Contextual Interpretation of Nasa Yuwe Language Expressions into Spanish Using Natural Language Processing Techniques

Abstract

The Nasa Páez Indigenous community, located mainly in the department of Cauca, Colombia, faces a cultural risk due to the progressive loss of its ancestral language, Nasa Yuwe, particularly among younger generations. As language constitutes the core of their worldview, oral tradition, and community life, this study draws on Natural Language Processing techniques to propose technological tools aimed at preserving Nasa Yuwe. While Western languages such as English, Spanish, and French benefit from numerous automatic translation solutions, implementing such technologies for Nasa Yuwe represents a significant challenge, as it is an ultra–low-resource language in digital terms when compared to those languages. To ensure coherence in the collection and systematization of Nasa Yuwe expressions, this research is delimited to the San Lorenzo de Caldono Indigenous Reserve, allowing the study to focus on a specific linguistic variant. To evaluate the developed interpreter, an intention confusion matrix was employed, yielding an efficiency of 100%, measured through the F1-score, and a precision of 100% in the recognition of cultural expressions, based on a total of 144 evaluated samples. Additionally, the evaluation of cultural entity extraction, conducted using the confusion matrix generated by the DIETClassifier model, showed an average (macro) precision of 91.1%, a recall of 95.3%, and an overall efficiency of 92.8% over 955 annotated entities, confirming the system’s effective performance in identifying semantic components inherent to the Nasa Yuwe language.