Rule-based machine translation for Aymara

This paper presents the ongoing result of an approach developed by the collaboration of a computational linguist with a field linguist that addresses one of the oft-overlooked keys to language maintenance: the development of modern language-learning tools. Although machine translation isn’t commonly thought of as a language learning tool, it can be a useful way for learners to better visualize the word-formation process and explore the structure of their own language, particularly in highly agglutinative languages like Aymara and Quechua. Moreover, the availability of translation software could eventually facilitate greater interlinguistic communication between speakers of minority languages and serve to further legitimize written production by such marginalized speakers. We provide an overview of how this software functions, describing the process by which the morphological analyzer (tagger) provides an input for the syntactic analysis (parser). We also give an overview of further possible applications of this work and show how this approach can be easily tweaked to account for different varieties within a given language, thereby preserving intervariant and dialectical differences along with the richness of variation. Moreover, we show how this model can be replicated for other languages and highlight the possibility of collaborative development with native speakers.