While it is preferred to speak in English or use a human translator whenever possible, sometimes machine translation is necessary. 

When a client’s primary language is not English, MechaSqueak provides integrated translation capabilities that allow dispatchers to translate without resorting to external tools.

Establishing the Client’s Language

When a case comes in, the "case language" will be set to the preferred language setting of the client's web browser. If the language setting does not match what the client is speaking, you must change the language of the case for translation tools to work correctly.

!lang <language-code>

You can find a list of supported languages for translation here with their language code.

If the language remains at en, MechaSqueak will assume translation is unnecessary.

Receiving Live Translations of Client Messages

To view an English rendition of the client’s messages in real time, activate live subtitles:

!tsub notice   # translations delivered as IRC notices (recommended)

A translated message will appear in IRC under the original message like this:

<spacedawg> отправил запрос дружбы

-MechaSqueak[BOT]- <spacedawg> Sent a friend request

Notices appear in the active channel window but are visible only to the subscribing dispatcher. Should your IRC client handle notices poorly, private‑message delivery is available:

!tsub pm       # translations delivered via private message

The subscription persists until explicitly cancelled:

!tunsub        # terminate the subscription

Live translations for the client remain active even after the case has been formally closed until the case number is re-used by a new case. When a new case is assigned to the number, Mecha will 'forget' the old client.

Communicating with the Client via /tr

The /tr slash command allows translation to be provided by mecha while posting under the dispatcher’s own nickname, preserving the natural flow of conversation.

/tr <case‑id> <message>

Example :

/tr 4 Please drop out of supercruise

<dispatcher> SpaceDawg: Veuillez quitter la supercroisière.

 When the dispatch uses this command, Mecha will send the original english message to anyone else currently subscribing to translations via !tsub so rats and other dispatchers are aware of what is being said. Dispatch messages are not translated if they are typing in a foreign language themselves or using an external translation tool.

Addressing a Channel Directly

Sometimes you may need to send a translated message in the channel to someone without a case ID, the /tr command allows you to define a target channel and language.

/tr <channel> <language‑code> <message>

Example:

 /tr #fuelrats es Frame shift complete; slowing to sub‑light.

Prerequisite:  The IRC client must allow the transmission of raw text lines. HexChat, WeeChat and Polaris meet this requirement; certain web clients do not. Some clients like IRSSI may require you to change a setting or add your own alias to make it work.

Alternative (Legacy) Commands

Althrough /tr is recommended, there are also regular !commands that can be used in the channel or in a PM with MechaSqueak in situations where IRC server /commands cannot be used.

More information about these Mecha commands is available on the Mecha help page

Quoting Foreign‑Language Messages

Executing !grab on a message written in a foreign language automatically stores an English translation alongside the original text.

Best practices

• Use short, unambiguous sentences. Machine translation yields more accurate results when the source text is concise.
• Avoid language-specific or cultural idioms, these may not translate well and will confuse the client. 
• Avoid using roleplay terms, for example "your rescuers will be there soon" will translate better than "your rats will be there soon".

Q&A

What service do you use to perform machine translations?

• We send requests to OpenAI using their GPT 4o model these requests amount to about €0.50 a month and the cost is covered out of pocket by an administrator.

Are these translations reliable?

• All machine translation has the possibility of being incorrect, that is why we recommend avoiding it when possible and following the best practices when using it. 

Why did you choose to use GPT for translations?

• GPT is powered by a transformer based large language model which is a more advanced evolution than the simpler machine learning approaches used on older technologies like Google Translate and DEEPL, it has been shown to have a roughly 95% accuracy, though this accuracy will differ depending on the language.