. ## The editor Press **Edit…** beside the Prompt dropdown. * **Left:** the library as a tree, folders and prompts. Select one to open it. * **Right:** name, description, which product it is for, sort order, and the prompt text with Markdown headings and `{{PLACEHOLDERS}}` highlighted. * **Toolbar, left:** New, Save, Placeholder, AutoPrompt. * **Toolbar, right:** whether Pre-translate goes to Claude Desktop, the [Activity window](#the-activity-window), and Translation settings. Everything else is on the **File**, **memoQ**, **Settings** and **Help** menus. The **Claude Desktop** button on the right is a switch, not a command, and its caption says which mode is *on* rather than what pressing it would do. It decides whether Pre-translate spends your API key or hands the segments to the chat, so it is worth a glance before a long run. It is the same setting as **Settings → Pre-translate via Claude Desktop**, and as the checkbox in memoQ’s own dialog – change it anywhere and all three follow. Save with **Ctrl+S**. A prompt marked read-only in the library (the built-in defaults) can be read but not overwritten; make a copy under a new name instead. Any settings the file carries that the editor does not have a field for – tags, favourites, QuickLauncher flags set by the Trados plugin – are preserved untouched on save. The status line under the description says which ones the file has. ### What memoQ is using The bar under the toolbar opens with the **memoQ project** these apply to, because everything on it is recorded against a project. When it reads **no project yet**, in red, memoQ has not sent a translation request and the plugin does not know where it is – usually because Supervertaler is not selected as the MT engine in a newly created project. A memory bank chosen at that moment is filed against whichever project came before, silently, so it is worth a glance before changing anything. Then three things memoQ will apply to every translation: **Prompt**, **Glossary** and **Memory bank**. Click any of them to change it. Each opens a list with a filter box rather than a dropdown menu, because all three grow with the work – a prompt library reaches forty entries quickly, and a bank per client does the same. All three can also be set to nothing: the glossary list has a **(none)** row, and a **Browse…** row at the end for a glossary that lives outside the glossaries folder. These are the choices that change between jobs, which is why they are here rather than in Translation settings: they are what the model knows before it is shown a segment. Each is also the same setting memoQ’s own dialog shows, so either place can change it. The memory bank is remembered **per project**. Choose one while working on a job and it comes back when you return to that job – and a project you have never chosen one for does *not* inherit the last one, because a bank carries one client’s terminology and the wrong one is worse than none. What such a project gets instead is the row called **(no client bank – shared defaults only)**. Your `_shared` bank still travels: it is where the material that applies to every job regardless of client lives, so it is never switched off by not choosing a client. See [Memory banks](/memoq/mcp-server/#memory-banks) for where they live. A bank is sent whole with **every** translation request, up to about 32,000 tokens, and to AutoPrompt up to 40,000. Anything that does not fit is dropped by priority and named in the [Activity window](#the-activity-window) rather than lost quietly – if you see a file listed there, that is the budget, not a fault. The cost of carrying it is small because the same text is sent every time and providers cache it: on one 370-segment run the bank and prompt together came to 45,870 tokens, and caching turned roughly $10 into roughly $4. ### Placeholders Prompts use `{{SOURCE_LANGUAGE}}` and `{{TARGET_LANGUAGE}}` rather than naming languages, so one prompt serves every language pair. memoQ fills them in per project. **Insert placeholder** lists the ones memoQ can fill. A placeholder memoQ cannot fill – `{{SOURCE_SEGMENT}}`, say, which only the Trados plugin provides – is shown in red, and the editor warns that it will reach the model as empty text. Do not use those in a prompt meant for memoQ. ### Which product a prompt is for **Available in** can be *both*, *trados* or *memoq*. memoQ’s dropdown hides prompts marked for Trados only. Leave it on *both* unless a prompt genuinely depends on something one product cannot supply. A prompt tied to one product says so in its **filename**: `Patent claims EN-NL [memoQ].md`, `Define [Trados].md`. Prompts available to both carry no marker, so the absence of one is itself readable – which is the point, because in Explorer the metadata header is not visible and every prompt otherwise looks alike. The marker is written from the **Available in** field on every save and stripped again on every read, so it is a label rather than a setting. Renaming the file in Explorer does not change which product a prompt is for, and the next save puts the old marker back: change the field, not the filename. The editor’s tree and the Prompt dropdown show the same thing in words. ## Where the library lives `C:\Users\\Supervertaler\prompt_library\` – one Markdown file per prompt, with a small metadata header. **Open folder** in the editor takes you there. The files are plain text; nothing stops you editing them directly, and a folder synced between machines carries the whole library with it. ## AutoPrompt: drafting a prompt for the open project Press **AutoPrompt…** in the editor’s toolbar, or choose it from the **memoQ** menu. Supervertaler reads the document you are translating, your glossary hits in it and anything you have already confirmed, and has the AI write a prompt tailored to that job – domain, register, a locked glossary, the lot. The result is saved under **Translate** and opened for you to review; then pick it from memoQ’s **Prompt** dropdown. Before it runs you choose the document (if several are captured), and can add a briefing – client, audience, style, what to avoid – which the AI treats as authoritative. The briefing is the one input nothing else supplies: the filing route, a discrepancy you already know about, anything true of this job that is not in the document or the memory bank. The two checkboxes grey out when they have nothing to offer – no glossary is active, or nothing has been confirmed in this document yet – and say which it is, rather than sitting there ticked and doing nothing. **Preview context…** shows you exactly what will be sent, before anything is sent: the extract from your document, the glossary hits, the segments you have confirmed, the briefing you typed, and the instructions the AI is given about writing a prompt for memoQ. It makes no API call and costs nothing, and the briefing box stays open behind it – so the loop is look, add what is missing, look again, then generate. Three things to know: * **memoQ must be running with a Supervertaler engine active**, and the document must have been captured – one Pre-translate does it (free, with the [Claude Desktop box](/memoq/mcp-server/#the-checkbox) ticked). The plugin only sees what memoQ has sent it. * **It uses the provider, model and API key from your Supervertaler settings.** Two calls: a short one to classify the document, then a long one to write the prompt. Expect a minute or two. * **The prompt is written for memoQ, not copied from the Trados recipe.** Single-segment lookups are handled as well as batches; tag markers must be reproduced exactly; translations you have confirmed outrank the prompt’s own glossary; and it is kept to 1,500–3,000 words because memoQ re-sends the whole prompt with every ten-segment request. Draft it again later in the job and it gets better: by then it can see what you have confirmed, which is stronger evidence of how you want *this* document translated than the source text alone. ### A drafted prompt is the only source of terminology A prompt AutoPrompt wrote ends in a locked-terms table chosen for this document. So while one is selected, **the glossary’s preferred renderings are not sent to the model as well** – two lists of terminology that were never written to agree, with nothing saying which wins, is a worse position than one list. **Forbidden terms still go.** A preferred rendering is advice, and two sources of advice can contradict each other confusingly; “never use this word” is a constraint, and there are few of them. So they travel whatever prompt is selected. Nothing else about the glossary changes: it still drives the terminology pane, the QA check and AutoPrompt’s own reading of the document. Only the per-request injection stops, and the [Activity window](#the-activity-window) says so once per prompt. The consequence worth remembering: a term you forbid **after** a prompt was drafted is enforced immediately, but a preferred rendering you add afterwards is not – draft the prompt again to take it in. **Export glossary** is the other half of that loop: derive the glossary *from* the prompt and the two cannot contradict each other in the first place. Prompts saved from the chat over [MCP](/memoq/mcp-server/) count as drafted too, and are marked for the product you were connected to. ### List numbering reaches the model as structure The letters on the steps of a claim – a), b), c) – are not text. Word generates them from the paragraph’s list settings, so memoQ’s grid does not contain them and neither does anything the plugin is sent. Shown six unlabelled sentences followed by *steps a. to f.*, a model will flag the reference as a possible source defect, and it will do so on every lettered list in every document. With the [live document link](/memoq/mcp-server/#the-live-document-link) connected, the plugin knows which file memoQ imported, reads the numbering out of it – counted over the whole document, exactly as Word renders it, restarts and all – and sends each paragraph’s marker in front of its first segment as `[#e)]`. The prompt tells the model this is structure: use it to resolve cross-references and keep list items parallel, never translate it, never reproduce it. Every reply is checked before it reaches the document, and an echoed marker is removed, with a line in the [Activity window](#the-activity-window) saying so – that line is your evidence, per model, that the rule is being obeyed. There is no switch for it. Without the live link, or for a document with no lists, the model is instead told that numbering is supplied by the document and not to flag its absence. The Activity window says which of the two happened, once per document. On a project checked out from a server the file memoQ names is on the project manager’s machine, not yours; locate it once in [FigureLens](#where-the-documents-come-from) and the numbering is read from your copy. ### Translator comments Where a note is genuinely necessary – an ambiguity in the source, a term that could go two ways, a probable defect in the original – a drafted prompt has the AI put it inline at the end of the target as a `[[TC: …]]` marker. Supervertaler for Trados uses the same form, so a prompt written for one product reads correctly in the other. Nothing extracts these for you, and that is deliberate. You read them in the grid as you review, decide which are worth keeping, turn those into real memoQ comments on the segment, and delete the marker from the text. Search for `[[TC:` to find them all. ## FigureLens: what the figures show The model sees a document’s text and not its pictures. A claim that names *part 12* is translated by a model that has never seen part 12, and a figure’s caption is often the only description of it anywhere in the text. **FigureLens…** on the toolbar (also **memoQ → FigureLens…**) is the panel that closes that gap, in two steps. It is named for what it does beside [TermLens](/memoq/terminology/): that one shows the model the terms in a segment, this one shows it the pictures. **Step 1 – Extract images** copies every image out of the documents into the active memory bank’s `figures\` folder, named after their figure numbers – `Figure 01.png`, `Figure 02.png` – so a folder of drawings reads like the document. Free, no AI. The panel says how the labels were arrived at: paired by position and checked, taken from nearby text, or withheld when it could not tell. **Step 2 – Describe images with AI** shows each image to the model, together with what the text says about it, and saves the descriptions as `figures.md` in the memory bank – one paid request per image, and the panel states the count and the provider before you click. Every prompt reads that file from then on, so read it first: a wrong caption would be invisible and everywhere. Reference signs the model reads in a drawing that appear nowhere in the text are listed at the end, because that is a defect worth raising with the client before filing. **Describe from the text only** is the free alternative – what the document itself says about each figure, without looking at the images – and either replaces the other, after asking. The images folder is not chosen: it is inside the memory bank, because that is where `figures.md` goes and the two belong together. With the shared bank or no bank active, the Result line offers to create a bank named after the memoQ project and switch to it. ### Where the documents come from memoQ never keeps the original file in its project folder – a local project stores the filename as an empty placeholder, a project checked out from a server stores only memoQ’s own data – so the panel works from the file memoQ imported, wherever that was: * On a **local project**, the [live document link](/memoq/mcp-server/#the-live-document-link) reports the path memoQ imported each document from, and the panel finds the file there without any setting. * On a **server project**, memoQ records the path the file had on the project manager’s machine, which does not exist on yours. The panel lists those documents as *not on this computer* and offers **Locate the original document…** – point it at the copy you were sent, once, and it is remembered for that document in `C:\Users\\AppData\Local\Supervertaler.memoQ\document-files.txt`. Locating a document here also switches on [list numbering](#list-numbering-reaches-the-model-as-structure) for it. * **Add a document file…** is for a Word file memoQ has said nothing about at all. Its images are read from the file directly; the file is remembered for the active memory bank. Supervertaler finds the project folder by asking memoQ where it keeps its projects – the custom folder set under **Options → Locations → Projects**, the default `C:\Users\\Documents\My memoQ projects` when you have not set one, and memoQ’s own register of every project, which names each one’s actual folder. Moving your projects folder therefore needs nothing here, and projects left behind in the old location are still found. **Document images report** at the bottom writes a Markdown listing of every image in every document – label, size, caption, the text around it – into the memory bank and opens it. No AI call. ## Export glossary: the prompt’s terms as the project glossary An AutoPrompt draft ends with a locked-terms table – a dozen or so renderings chosen for this document. That table is exactly what the [terminology plugin](/memoq/terminology/) and the `check_terminology` QA tool should work from: a general glossary flags *application → aanvrage* in every paragraph of a software patent, a project glossary knows better. Choose **memoQ → Export this prompt’s terms as a glossary** with the prompt open. Supervertaler reads every table in it that names a source and a target column, turns notes of the form *never “apparatus”* into forbidden entries, and writes a tab-separated glossary file to `C:\Users\\Supervertaler\memoq\glossaries\.txt`. Answer yes when it asks and that file becomes the active glossary immediately, whether or not memoQ is running. The file is plain text – edit it freely; the plugin re-reads it whenever it changes. Any prompt with a table laid out the same way works, not only AutoPrompt’s. ### Settings **Settings → Translation settings** holds how Supervertaler translates: provider, model, endpoint, parallel requests, segments per request, and whether termbase hits and surrounding segments are sent to the model. These are the same settings as memoQ’s own Supervertaler dialog, reading and writing the same file, so either place can change them and both show the same values. The **Model** list is short on purpose: three to five models per provider, each with a line saying what it is for. A provider’s own catalogue runs to thirty or forty entries – image models, speech models, dated snapshots of the same model – and a list like that is one nobody in a hurry can choose from. A model that has been superseded is removed rather than annotated, so what is left is what is worth using today. **Fetch list** asks the provider for its full list, using the API key below. **Show all models** then shows everything it returned under the short list, which is where to look for a model released after your copy of Supervertaler was built. The fetched list is remembered, and so is the tick, so this is a decision you make once. The line under the button says when the list was last fetched and how much of it is beyond the short list. The box stays typeable throughout, so a gateway, a private deployment or a model that appears in neither list can be entered by hand – and a model already saved in your settings keeps working whether or not it is in the list on screen. Changing the provider changes three things together: the model list, the model itself – to that provider’s first recommendation, since a model belonging to another provider can only fail – and the API key, which is re-read for the provider you have just chosen. A key you type here is remembered per provider while the window is open. memoQ’s own **Configure plugin** dialog has the same three controls, reading and writing the same settings, so it does not matter which one you use. **Segments per request** can only lower what memoQ does, not raise it – memoQ hands a plugin about ten segments at a time during Pre-translate, however high this is set. Lowering it is still worth doing if a model keeps returning fewer translations than it was sent. **Settings → Pre-translate via Claude Desktop (MCP)** is on the menu itself, and on the right of the toolbar, because it is the one that gets flipped between jobs rather than set once. See [MCP server](/memoq/mcp-server/). ### API keys live in one file Every Supervertaler product reads the same file: ```plaintext C:\Users\\Supervertaler\settings\api-keys.json ``` One key per provider, plain text, editable in Notepad. A key pasted here works in Supervertaler for Trados and Supervertaler Sidekick as well, and rotating one means changing one line in one place. Before this file existed there were three dialogs in three products each keeping their own, which is how an hour goes missing to a key for one service pasted into another’s box. Keys are stored under the provider ids `claude`, `openai` and `gemini`. Sidekick keeps its machine-translation keys in the same file under their own names – note that `google` there is Google Translate, not Gemini. Plain text is deliberate, and the same choice Supervertaler for Trados has always made: a key that can be rotated by pasting a line into a text file is a key that actually gets rotated, and anyone who can read that file can already read everything else in your profile. The **API key** box shows the key for the provider you have selected and writes back to that file. If you had a key configured before the file existed – in memoQ’s own settings, or in Trados’s – it is copied in the first time Supervertaler needs it, so there is nothing to do. If you paste a key that plainly belongs to another service, the line under the box says so as you type: *This is an OpenAI key, not an Anthropic one.* That is worth more than the provider’s own answer, which is that the key is incorrect. ## The Activity window memoQ’s Pre-translate dialog is modal and says only *Processing*, for as long as the run takes: no engine, no model, no count, and no sign when something is wrong. **memoQ → Activity…**, or **Ctrl+L**, opens a window that shows what Supervertaler is actually doing. It is a window of its own rather than a panel so that it can sit over memoQ while that dialog holds the screen. Tick **Keep on top** and you can watch a Pre-translate run from the first batch to the last. What it shows: the engine and model each project starts with, the glossary as it loads and how many terms came out of it, warnings when the selected prompt or glossary faces the opposite language pair, every batch with the segments sent, the segments returned and the glossary terms matched, AutoPrompt drafts, and anything that failed. A batch that comes back short is called out rather than logged flatly, because that is the failure that quietly shifts every translation after it. Three lines are worth knowing by sight: * **Bank** – which memory bank a project switched to, and once per job how much of it is being sent. If it ends with a file listed as *not sent*, that is the budget trimming by priority, not a fault. * **Terminology** – said once when a drafted prompt is holding the glossary back, so a quiet change to what reaches the model is never silent. * **The token count on each batch** – `tokens: in 1,041 (cache write 45,870) out 1,233`. The prompt and the bank are identical on every request of a run, so providers cache them: the first batch writes, the rest read at a tenth of the rate. If *cached* never appears across a long run, something is re-sending the block at full price. **Show everything** un-hides the per-request diagnostics – memoQ’s capability probes, lookup sessions, single-segment translations – which are what you want when something is wrong and noise the rest of the time. The window reads the plugin’s own log, `C:\Users\\AppData\Local\Supervertaler.memoQ\plugin.log`, rather than being fed by the plugin. So it shows what happened before you opened it, it works whether or not memoQ is running, and closing it costs nothing. Its position and size are remembered. ## Drafting prompts with Claude If you use the [MCP server](/memoq/mcp-server/), Claude can write into this library: *“Draft a translation prompt for this project and save it.”* It reads the captured document, your confirmed segments and your glossary, saves the result as a new prompt, and you pick it from the dropdown. The editor is where you review and tune what it wrote. # Self-learning Translation Supervertaler remembers the segments you confirm, and shows the most relevant ones to the model when it translates later segments in the same document. This is what makes the second half of a document read like the first. Settle on a rendering once, confirm it, and everything that follows is translated with your choice in front of the model rather than against a blank slate. ### Turning it on **Resource console → MT settings → your resource → Settings tab → Self-learning MT → Supervertaler.** Then restart memoQ. Until this is set, memoQ never passes confirmed segments to the plugin, and Supervertaler translates without memory. ### What gets remembered Only segments **you confirm**. Not the AI’s raw output. That distinction is deliberate. Feeding a machine its own guesses compounds its mistakes; the value of these examples is precisely that a human approved them. For each confirmation Supervertaler stores the source text and the target text, and nothing else – no tags, no formatting, no metadata. Re-confirming a segment replaces the earlier version, so your latest decision is the one that counts. ### How much is sent to the AI Not all of it. For each new segment, Supervertaler picks the **five** stored pairs sharing the most vocabulary with it. A pair with no words in common is never sent. So a document with hundreds of confirmed segments still contributes only a few short examples per request – the ones most likely to matter. ### What is stored, and where Memory is kept per document *and* language pair, and written to: ```plaintext C:\Users\\AppData\Local\Supervertaler.memoQ\document-memory\ ``` It survives closing memoQ, so picking a job back up the next morning keeps yesterday’s decisions. Limits: 500 pairs per document, 200 documents, and anything untouched for 60 days is discarded. This is client text on your disk These files contain source and target segments from real work. They are stored under your own user profile, are not synced anywhere, and never leave your computer. The plugin’s options dialog shows how many documents and how much space are held, with a **Forget stored context** button that deletes all of it. Your translations in memoQ are not affected. ### What it is not This is not adaptive machine translation in the sense that ModernMT or Lara mean it. No model is trained or fine-tuned, and nothing is sent to Supervertaler – the examples are simply included in the request to the AI provider you chose, alongside the segment. The practical differences: it works within a document rather than across your whole history, it is lost if you clear it, and it is entirely inspectable. Nothing happens that you cannot see in the prompt. # Terminology Supervertaler reads a glossary file and uses it in two places at once: memoQ’s terminology pane, and the AI’s prompt. ### Why a file, and not a memoQ term base memoQ does not let a plugin read its own term bases. A term base attached to your project is visible to you and to memoQ’s QA, but not to a machine-translation plugin – so a term marked forbidden in memoQ will not, on its own, stop the AI using it. Supervertaler works around this by being a terminology source in its own right. Terms in its glossary reach the model because Supervertaler puts them there. ### Setting it up **Options → Terminology plugins.** 1. Tick **Perform terminology plugin lookups while working in the translation grid**. Nothing happens until this is on. 2. Find **Supervertaler terms** in the list. It reads *Not configured* until a glossary is set. 3. Click its **Options**, choose your glossary file, press OK. 4. Tick **Enable plugin**. Restart memoQ. The same glossary setting is reachable from the translation engine’s options dialog – both halves of the plugin read one file. ### Which glossary is active One setting, three consumers: the terminology pane, the prompt sent to the model, and the terminology QA check all read the same file. It is shown in four places, so you never have to guess which one is answering: * **The engine’s options dialog** (Options → Machine translation → Supervertaler → Options) has a *Glossary* row naming the active file with its full path, in red if the file has gone missing. *Change…* opens the same chooser the terminology plugin uses. * **Every hit in Translation results** carries the glossary’s file name in its grey footer (*Supervertaler · patent eng-dut.txt*). * **The [prompt editor](/memoq/prompt-editor/)** names it in its status bar, where clicking it changes it, and works with memoQ closed. * **Claude’s project report** (`get_project` over the [MCP server](/memoq/mcp-server/)) includes the path under *activeGlossary*. Exporting a glossary from the [prompt editor](/memoq/prompt-editor/) makes that file the active one immediately; the options dialog and the footer show the new name on the next lookup. ### What you get **In the grid.** Matched terms are highlighted in the source segment: green for approved terms, red for forbidden ones. **In Translation results.** Each match appears as an entry showing the target term, the source term it matched, and – for a forbidden term – the wording struck through under *Do not use*. **In the prompt.** Approved terms are sent as the client’s preferred wording; forbidden terms as absolute constraints. **For Claude, if you use the [MCP server](/memoq/mcp-server/).** Every row your cursor lands on is looked up by this plugin whatever MT engine is selected, and Supervertaler remembers each one – so a document you pre-translated with Google or from TM alone still becomes visible to Claude as you walk through it. Claude can also read and add glossary entries directly (*“we agreed* draagarm *=* support arm *– add it”*). ### Preferred, not mandatory Approved terms are given to the model as a strong steer it may override when an entry is clearly wrong for the sentence at hand. That asymmetry is deliberate, and it comes from a real failure. A patent term base may quite correctly render *applications* as *aanvragen* – in the sense of a patent application. Told to use terminology verbatim, the model translated “Mashup applications” as “Mashup-aanvragen”, which is nonsense. A translator treats a term base as guidance they may set aside with reason, and the model is asked to do the same. **Forbidden terms are not softened.** They are stated as absolute, because that is what a forbidden term is for. ### Format See [Glossary format](/memoq/glossary-format/). # Troubleshooting ### Supervertaler does not appear at all Check `Supervertaler.MemoQ.dll` is in memoQ’s `Addins` folder, and that you answered **Yes** to the unsigned-plugin prompt on startup – its default button is *No*. If memoQ was recently upgraded to a new major version, the add-ins need copying into the new program folder. See [Installation](/memoq/installation/). ### It translates, but never learns **Self-learning MT** is not set. Resource console → MT settings → your resource → **Settings** tab → **Self-learning MT** → *Supervertaler*, then restart memoQ. Advertising the capability only makes the engine eligible; memoQ does not send confirmations until it is actually selected there. ### Terminology is not showing Three things must all be true, under **Options → Terminology plugins**: 1. **Perform terminology plugin lookups while working in the translation grid** is ticked 2. **Supervertaler terms** does not read *Not configured* – i.e. a glossary file is set 3. **Enable plugin** is ticked for it ### The glossary loads no terms Almost always spaces where tabs should be. The glossary options dialog reports how many terms it parsed; if that reads zero with a file selected, open the file in an editor with whitespace visible and check the separators. ### The panel names the wrong project, or nothing reaches the model Supervertaler learns which project is open from two channels: a translation request from memoQ, and the [live document link](/memoq/mcp-server/#the-live-document-link), which reports every document memoQ shows. With the link connected, the panel follows within a couple of seconds of your clicking into a segment of the new project. Without it, the plugin has only translation requests – an add-in cannot ask – so until the first request of a session the panel still names the last project memoQ *did* ask about, and a memory bank chosen at that moment is recorded against that earlier project. **Click the project name** in the panel (or **memoQ → Sync with memoQ now**) to take the project from the document memoQ is showing this instant. If it cannot, it says why: the live link is not connected, memoQ has not reported a document yet, or no project folder holds that document. Where Supervertaler looks is memoQ’s own answer – the custom folder set under **Options → Locations → Projects**, the default `C:\Users\\Documents\My memoQ projects` when none is set, and memoQ’s register of every project, which names each one’s actual folder – so moving your projects folder needs no setting here, and projects left behind in the old location are still found. If clicking into a segment does not update the panel, memoQ is not calling the engine at all. Open **Project home → Settings → MT settings** and look for the line **“MT plugins are currently disabled.”** A project checked out from a memoQ server can have MT plugins switched off by the project manager, and there is no client-side setting that overrides it. The terminology provider still works in such a project, because it is not an MT plugin; translation through Supervertaler does not, in any mode, and neither does staging from Claude Desktop, which enters the grid through the same engine. Ask the project manager to allow MT plugins, or take the document out through a bilingual export. If you chose a memory bank while the panel was stale, open the bank chooser again once the right project is shown – that records the choice against the right project – and check the previous project’s row in `C:\Users\\AppData\Local\Supervertaler.memoQ\memory-bank-projects.txt`, one project GUID per line, in case it now names a bank it should not. ### The log The **Activity** window in the prompt editor (**memoQ → Activity**, or Ctrl+L) shows the same log live. On disk it is: ```plaintext C:\Users\\AppData\Local\Supervertaler.memoQ\plugin.log ``` with a fallback at `C:\Users\\AppData\Local\Temp\Supervertaler-memoQ.log` if that folder cannot be written. It records what memoQ asked for and what was sent – segment sizes, how many glossary terms matched, how many remembered segments were used, and any errors. It does not contain the text of your translations. A typical healthy line: ```plaintext translate: 199 src chars, 0 tag(s) -> 239 target chars, 0 tag(s) | recall: used 2 of 7 held | terms: 7 ``` meaning: a 199-character segment; two remembered segments and seven glossary terms sent with it; a 239-character translation returned.