Translation

Machine translation via a free API (internet required). Free, no upload. 1 GB max · Up to 1 GB · Processed locally, never sent to a server.

This page translates text between thirty languages, from English to Arabic, Japanese or Mauritian Creole. Before anything else, you need to know where your text goes: it leaves your browser. On every run, the contents of the input area are sent to api.mymemory.translated.net, the public endpoint of MyMemory, the translation memory operated by the company Translated. This is the one tool in the category that talks to a third party.

The mechanism is easy to describe. Text is refused beyond 4,500 characters. Below that limit it is cut into pieces of roughly 480 characters at sentence boundaries, to stay within what a single request accepts. Each piece leaves in a separate request with a fifteen-second timeout, and the button reports progress in the form "Translating 2/3". The translations that come back are then glued end to end, separated by a single space.

MyMemory is not a neural engine like the others. It is first and foremost a memory, meaning a database of segments that have already been translated, fed by public corpora and by contributions from its users, and topped up with machine translation when no stored segment matches. That architecture explains most of what you will observe in practice: excellent output on common sentences it has met before, and much less certain results as soon as your text is unusual.

How to use it

  1. Set the two languages The From and To lists offer the same thirty entries, and open on French as the source and English as the target, so English users normally change both. Choosing the same language twice stops the process before any request is sent.
  2. Paste at most 4,500 characters Beyond that a "Text too long" message appears and nothing is sent. No automatic truncation is applied, so splitting a longer document is your responsibility.
  3. Press Run The button is disabled for the duration and counts the pieces as they are processed. On a two-thousand-character text, expect four or five successive round trips with the service.
  4. Check the joins Each piece is translated with no knowledge of the others. The seams between two blocks are where a pronoun loses its antecedent, or where a technical term changes translation halfway through.
  5. Copy the result The copy button appears under the translated block. The source text stays on display above it, which makes a segment-by-segment check straightforward.

What travels, and in what form

The request is a GET, which means your text travels inside the address itself, encoded in the query parameter, alongside the language pair written in the form "en|fr". That detail has a concrete consequence. The body of a request is rarely retained by the equipment it passes through, whereas an address appears by construction in the destination server logs and, potentially, in those of network intermediaries along the way.

Nothing else is added to the transmission: no account identifier, no filename, no reference to the page you came from. As with any web request, however, your IP address and the usual browser headers reach the recipient. That server does not belong to NexavoTools and applies its own policy. What it logs, and for how long, is not something this site can speak to.

The rule that follows is simple enough. A contract clause under negotiation, a medical report, an extract from a client file or an embargoed announcement have no business in this field. Strip out names and amounts first, or use a service whose hosting you control.

The chunking, and what it costs you

Pieces are assembled sentence by sentence until they approach 480 characters. That approach carries a limit worth knowing about: a single sentence longer than 480 characters on its own is cut at that length, and the remainder is never sent. A paragraph written as one continuous block with no full stops will therefore lose its ending silently, with no warning and no error.

The second cost is contextual. Because each block is translated in isolation, the rendering of a technical term can drift from one block to the next, and a pronoun sitting at the head of a piece arrives with no referent behind it. Rereading the joining areas is generally enough to catch these breaks before they reach anyone else.

Quality varies sharply from one pair to another

A translation memory serves you in proportion to how well it has been fed. The major European pairs, English into French, Spanish, German or Italian, rest on very large corpora and often return usable output on standard sentences. The less-resourced languages in the list, Mauritian Creole or Thai among them, draw on far thinner material, and the fallback machine translation carries more of the load.

The collaborative sourcing of the database has a second consequence. A faulty segment submitted by a contributor is returned with exactly the same confidence as a professional translation, and nothing on screen distinguishes the two. No translation obtained here should be published without a reread by somebody who knows the target language, and a contractual or medical document is work for a professional translator, not for this page.

The three ways a translation fails

The first failure is local and immediate: identical source and target languages, or a text over the character limit. Nothing is sent, the message is explicit, and the correction is yours to make.

The second is the network. Past fifteen seconds with no answer for a given piece, or with no connection at all, the box states that the service is unreachable and reminds you that this tool, unlike the rest of the category, does not work offline.

The third comes from the service itself. When the response carries an error status the HTTP code is reported, and when MyMemory returns its own explanation that text is displayed exactly as received, with no rewording on our part. MyMemory writes those messages in English, so they arrive readable, but they are the service speaking, not this site. The daily quota applied to anonymous calls, calculated on your IP address, is the message you are most likely to see.

Frequently asked questions

Why does this page not work offline?

Because it contains no translation engine of its own. It prepares a request, sends it, and formats the answer. With no network there is simply nothing to display. It is the only tool in this category in that position.

How many translations can I run per day?

This site imposes no quota. MyMemory applies one to anonymous calls, and when you reach it, its own warning message appears in the error box unchanged, which is why the wording will not match the rest of the page.

The end of my sentence was not translated, why?

Because that sentence ran past 480 characters in one unbroken block. The chunking cuts it at that length and the remainder is not transmitted. Add strong punctuation, or split the sentence yourself, then run it again.

Why does the result differ from another online translator?

Every service has its own engine and its own database. MyMemory favors a segment that has already been translated whenever it finds a close match, where a purely neural engine recomputes the entire sentence from scratch each time.

Can I translate a document rather than pasted text?

No, the field accepts pasted text only. A PDF or a Word document has to be converted first, then cut into slices of under 4,500 characters, which also gives you a chance to remove anything confidential.

Is my text retained by the service?

This site has no control over that and cannot make a promise on another company behalf. The MyMemory database is fed by contributions, so assume that anything you send may be kept, and treat this field as a public space.

Similar tools

Discover other tools

Categories