cognate-tables
nathanschaumann/cognate-tables
Cognate tables across English, Spanish, Portuguese, French, Russian and Armenian (111,711 rows, 15 pairs)
cognate-detectionmultilingual-datalanguage-learninglinguistic-datasetword-similaritytransliterationromanianizationlemmatization
Summary
A dataset of 111,711 cognate pairs across six languages (English, Spanish, Portuguese, French, Russian, Armenian), formatted as CSV and Markdown tables. Created for a language-learning application, it identifies words that are visually similar after romanization and share meaning. The dataset was built using a multi-step pipeline involving frequency lists, AI model annotation (Claude), mechanical filtering, and manual audits.