See how translation really happens.
The Stone Guest ToolKit is a complete research platform for studying how people translate, alone or in teams, from wherever they are. It handles everything from sign-up and consent to recording the work and analysing it, so researchers can focus on the questions instead of the plumbing.
What this is, in plain words
No technical background needed for this part.
A finished translation tells you very little about how it was made. Where did the translator hesitate? What did they look up? When did the team talk, and did it change the text?
Translation-process research answers these questions by observing the work as it happens. Traditionally that meant bringing people into a lab. The Stone Guest ToolKit lets the same kind of study happen remotely: participants work on their own computers, at home or in class, and the observation still happens carefully, with their consent, and in a way that can be analysed rigorously afterwards.
The name comes from the stone guest of the Don Juan story: an uninvited figure who quietly sits at the table. Our "guest" is invited, polite, and leaves the moment you ask it to.
What a study can observe
Who it's for
Researchers
Translation studies, cognitive science, applied linguistics and writing research. Design a study in a web panel, run it, and get clean, documented data out.
Participants
Students, professional translators and volunteers. Sign up in a few minutes in Turkish or English, install one small program, and take part from home.
Ethics boards & supervisors
Versioned consent, an encrypted identity vault, a full audit trail and automatic deletion after the retention period. Everything is recorded and checkable.
Where everything lives
Each part of the platform has its own address on translation-cognition.com, all on our own server. Participants only ever need the first one; we send them the rest when they need it.
How a study works
Here is a typical team-translation study from a participant's point of view. Researchers can switch any part on or off; a simple questionnaire study uses only a few of these steps.
- Sign upOpen the study link, check eligibility and read the information sheet in Turkish or English.
- Give consentTick each item you agree to. The exact version you saw is stored with the time.
- Get your IDConfirm your email and receive a participant ID (PID). From now on you are known only by it.
- Quick testsA short vocabulary test (LexTALE) and background questions, scored automatically.
- Meet your teamYou are placed in a balanced team and get a link to set your own chat password.
- Install Stone GuestDownload it for your system and run the installer. A self-test checks it works.
- TranslateWork as usual. Stone Guest records your session; the team chats and drafts together.
- Upload & finishUpload each session file, answer the closing questions, and download your own data if you like.
Behind the scenes, the researcher
All of this happens in the panel at toolkit.translation-cognition.com, with no command line or programming.
Create the study, or clone an earlier one
Phases, conditions, teams and which parts the study uses: team chat, recording, questionnaires, shared drafts, uploads.
Studies → New studyWrite the consent text and sign-up form
Side by side in Turkish and English, with a live preview. Publishing creates a fixed version; changes become a new version, and nobody's answers are mixed up.
Study → Consent · RegistrationChoose questionnaires and tests
Pick from the library (LexTALE, background, NASA-TLX, reflections) or build your own with Likert scales, choices and free text.
Study → InstrumentsSet up the team chat spaces
One click per team, or all teams at once. The panel creates the space, its channels, a friendly study bot and the welcome messages.
Zulip → Realms → New realmConfigure Stone Guest for this study
Task names, instructions, what counts as the "translation window" and how private the recording is. Participants download a version already set up for your study.
Study → Stone Guest packageRehearse in pilot mode
Clone the study into a sandbox with pretend participants and fast-forward the clock to check every scheduled message and team change before anyone real is involved.
Study → Pilot
Open recruitment and watch sign-ups live
Share one link. A live funnel shows how many people reached each step; a waitlist fills automatically when the study is full.
Study → Live registrationsForm balanced teams
Allocation spreads language, experience and other factors evenly across teams, with a preview before you commit. Plan a mid-study reshuffle and see who carries which thread into the new team.
Study → TeamsLet the schedule run itself
Weekly briefs, reminders and deadline nudges go out on time by email or chat, respecting quiet hours, with a dry run first. Participants can ask the bot for their deadline or status.
Study → AutomationKeep an eye on data quality
Flags appear when someone hasn't uploaded for days, a session looks too short, or someone is active in chat without a recording.
Study → Data → Quality flagsRead the conversation as it happens
Browse every team's chat by channel and topic, see who is online, and view each person's active hours as a heatmap.
Study → Zulip → Messages · ActivityKeep field notes
Tagged memos and incident reports. Every message you send to participants is automatically marked on the timeline, so its influence can be separated in the analysis.
Study → Notes
Replay a session
Watch a participant's keystrokes, the team's draft and the chat play back together on one timeline. Useful for coding and for cued interviews.
Data → Session replayRun the Task Segment Framework analysis
Hundreds of indicators per participant and task, with the framework's rules on what each may be used for enforced automatically.
Analysis → TSFRate the translations
Blind rating rounds with rubrics and multiple raters, and inter-rater agreement computed for you.
AssessmentExport
Tidy tables keyed by participant ID, ready for R or Python, plus formats compatible with CRITT TPR-DB and Inputlog.
Data → ExportsWrite it up
A generator turns the study's setup, allocation, instruments and logged deviations into a draft methods section.
Study → Protocol reportShare the dataset
Build an anonymised package with data, codebook, manifest and README, ready for OSF. Identities are purged automatically when the retention period ends.
Data → Dataset release
Stone Guest, the recorder
A small program participants install on their own computer. It records how the translation is produced and saves it as a file on that computer. It never connects to the internet: participants decide when to upload the file themselves.
What it records
- Each key press and release, with exact timing, correct for Turkish keyboards
- Mouse clicks and scrolling; copy, cut and paste
- Which kind of window is in front: the translation, the source text, a search, the team chat, an AI tool
- Task start and switch markers you press yourself
What it never does
- Send anything over the internet; there is no networking code at all
- Record what you type into password fields
- Run in the background without a visible indicator
- Keep recording when you press pause or stop
Getting started as a participant
Download
Use the link from your study, or stoneguest.translation-cognition.com. The page picks the right version for your computer.
Install
Unzip and double-click the installer. On a Mac, allow Accessibility and Input Monitoring when asked. A quick "press any key" test confirms it works.
Record & upload
Start a session, translate, stop. The end screen shows your file and a short integrity code. Upload it at upload.translation-cognition.com.
Everything in one place
Each study switches on only the parts it needs. A one-session questionnaire study and a five-week team study run on the same platform.
Sign-up & consent
Form builder with eligibility rules, waitlists, email verification and versioned bilingual consent. Answers save as people go, so they can come back later.
Team chat (Zulip)
A private space per team, created from the panel. Accounts are made for consenting participants only, with nicknames that hide real identities.
Balanced teams
Minimisation or balanced batch allocation, a drag-and-drop team board, and a reshuffle planner with a visual map of who moved where.
Scheduling & bots
Rules like "the day before phase 2, at 9:00, post the brief to every team". Bots answer participant questions and forward the rest to you.
Questionnaires & tests
A library and builder with automatic scoring. Any instrument can be sent as its own personal link, by email or chat.
Session studies
Step-by-step flows with timed activities and timestamped notes, repeated weekly if needed. Every answer appears in the panel live.
Shared drafts (Etherpad)
Pads per team or per person; every revision is aligned with keystrokes and chat on one timeline.
More data sources
Wristband exports (e.g. Empatica EmbracePlus), AI-chat logs and other files are aligned using sync markers.
MT & AI use
Set a policy per study, tag DeepL, ChatGPT and similar tools automatically in recordings, and ask about them in a self-report.
Session replay
Keystrokes, drafts and chat played back together, with a segment viewer for manual coding and a second coder.
Interaction analytics
Reply networks, response times, participation balance across team members and message volume per phase.
Rating & interviews
Blind rating rounds with inter-rater reliability, interview scheduling, and certificates or compensation for participants.
Analysis built in: the Task Segment Framework
The platform's main analysis follows the Task Segment Framework (Muñoz & Apfelthaler, 2022).
In plain words: the framework cuts a translation session into short stretches of activity, separated by pauses. A pause is only counted as "long" relative to that person's own typing rhythm, so fast and slow typists are compared fairly. Each stretch then gets a label, such as adding text, changing text or searching.
For researchers: thresholds are computed per participant and task (lower: 2 × median within-word interval; upper: 3 × median between-word interval). Every indicator carries its role, expected direction, epistemic level and unit, and the platform enforces them: only core indicators can be outcomes, interpretive ones are marked, and incompatible units never combine silently.
The indicator table is imported as data, not hard-coded. New versions of the table, or other frameworks, can be added alongside, and every result records which version produced it.
A session, segmented
Exports come out as tidy tables (participant × task × segment, indicator or progression bin) with a generated codebook, ready for mixed-effects models in R (lme4) or Python (statsmodels).
Privacy & ethics
Designed for KVKK and GDPR from the first line. In short: we collect only what a study needs, keep names apart from data, and delete what we promised to delete.
You are a number
Your name and email are stored encrypted in a separate vault that only the study owner can open, and every opening is logged. All research data uses your participant ID.
Consent you can see
Each consent item is a separate tick box. The exact text you agreed to is kept, and team chat accounts are only created if you agreed to the chat-data item.
Leave anytime
Withdraw from the My Study page at any point, even after finishing. Your data can then be erased; chat messages are deleted or anonymised as the study promised.
Your copy
Download everything the study holds about you as a file from the My Study page. It never includes other people's data.
Nothing kept forever
Each study has a retention period. When it ends, identities are purged automatically; owners are warned two weeks ahead.
Locked down
Two-factor sign-in for researchers, roles (Owner, Researcher, Viewer), an append-only audit log, encrypted nightly backups and our own server; no third-party trackers.
Questions people ask
Is Stone Guest a keylogger? Should I be worried?
Technically yes: it records keystrokes, which is exactly what translation-process research needs. But it records only while you have a session running, shows a visible indicator, redacts password fields, never connects to the internet, and saves the file on your own computer. Nothing leaves your machine until you upload it yourself.
My antivirus or macOS warns me. Is that normal?
Yes. Any program that reads the keyboard needs special permission. On a Mac you grant Accessibility and Input Monitoring in System Settings; the program tells you exactly where. Stone Guest is open about what it does and is not code-signed, which is why some systems ask twice.
Do I need to be good with computers to take part?
No. Sign-up is a normal web form, the installer is a double-click, and the download page has an illustrated walk-through. If something goes wrong, the Help tab on your My Study page reaches the researchers directly.
Will my teammates know who I am?
No. In the team chat everyone has a fixed nickname, email addresses are hidden, and names cannot be changed. Researchers see your participant ID, not your name, in all study data.
Can I use DeepL, ChatGPT or a dictionary?
That depends on the study, which sets its own rule. Either way, the information sheet tells you before you start, and recordings note which tools were in front so the analysis can take them into account.
I'm a researcher. Can I run my study here?
Yes. Get in touch with a short description of your study. We set up the study and a panel account with two-factor sign-in for you, and your study gets its own sign-up link, download page and team spaces on translation-cognition.com. Everything study-specific is configuration, so no programming is needed.
Can we install the platform at our own institution?
Yes. The whole stack installs on a single server next to an existing Zulip with one documented procedure, upgrades in place with automatic rollback, and studies can be exported from one installation and imported into another. Contact us for the installation guide.
Which languages are supported?
Every participant-facing page, email and bot message is in Turkish and English; participants choose. Study materials can be in any language. The researcher panel is in English.
How should we cite it?
Cite the Stone Guest ToolKit (Kahyaoğlu) and the Task Segment Framework (Muñoz Martín & Apfelthaler, 2022) for any TSF-based analysis. Write to us for the current citation.
Run a study, join one, or just ask.
Whether you're a researcher planning a translation-process study, a participant with a question, or an institution interested in hosting the platform, write to Serhat Kahyaoğlu.
✉ serhat.kahyaoglu@yahoo.com