Speaker
Description
Lexicographic practice has often balanced comprehensive coverage with systematic ordering, privileging alphabetical arrangement and encyclopedic completeness over clearly defined user needs. This paper argues that user-oriented specialized dictionaries, supported by corpus-driven and AI-augmented methods, can better serve users who require targeted lexical knowledge rather than exhaustive inventories. The study uses Śabdakalpadruma as a Sanskrit lexicographic case study and reports its conversion into a structured digital corpus of 42,824 headword entries through OCR, post-OCR cleaning, Excel inspection, SQLite structuring, JSON representation, and searchable HTML implementation. Building on this corpus, the paper develops a pilot rule-based recommender model for ritual vocabulary. Twenty ritual query terms were searched in headword and definition fields, and the extracted entries were automatically annotated through semantic tagging, co-occurrence detection, relation scoring, confidence labelling, and relevance classification. A stratified validation sample of 100 entries, with five entries from each query term, was manually evaluated. The results show that 67% of annotations were correct, 19% usable with correction, and 14% wrong. The study demonstrates a staged pathway from digitized lexicographic corpus to rule-based recommendation, human-validated training data, and future AI-assisted semantic prediction.