Free Online Translator — 100+ Languages_

Over 100 languages — from widely spoken ones like English, Spanish, French, German, Italian, Portuguese, Dutch, Russian, Mandarin Chinese, Japanese, Korean, Arabic, Hindi and Turkish, to smaller and regional ones. Pick source and target from the searchable dropdowns, or use the quick-language buttons. Leave the source on Auto-detect and it identifies the input language for you, and ⇄ swap (or Alt+S) reverses the direction in one click.

It is a lighter, ad-free front end for quick text translation — no account, no cookie banner, no upsell, just a clean two-pane screen with auto-detect, text-to-speech, romanization, history and shareable links. If you need to translate whole documents or images, or want the very widest coverage, Google Translate or DeepL will serve you better. For fast copy-and-paste text, this stays out of your way.

Very accurate for everyday text — emails, messages, social posts, product descriptions and general communication, especially between major European languages. Idioms, slang, wordplay and highly technical, legal or medical content still need a human check, so review anything where a mistranslation would matter. Because results appear as you type, you can rephrase the source and watch how the translation changes, which is often the fastest way to get a cleaner result.

Yes for most everyday needs — emails, website copy, marketing text, support replies and internal notes. For legally binding contracts, regulatory filings, or medical and safety-critical documents where an error carries real liability, have a professional translator sign off. And note the privacy point below: your text is sent to a translation service, so do not paste confidential material.

No — this handles text you type or paste, up to about 5,000 characters at a time. It does not read files or images. Copy the text out of the document, email or chat and paste it in. If you need to get text out of a PDF first, the PDF toolkit on this site can extract it.

Yes. Copy (or Ctrl+D) grabs the result, Export downloads it as a .txt file, and Share link builds a URL that reopens the tool with your text and language pair already filled in — useful for sending a translation to someone else. Recent translations stay in your history so you can go back to them.

Be clear-eyed about this one: your source text is sent to an online translation service to be translated, so it is not private in the way a fully local tool would be. Do not paste passwords, personal data, medical details or confidential business material. Everything else — your preferences, your history, the interface — stays in your browser, and there is no account, login or tracking. It needs a connection, because the translation itself happens remotely.

Very. Translate a phrase in both directions to see how meaning and word order shift, press 🔊 Listen to hear it spoken, and turn on romanization to read non-Latin scripts in the Latin alphabet. Your history doubles as a vocabulary list of everything you looked up.

Yes — completely free, no ads, no account, and no limit beyond the per-translation character count. It is one of the free browser tools at jasperbernaers.com.

Requests go to Google's public translation endpoint, with MyMemory as an automatic fallback if that fails or rate-limits. You get the same underlying quality as the big free engines, without the interface built around it. Nothing here is a home-grown model — this page is the front end, not the translator.

Almost always a rate limit. The free endpoints throttle heavy use from one address, and the page falls back to the second engine when that happens. Wait half a minute and try again, translate in smaller chunks, or reload. A failed translation is never silent — you will see an error rather than a wrong answer.

Because machine-translation models are updated continuously, and the fallback engine can produce different wording from the primary one. Neither is wrong; they are different systems trained on different data. If you need a translation to stay fixed — for published copy, say — save the output rather than relying on regenerating it later.

About 5,000 characters per translation, which is roughly 800 words or two pages of prose. Longer text should be split into chunks. Splitting at paragraph boundaries rather than mid-sentence gives noticeably better results, because the engine uses surrounding words for context.

Yes — translation is debounced, so it fires shortly after you stop typing rather than on every keystroke. That is what makes rephrasing so useful: change one word in the source and the result updates, so you can feel out which phrasing translates cleanly.

The engine infers the language from the text itself and shows what it decided. It is reliable on a full sentence and much less so on a single word — gift is English for a present and German for poison, and there is no way to tell which from one word. Short input, heavily mixed languages, or text that is mostly names and numbers are the cases where you should set the source language manually.

Romanization writes a non-Latin script in the Latin alphabet so you can pronounce it — Chinese 你好 becomes nǐ hǎo, Japanese ありがとう becomes arigatō, Russian спасибо becomes spasibo. It appears automatically when the target uses a non-Latin script. It is a pronunciation aid, not a second translation, and different romanization systems exist for the same language.

Yes — 🔊 Listen uses your device's built-in speech synthesis, so the available voices come from your operating system rather than from this page. Quality varies a lot by language and platform: major languages sound natural on most modern systems, while less common ones may have no installed voice at all, in which case nothing plays. Installing additional language voices in your OS settings fixes it.

In your browser's local storage on this device only. It is not synced, not uploaded and not visible to anyone else, and clearing your browser data removes it. Use Clear history to wipe it deliberately — worth doing on a shared or public computer, since the entries are readable by anyone using that browser profile.

Your source text and the two language codes, encoded into the URL itself. That means the text is visible in the link — so treat a share link exactly like the text it carries, and do not send one containing anything sensitive. Anyone opening it gets the tool preloaded with your text, and the translation runs on their device.

Alt+S swaps the source and target languages, and Ctrl+D copies the result. The language dropdowns are searchable, so typing the first letters of a language name is faster than scrolling a list of a hundred.

Feed it complete sentences with normal punctuation — machine translation leans heavily on grammatical context, and a fragment gives it nothing to work with. Avoid idioms if the meaning matters, spell out ambiguous pronouns, and split long paragraphs. If a result reads oddly, rewrite the source more plainly and translate again; you will usually fix it in one pass.

Usually not — modern engines translate directly between most pairs, and routing through English adds a second chance to lose nuance. The exception is genuinely rare pairs, where the engine may already be pivoting through English internally. If a direct result looks wrong, translating each direction separately and comparing is a reasonable diagnostic.

Back-translate it: paste the result into the source box and translate it back. If it returns roughly your original meaning, the translation is probably sound; if it comes back garbled, something was lost. It is not proof — a wrong translation can back-translate cleanly — but it catches the majority of serious errors.

Because English does not mark the distinction, so the engine has to guess. German du and Sie, French tu and vous, Spanish tú and usted all carry a politeness level that English simply lacks. If it matters, check the output and adjust manually — this is one of the most common ways machine translation goes subtly wrong in professional email.

Yes — Arabic, Hebrew, Persian and Urdu translate normally and display in the correct direction. Romanization is offered for them too. The one thing to watch is copying RTL text into a system that does not handle bidirectional text well, where punctuation can appear to jump to the wrong end of the line.

Emoji pass through untouched. Code and markup are riskier: the engine may translate words inside tags, attributes or string literals, which quietly breaks them. If you need to translate a UI string table or an HTML page, translate only the human-readable text and leave the surrounding syntax alone.

For understanding something, generally yes. For publishing something, generally no. The gap is not vocabulary — it is register, cultural context, brand voice and the ability to notice that a literal translation, while correct, would land badly. A common professional workflow is machine translation followed by a human edit, which is faster than translating from scratch and much better than raw output.

Increasingly little. Today's machine translation is neural — the same broad family of models behind large language models — so the old distinction between statistical translation and AI has mostly dissolved. General-purpose AI chat tools can do more with an instruction ("translate this formally, for a legal audience") but a dedicated translation engine is faster and more consistent for straight text.

Broadly, those with the least training data. Major European and East Asian languages are excellent because decades of parallel text exist. Low-resource languages — many African, Central Asian and indigenous languages — are noticeably weaker, and languages with rich morphology or very free word order are harder still. Treat output in a low-resource language as a rough draft.

That depends on the upstream provider's policy, not on this page, and it is a fair question to ask of any free translation service. MyMemory in particular is explicitly a translation memory — it is built on stored translations. This is the practical reason for the advice above: keep confidential material out of any free online translator, including this one.

For your own emails, listings, descriptions and marketing copy, yes — nobody claims copyright on machine output of your input. Two caveats worth knowing: you have no rights to the source text if it was not yours to begin with, and for anything customer-facing you should have a fluent speaker read it before it goes out.

Yes. The two panes stack on small screens, the language pickers are searchable rather than long scrolling lists, and Listen uses the phone's own speech engine. It works fine as a travel translator provided you have a connection — the translation itself happens online, so it will not help you offline abroad.

Because translation quality depends on models far too large to ship to a browser — the ones behind these results are gigabytes in size and run on servers. Offline translation apps exist and download compact per-language models, but the quality trade-off is real. If you need offline translation while travelling, download language packs in a dedicated mobile app before you go.