dsh-bilingual-reader

Reading a dense English paper in the DSH sidebar, you hit a term or a convoluted sentence and the instinct is to select, copy, and ask the LLM for a translation—but that drags reading fragments into your main conversation, polluting context and tainting every follow-up question. dsh-bilingual-reader exists for exactly this moment: it delivers inline, selection-triggered translation right inside the sidebar, each call a one-shot, isolated LLM invocation that never writes back to your session, so your reading aid and your working conversation stay cleanly separated.

The core experience is built around accuracy and speed. PDFs render through the browser-native engine with perfect layout, paging, and text selection. On desktop (Electron), simply select and copy and translation fires automatically; on web, one click on the button does it. Source and target languages include Chinese, English, Japanese, and Korean out of the box, with a free-form add-language input for anything else. Setting the source to Auto-Recognise first runs a zero-cost character-set heuristic, falling back to a single LLM guess when the heuristics are ambiguous. The plugin locates the unique occurrence of your selected text in the full document and pulls surrounding sentences as context; if the phrase appears multiple times or cannot be matched, it deliberately drops context to avoid misleading the model. Both original and translation appear side by side below, with one-click copy on the translation.

All preferences—languages, context window size, custom language entries—are persisted across sessions. This plugin is ideal for researchers and engineers who read English papers, technical docs, or foreign-language literature heavily inside DSH and want a lightweight, zero-pollution reading aid that gets out of the way the moment the translation is done.