OneStopEnglish Corpus

OneStopEnglish is a parallel readability corpus built by Sowmya Vajjala and Ivana Lucic at Iowa State University, released with their 2018 paper at the BEA workshop on NLP for educational applications. It draws on material from onestopenglish.com, an English language teaching resource where teachers rewrote Guardian articles for learners. The corpus holds 189 source articles, each provided at three levels, Elementary, Intermediate and Advanced, for 567 texts in total. Because the three versions of each article carry the same meaning at different complexity, the levels are directly comparable, which is the whole point of the collection.

You access it as a public GitHub repository, cloned or downloaded as plain text files organised by level plus CSV files, one per article with the three levels side by side, and sentence-aligned versions produced by cosine similarity matching. There is no API, account or application wall, so the access tier is open. It is small, well under a gigabyte, and lightly processed rather than raw scrape.

For RAG it is a compact, clean set of level-labelled passages useful for readability grading and simplification retrieval. For an AI tutor it is more pointed: the aligned triples let a tutor pitch the same content at a learner's level, generate or check simplifications, and model reading difficulty for level-appropriate delivery, all things a generic corpus cannot support.

The licence is Creative Commons Attribution-ShareAlike 4.0, so commercial use is permitted, but attribution is required and any derivative corpus must carry the same share-alike terms, which matters if you redistribute processed versions inside a product. Note the underlying articles originate from the Guardian via onestopenglish.com, so treat provenance carefully at scale.

It sits in our education set alongside broad curricula such as openstax, khan-academy, ck-12, siyavula, common-core and ngss, but is narrower and language-focused, complementing rather than overlapping the maths sources amps, megamath, stackmathqa, naturalproofs and openthoughts3.