PDF-to-HTML Layout-Preserving Machine Translation for Multilingual Open Education Resources: Addressing Character Density Expansion Across Languages

Abstract

Open Educational Resources (OER) are widely promoted as a mechanism for democratizing knowledge, yet, global OER ecosystems are heavily skewed toward English language content. This imbalance has been formally recognized through the 2012 Paris OER Declaration, the 2017 Ljubljana OER Action Plan, and the 2019 UNESCO Recommendation on OER, each calling on governments to prioritize content in local languages. A core technical barrier sustaining this inequity is the translation of Portable Document Format (PDF) documents across language pairs with substantially different character densities, such as Japanese to English or English to Portuguese. Existing tools either discard visual structure or produce fixedlayout outputs where translated text overflows its original bounding boxes, rendering documents unusable without manual correction. This paper presents a PDF to HTML translation pipeline that converts each PDF into an HTML representation, enabling adaptive vertical reflow that preserves visual fidelity while accommodating the lexical expansion inherent in crosslanguage translation. The system uses coordinate-aware text extraction, background rasterization, a configurable translation backend, and a layout adaptation procedure comprising horizontal slot redistribution and vertical cascade adjustment. The result is a self-contained HTML output that closely approximates the original document's visual appearance while allowing translated text to expand beyond the fixed constraints of the source PDF. The design is motivated by the documented inequity of English-dominant OER ecosystems and positions open-source and locally deployable AI tools as a practical instrument for multilingual educational access in lowconnectivity and low-resource environments.