Translate & Summarize

A summary in the target language before the full translation. The point is to decide whether the document is worth reading in translation at all, which for a long contract or report is a saving worth having — and the summary is produced in the language you are going to read it in.

Ordering the Decision Before the Work

The expensive mistake with long foreign-language documents is translating them. A forty-page contract that turns out to be a standard renewal has cost an afternoon of a translator's time and produced nothing. Reading a summary first is the step that prevents it, and the summary is only useful if it arrives in the language the reader actually reads fluently.

• The workflow runs in two labelled steps and the first one is always the summary. The tool produces a condensed version of the document in the selected length and target language before any phrase substitution happens to the full text, so the decision to continue is made on the summary rather than on a partially processed whole.
• Summary length has three settings: a short version of about three lines, a medium version of about four, and a detailed version of about six. The difference between them is how many of the ranked sentences are kept, not how deeply the content is analysed.
• Sentence selection during summarising uses a stop-word list to discount the words that appear in every document, so the ranking is driven by vocabulary that is actually characteristic of this one.
• Six target languages are available — Chinese, Spanish, French, German, Portuguese and Japanese — and the full-document phrase pass runs against a table of around a hundred and sixteen conventional phrases grouped by language.
• Only after the summary has been read does the second step handle the body, substituting recognised phrases and reporting how many it dealt with.
• The whole document text stays in the page during both steps, and the tool reports the substitution count so the extent of the local processing is visible rather than assumed.

Using the Summary as a Gate

1. Select the target language you read, not the language of the document. A summary in a language you are translating into rather than reading in defeats the purpose of the step.
2. Start at the short summary length. If three lines tell you the document is a standard agreement, you have saved the whole exercise.
3. Choose the detailed length only when the short version leaves a real question open — the difference is more sentences from the same ranking, so it adds coverage rather than depth.
4. Decide explicitly whether to continue. The reason the summary comes first is to make that a decision you take rather than a step you skip past.
5. If you continue, run the body pass and read the substitution count against the document length. A hundred-page contract with a dozen substitutions has been almost untouched, and the honest conclusion is that the terminology was handled and the translation was not.
6. Keep the summary. Where a document is going to be discussed with colleagues, a short version in a shared language is often the artefact that gets used, regardless of whether the full translation happens.

A Contract Nobody Needed to Translate

A forty-page supplier agreement arrives in a language the procurement team does not read, and the question is whether it needs a translator.

• The short summary, three lines in the reader's own language, identifies the document as a renewal of an existing agreement with terms matching the previous one plus an updated price schedule. That is the decision: no full translation required, one clause to check.
• The medium summary adds roughly a line and surfaces the clause that changed, which is the part the team actually needed. Note that the additional sentence came from the same ranking rather than being new analysis — the longer setting shows more of what was already selected.
• Running the detailed setting would add two more lines of context and no new information about the change, which is the general shape of this trade-off: the marginal value of each extra line falls quickly.
• If the document had been an unfamiliar agreement instead, the summary would have established that too, and the decision would have been to proceed — at which point the phrase pass handles terminology and a translator or a connected provider does the body.
• In this case the phrase pass is never run, and the substitution count is irrelevant. That is the intended outcome: the tool's most valuable use is the one where it stops the work.

Four Ways to Approach a Long Foreign Document

Alternative	Notes
Summary in your language, then decide (this tool)	Costs seconds and frequently ends the process. The limitation is that a summary cannot tell you whether a clause is acceptable — only whether the document is the kind of thing you expected, which is the question that decides most of these cases.
Full machine translation, then read	Produces everything and costs the reading time for all of it. Worth it when the document is definitely needed in full and the reader can work from a rough version; wasteful when the answer was going to be that it is unremarkable.
A human translator	The right answer when the document has legal or commercial consequences and the wording of individual clauses matters. The summary step still applies — it is what a translator would ask you for before starting, so having it improves the brief.
Reading the original with a dictionary	Feasible for a single page and produces the most reliable understanding of any method here. Not feasible past a few pages, which is where the summary step takes over as the way to decide which few pages deserve this treatment.

What Runs Locally and What Does Not

The summary ranking, the stop-word discounting and the phrase substitution all happen in the page, using tables compiled into it. Nothing is transmitted on either step, and the document text never leaves the browser — a property that matters considerably here, since the documents people summarise before translating are disproportionately contracts, internal reports and correspondence that has not been published. The limit is the ceiling on quality rather than on privacy: sentence selection by word frequency produces a competent summary of conventional documents and a poor one of anything where the important content is carried by structure rather than by distinctive vocabulary.

• The summary is extractive, so every line in it comes from the document. Nothing is paraphrased or inferred.
• The phrase table is grouped by target language, and the pass runs longest phrase first to avoid breaking longer expressions.
• No key, no account and no connectivity are required for either step; the tool works with the network disconnected.
• Nothing persists between runs. If the summary is worth keeping, copy it before closing the tab, because there is no history.

Questions About Summarising Before Translating

Is the summary reliable enough to skip the document entirely?

For establishing what kind of document it is, usually yes. For establishing whether a specific clause is acceptable, no — the summary selects sentences, and a short version may omit the only sentence you needed. Use it to decide whether to engage with the document at all, and treat any conclusion about specific terms as requiring the passage itself.

Why is the summary in the target language rather than the source?

Because the person deciding is the person who cannot read the source. A summary in the original language would need translating before it could be used, which returns the problem to where it started. Producing it in the reader's language is the whole reason the step comes first.

How much does the summary length change the content?

It changes how many ranked sentences are shown, not what is looked for. The short version keeps about three lines, the medium about four and the detailed about six, all drawn from the same ordering. Moving from short to detailed therefore adds the next-most-prominent sentences rather than a deeper reading, and the marginal value of each line falls quickly.

Does the document get sent anywhere?

No, on either step. Ranking, summarising and phrase substitution all run in the page, so an unpublished contract or internal report is never transmitted and the tool works offline. Since the documents worth this treatment are disproportionately confidential, that constraint is what makes the summary gate usable at all rather than a step that requires uploading first.

Last verified: October 3, 2026

Launch Translate And Summarize

All Tools · AI Tools · About · How We Test · Contact · Privacy

Maintained by Huang Bin · Last updated October 3, 2026