Portuguese to English

What Machine Translation Gets Wrong Between Portuguese and English (and How to Catch It)

Last reviewed on May 7, 2026.

Automated Portuguese-to-English translation has improved dramatically in the last decade. For most everyday text it produces output that's grammatical, fluent, and broadly correct. But for a translator that's running translations all day, the question isn't "is it good?" — it's "where does it slip up?" That question matters because the failures aren't random: they cluster around specific linguistic situations where the source is ambiguous, the systems lack context, or the two languages encode meaning differently.

This page maps the patterns. None of it is meant to undersell machine translation; it's the inventory of places to slow down, double-check, or rewrite the source. The translator on this site is reliable for the bulk of your text — the value is in knowing which sentences to read twice.

The big six error categories

From real PT–EN output, recurring slips fall into six clusters. The rest of the page works through each with examples.

  1. Ambiguous gender and pronoun resolution.
  2. The "ser" vs. "estar" distinction collapsing into English "be".
  3. Verb tense and aspect mismatches (especially present perfect).
  4. Idioms and fixed expressions taken literally.
  5. Variant-specific vocabulary (Brazilian vs. European Portuguese).
  6. Register and tone — formal/informal pronouns flattened to "you".

1. Ambiguous gender and pronoun resolution

Portuguese marks gender on adjectives, articles, and many noun endings. English largely doesn't. When the source uses a gendered word, the English may flatten it. The reverse direction is worse: English "the doctor" or "they" is gender-ambiguous, but a literal Portuguese rendering forces a guess.

Example (PT → EN):

  • "A médica disse que ela vem amanhã." → "The doctor said she's coming tomorrow." Here the Portuguese is unambiguous: a médica is female. A literal pass that translates "ela" as a generic "she" is fine; a careless model may switch to "he" if it loses track.

Example (EN → PT):

  • "The lawyer reviewed her contract." → A model has to pick "advogada" or "advogado" for the lawyer, and decide whether "her contract" refers back to the lawyer or to a different person. Bad guesses are common when the noun is far from the pronoun.

Catch it: if the source uses gendered nouns, scan the English for any third-person pronoun and confirm it points to the right person. If you're going EN → PT and the subject is a profession, double-check the gender ending.

2. "Ser" vs. "estar" collapsing into "be"

Portuguese has two verbs for "to be": ser for inherent or permanent qualities, estar for states or temporary situations. English has one. Going PT → EN, no information is lost; going EN → PT, the model has to guess.

  • "He is sick." → Ele está doente. (estar — temporary state). Wrong choice "ele é doente" means "he is a sickly person, chronically ill".
  • "She is tall." → Ela é alta. (ser — inherent). Wrong choice "ela está alta" sounds like "she's grown tall lately".

Most modern systems handle the obvious cases. Errors appear when the English is genuinely ambiguous — "the door is open" can be permanent ("an always-open door") or temporary ("the door has been opened"). PT requires you to choose.

Catch it: when translating EN → PT, if you see "is" or "was" before an adjective, ask: is this how the subject usually is, or how it is right now? Pick ser for the first, estar for the second.

3. Verb tense and aspect mismatches

Portuguese and English both have a present perfect, but they use it for different things. The Portuguese tenho feito roughly means "I have been doing recently / repeatedly", while the English "I have done" usually marks a completed action with present relevance. A literal translation in either direction can mislead.

  • "Tenho lido muito ultimamente." → Best: "I've been reading a lot lately." Literal "I have read a lot lately" is tolerable but loses the repeated/ongoing sense.
  • "I have lived here since 2019." → "Moro aqui desde 2019." Note the simple present in Portuguese, where English uses present perfect.

The simple past also misaligns. English uses "did" for completed actions at a specific past time; Portuguese has both pretérito perfeito ("fiz") and pretérito imperfeito ("fazia"). Pick the wrong one and the meaning shifts: "eu fazia exercício" = "I used to exercise" (habitual), not "I exercised" (specific past event).

Catch it: if the source mentions a specific past time ("yesterday", "in 2019"), expect simple past in both languages. If the source describes a habit or a backdrop, expect imperfeito in PT and "used to" or "would" in EN.

4. Idioms taken literally

Idioms are the classic failure mode and the easiest to spot. A literal pass on a fixed expression produces something either nonsensical or comically wrong.

  • "Pagar o pato" — literally "pay the duck"; means "take the blame for someone else's mistake".
  • "Cair a ficha" — literally "the token falls"; means "the penny drops" / "to finally understand".
  • "Engolir sapos" — literally "to swallow frogs"; means "to put up with annoying things without complaining".
  • "Chutar o balde" — literally "to kick the bucket"; in Portuguese it means "to give up / abandon a goal", not "to die".

Going the other direction:

  • "It's raining cats and dogs" — in Portuguese, "está chovendo canivetes" (BR) or "está a chover a cântaros" (PT). A literal "chuvendo gatos e cachorros" is not idiomatic.
  • "To be on the same page" → "estar na mesma página" works in modern Brazilian Portuguese (a calque), but a more native rendering is "estar de acordo" or "estarmos sintonizados".

Catch it: if a translation reads as bizarre, hunt for an idiom in the source. Conversely, if you wrote an English idiom, expect the literal Portuguese rendering to look odd.

5. Variant-specific vocabulary

A model trained mostly on Brazilian Portuguese may render a European text in Brazilian-flavored English (or vice versa). The English itself will be correct, but words from a domain may shift in unexpected ways:

  • propina in PT-PT means "tuition fee"; in PT-BR it means "bribe". A model that defaults to one variant may mistranslate the other.
  • banheiro in PT-BR is "bathroom"; in PT-PT it can mean "lifeguard". A literal output of "bathroom" for a beach safety notice would be wrong.
  • café da manhã (PT-BR breakfast) vs. pequeno almoço (PT-PT breakfast) — an unfamiliar form may be retained in the English ("small lunch"?) by a less-trained model.

Catch it: identify the variant first. The Brazilian vs European guide has tells you can spot in two seconds. If the variant differs from the model's default, slow down and read the output critically.

6. Register and tone flattened to "you"

Portuguese has tu/você/o senhor for politeness levels; English has one "you". Translating PT → EN, the politeness layer disappears. Going EN → PT, the model has to guess whether to use tu, você, or formal.

  • "Could you send the report?" → BR likely "Você pode enviar o relatório?"; PT likely "Pode enviar o relatório?" or "O senhor / a senhora pode enviar..." in a strict formal context.
  • "Hey, send the report." → "Manda o relatório aí" (very informal BR). The model needs to know the context to land here.

Catch it: when output goes into a customer email, a contract, or a publication, check the register manually. Even a perfect grammatical translation may sound stiff or too casual for the audience.

Less-frequent but high-impact errors

  • False friends. A small subset of cognates means different things; a model can pick the wrong sense. See the false friends reference.
  • Numerical formatting. Portuguese uses "," for decimals and "." for thousands ("1.234,56"); English reverses it. Translation tools usually handle text but may leave numbers untouched.
  • Date order. "11/03/2025" is March 11 in American English and 11 March in Portuguese (and British English). A bare number string may be read wrongly.
  • Acronyms and proper nouns. "DETRAN" (Brazilian DMV equivalent) may be translated literally or left untouched; either may need adjustment depending on audience.
  • Long sentences with embedded clauses. Portuguese tolerates long sentences with multiple clauses linked by commas; English prefers shorter ones. Literal output can read as a wall of text.

A worked review

Source (PT-BR): "O Marcos disse que ele tem trabalhado muito, mas que pretende tirar férias em janeiro. Antes disso, ele precisa terminar o projeto e depois discutir com o chefe."

A literal translation might give: "Marcos said he has been working a lot, but that he pretends to take vacation in January. Before that, he needs to finish the project and then discuss with the boss."

Two errors:

  • "pretende" — false friend. Should be "plans to take a vacation" or "intends to take a vacation", not "pretends to".
  • "discutir" — in business English, "discuss" works, but if the original implies friction, "argue" or "have a difficult conversation" might be closer to the intent.

Better: "Marcos said he's been working a lot but plans to take a vacation in January. Before that, he needs to finish the project and then talk it over with his boss."

Most production tools today get this almost right. The "pretends" mistake still surfaces in edge cases. The check is fast: read the output once with the source visible and ask whether each verb sense is correct.

A practical review checklist

  • Identify the variant. A few seconds to confirm PT-BR or PT-PT pays off later.
  • Scan for false friends. "atual", "pretender", "realizar", "assistir", "discutir", "sensível" deserve a glance every time.
  • Check tense for habit vs. specific event. Imperfeito vs pretérito perfeito; present perfect vs simple past.
  • Re-read any sentence with a third-person pronoun. Does it still point at the right person?
  • Spot the idioms. If a phrase reads as oddly literal, an idiom is hiding in the source.
  • Match the register to the audience. Formal client mail and a slack-style note can't share the same English.
  • Verify numbers and dates. Especially in financial or scheduling text.

When to stop trusting the machine

Some kinds of text remain best handled by a human translator regardless of how good the model is:

  • Legal and contractual text. A single mistranslated word can change rights and obligations.
  • Medical instructions. Dosage, frequency, and adverse-effect language must be exact.
  • Marketing slogans and brand voice. Tone and cultural feel matter more than literal accuracy.
  • Literature and poetry. Rhythm, ambiguity, and cultural reference rarely survive automated translation.
  • Anything with regulatory or audit consequences. Financial filings, tax language, immigration documents.

For everyday use — understanding an article, drafting an email, looking up a phrase, helping a friend — the translator on this site, used with the checks above, is more than enough. The point of this page is to make the reading-twice habit a default. For broader background on how Portuguese and English diverge, see Learn English for Portuguese speakers and the false friends reference.