Guide
Aligning a text with its translation, sentence by sentence
A text and its translation side by side, sentence by sentence, is useful to many people. Language learners read a story in a new language with the original as a guide. Translators turn their past work into a translation memory so their tools can suggest matches. Researchers build parallel corpora, and editors check that nothing was left out. Aligning by hand takes hours; this tool does it in a moment and lets you correct the result.
Splitting into sentences
Both texts are split into sentences first. A sentence ends with a full stop, question mark, exclamation mark or ellipsis, including Arabic and Devanagari marks, when the next word starts with a capital, a digit or a quotation mark. Common abbreviations such as Dr., Mr., e.g. and single initials do not end a sentence. In Chinese and Japanese, the full-width marks 。！？ end sentences. Paragraph breaks are always boundaries.
How the alignment works
The method, published by William Gale and Kenneth Church in 1993, rests on a simple observation: long sentences tend to translate into long sentences and short ones into short. It compares the lengths of sentences on both sides, measured in letters and digits, against the overall length ratio of your two texts, and finds the most likely pairing. Most pairs are one sentence to one, but it also allows one sentence to become two, two to become one, and sentences with no counterpart. No dictionary is needed, so it works for any pair of languages.
Checking and correcting
The aligned rows appear side by side. Rows where the lengths fit badly — the least certain matches — are highlighted. If a row is wrong, use the small buttons in each cell: split moves the last sentence of a cell into its own row, and merge pulls the next sentence up into the current row. Undo returns to the automatic alignment. Turn on match scores to see the cost of each row; lower is more confident.
Exporting
TMX (Translation Memory eXchange) is the standard file for translation memories, read by OmegaT, Trados, memoQ, Wordfast and others; set the language codes first. CSV opens in Excel with the two languages in two columns, and TSV pastes into any spreadsheet.
A worked example
The sample is a short English story about a lighthouse keeper and its Spanish translation. Both have five sentences and align one to one, so “Ships passed safely, and nobody ever thanked him.” sits next to “Los barcos pasaban sin peligro y nadie le daba nunca las gracias.” The TMX starts with the pair beginning “El viejo faro se alzaba al borde del acantilado.”
Getting the best results
Paste plain text with the same paragraph breaks on both sides, since paragraphs are hard boundaries that keep errors from spreading. Remove headers, footnotes and page numbers that appear in only one version.
Using aligned texts for language learning
Read the new language first and glance across only when you are stuck, rather than reading both columns line by line. Exported to CSV, the pairs make ready-made flashcards for apps such as Anki: one column on the front, the other on the back.
Related tools
Compare two versions of a text with the Diff Checker, count words in each language with the Word Counter, listen to the translation with Text to Speech, or dictate a translation with Speech to Text.
Checked against the current version of the Parallel Text Aligner (Bilingual) on .