Low-Resource Languages

Un pipeline en Google Colab que corrige transcripciones ruidosas de Whisper en maya yucateco combinando reglas morfológicas, un léxico de 2,988 entradas y Gemini 2.0 Flash como modelo corrector.

A Google Colab pipeline that corrects noisy Whisper transcriptions of Yucatec Maya by combining morphological rules, a 2,988-entry lexicon, and Gemini 2.0 Flash as a corrector model.

An openly licensed parallel corpus of 14,332 aligned phrase pairs in Yucatec Maya and Spanish, organized into 31 everyday communicative contexts — the first open, phrase-aligned corpus of its kind.

An openly licensed parallel corpus of 14,332 aligned phrase pairs in Yucatec Maya and Spanish, organized into 31 everyday communicative contexts — the first open, phrase-aligned corpus of its kind.

Un benchmark de similitud de palabras cultural y lingüísticamente adaptado al maya yucateco, construido a partir de una Lista de Swadesh filtrada, que muestra que la elección del benchmark pesa más que el ajuste de hiperparámetros al evaluar word embeddings.

A culturally and linguistically adapted word similarity benchmark for Yucatec Maya, built from a filtered Swadesh List, showing that benchmark selection matters more than hyperparameter tuning when evaluating word embeddings.

Un algoritmo fonético tipo Soundex diseñado para la lengua maya, que genera códigos consistentes para palabras de sonido similar y permite una recuperación de información robusta frente a la variación ortográfica.

A Soundex-style phonetic algorithm designed for the Maya language, generating consistent codes for similar-sounding words to enable robust information retrieval despite orthographic variation.