Modern view can show a host’s screens in another language. The title, field labels, messages, menu choices, column headings and what each function key does appear in the language your organization chose. What you type, and everything sent to the host, stays exactly as it was. The host never knows.
Inicio de sesión Translated to Español · Show original
Usuario . . . . . . WNDUSER
Contraseña . . . . .
- Hover over a translated text to see the host’s own words.
- Show original (or Ctrl+Shift+L, ⌘⇧L on a Mac) switches between the translation and the host’s text. WND Terminal remembers your choice.
- Screen readers are told which language each translated text is in, so they read it with the right voice.
What is translated, and what isn’t
WND Terminal translates the screen’s title, the labels next to fields and the hints after them, the host’s messages, other text on the screen, the labels of "label: value" pairs, what each option in a list’s legend does, menu choices, a list’s column headings and instructions, IBM i’s buttons and choices, and what each function key does.
Left as the host sends it:
- Input fields and what’s in them.
- The values next to labels (a name, a balance, a status) and the rows of a list.
- The status line: keyboard errors, "Row 1 of 600", "More...".
- The screen’s short name ("CADD", "WRKACTJOB") and the system name.
- Codes such as abends, message IDs and SQL codes (S0C7, ASRA, CPF9801, IEF450I, SQLCODE -803). They stay as they are inside a translated sentence too, so you can look them up.
- Values hidden by your organization’s privacy settings.
- TSO’s READY screen and other screens without fields.
Where translations come from
Your organization’s glossary comes first. It’s a CSV file of screen text and its translation, read on your computer. It always wins over the AI model.
Sign On,Inicio de sesión
User,Usuario
Work with,Trabajar con,prefix
A line matches a screen text exactly, ignoring capitals, extra blanks and a closing colon or period. With prefix in the third column, it matches texts that start with it. "Work with" turns "Work with Active Jobs" into "Trabajar con Active Jobs", unless "Work with Active Jobs" has a line of its own. The longest match wins.
One file can hold several languages. A first line starting with source names the columns, and each person gets the column for their language (es serves es-MX too):
source,es,fr,match
User,Usuario,Utilisateur,
Work with,Trabajar con,Gérer,prefix
Comma, semicolon and tab all work, as does a file saved from Excel.
The AI model translates the rest, when your organization allows it. It’s the same model as Explain this screen and Ask WND. Before anything is sent:
- Card numbers, Social Security numbers, e-mail addresses, long account numbers and passwords in commands are taken out, as for Explain. So is anything matching your organization’s
MaskPatterns, and anything already hidden on the screen. - Codes are taken out too.
Each one becomes a marker such as ⟦1⟧. The model sees "Card ⟦1⟧ refused, see ⟦2⟧" and the real values go back in on your screen only. A translation that loses a marker or adds one isn’t used, and that text stays in the host’s words.
The whole screen goes in one request. The answer is kept in your data folder (translations.json), under a fingerprint of the screen’s text, so a screen is translated once. The same screen with other data in it has the same fingerprint. The file holds only the text with the markers in it, never the values.
The glossary’s translations appear as soon as the screen does. The AI model’s follow a moment later, the first time a screen is seen.
Settings
On a small screen, swipe the table to read all columns.
| Setting | What it does |
|---|---|
| TranslateTo | The language, as a tag: es, fr, de, pt-BR, ja, zh-TW and so on. Empty: screens show as the host sends them. |
| TranslationGlossary | The glossary’s CSV file, for example \\fileserver\WND$\glossary.csv. |
| AllowAITranslation | The AI model translates what the glossary doesn’t. Off unless turned on. AllowAI = false turns it off too. |
| AIEndpoint, AIModel | The model server and model, as for Explain this screen. |
| MaskPatterns | Values that never reach the model, as for Explain and the help desk. |
With AllowAITranslation off, only the glossary is used and nothing leaves the computer. When the glossary can’t be read, the reason shows on the header’s tooltip and screens are translated without it.
Each request to the model is in the audit log as ai.translate: the model, the language, how many texts and characters were sent and how many values were masked. Never the text itself.
In a browser, WND Terminal uses the gateway’s policy and the organization’s model.