Sense-Level Vocabulary State and In-Reading Inference for Assisted Reading: Methods and System

This technical disclosure describes methods and a system for assisted reading of foreign-language text in which a reader's vocabulary knowledge is recorded per word sense rather than per word form. During offline preprocessing, every content-word occurrence in a book is mapped to an ID in a global, incrementally built sense inventory (dictionary defaults for unambiguous words; a language model for the rest, which also reviews the defaults). While reading, the reader clicks the words they do not know in a paragraph and confirms the paragraph; the senses of clicked tokens are recorded as unknown and the senses of all other content tokens as known. Annotations are then shown automatically wherever a sense the reader marked unknown occurs, without calling a model at reading time. The disclosure also covers rare-sense flags, sense-ID merging with aliases, intensive and extensive reading modes, cross-device synchronization with drafts, reading position, difficulty-weighted reading volume, vocabulary-size and suitability estimates, persona-specific commentary layers, annotation provenance, text-free annotation layers anchored by normalized-text hashes and token indices, and many variants. It is published as prior art; no patent rights are claimed. English and Chinese versions are attached. 本文公开一种外语文本的阅读辅助方法与系统:读者的词汇掌握状态按义项而不是按词形记录。离线预处理时,书中每个实词的每次出现都映射到一个全局、增量构建的义项表中的义项 ID(无歧义的词用词典默认值,其余由大语言模型判定并复核默认值)。阅读时,读者把一段里不认识的词点出来并确认本段读完,点了的词所指的义项记为不认识,其余实词所指的义项记为认识;此后凡读者标为不认识的义项出现,自动显示注释,阅读时不调用模型。本文还公开了熟词僻义标志、义项 ID 合并与别名、精读 / 泛读两种模式、带草稿的跨设备同步、阅读位置、按难度折算的阅读量、词汇量与适合度估计、按读者画像的讲解层、标注来源记录、按规范化文本哈希与 token 序号锚定的不含原文的外挂层,以及多种变体。本文作为现有技术公开,不主张任何专利权。附中英两个版本。