Educational Tools for Mapuzugun

Cristian Ahumada, Claudio Gutierrez, Antonios AnastasopoulosNAACL 2022

TL;DR

This paper introduces a suite of NLP tools designed to support the revitalization and education of Mapuzugun, an endangered Indigenous language. The system includes an orthography detector/converter, a morphological analyzer, and an informal translator, all tailored to Mapuzugun's unique linguistic features and made accessible through a user-friendly interface.

What Problem Does This Paper Address?

Mapuzugun, the language of the Mapuche people, faces a significant decline in speakers due to historical and political factors, leading to its exclusion from the educational system in Chile and Argentina. While Natural Language Processing (NLP) has seen remarkable progress, this advancement has largely benefited a handful of dominant languages, leaving many endangered languages like Mapuzugun without adequate technological support for preservation and education. Existing resources are scarce, and the lack of standardized orthography further complicates computational processing. This paper aims to bridge this gap by developing practical NLP tools specifically for Mapuzugun, empowering its speakers and learners.

Methods

The system is built around a pipeline of interconnected components:

  • Orthography Detector and Transliteriterator: Mapuzugun has three main orthographies (Unificado, Ragileo, and Azümchefe). This module identifies the input orthography and can convert between them using pre-defined conversion tables. The detection process involves a round-trip translation to verify consistency.
  • Morphological Analyzer: This core component segments words into their constituent morphemes. It's implemented using hand-crafted regular expressions derived from established grammars and linguistic resources. The analyzer considers the order and compatibility of morphemes, crucial for Mapuzugun's agglutinative nature. It assumes input in the Ragileo orthography for simplicity due to its 1-to-1 phoneme-to-grapheme mapping.
  • Informal Analysis Translator: This module translates individual morphemes or combinations into plain Spanish definitions, avoiding technical jargon. The goal is to make the analysis accessible to learners, with definitions for nouns, verbs, and adjectives sourced from a Mapuzugun-Spanish dictionary.
  • User Interface: A web-based interface provides access to these tools, available in both Mapuzugun and Spanish.

Datasets

The paper does not explicitly list specific datasets used for training the morphological analyzer. However, it mentions leveraging "established grammars of Mapuzugun" and a "compilation... of the grammar of the language" as sources for the regular expressions. Additionally, a "Mapuzugun-Spanish dictionary" was used for the informal translator. No new benchmarks were introduced; instead, the focus was on creating functional tools for an under-resourced language.

Key Insights and Ideas

  • Tailoring NLP to Endangered Languages: The project demonstrates that NLP tools can be effectively developed for languages with limited resources by leveraging linguistic expertise and existing grammatical knowledge, rather than relying solely on large data-driven approaches.
  • Importance of Orthography Handling: The existence of multiple orthographies for Mapuzugun highlights a common challenge for many Indigenous languages and underscores the necessity of tools that can bridge these variations.
  • "Informal" Translation for Accessibility: The concept of an "informal analysis translator" that provides simple, jargon-free explanations of morphemes is a key innovation for educational purposes, making complex linguistic analysis understandable to beginners.
  • Bridging Linguistic Research and Community Needs: The work successfully transforms linguistic research (grammars) and minimal community resources (dictionaries) into practical tools for language preservation and revitalization.

Key Results

A usability study was conducted with initial learners of Mapuzugun.

  • Translation Accuracy: For two specific words, "pemurpayafuyu" and "kujinerkeeiñmu," using the tool significantly improved translation accuracy across all user groups. The phrase translation also improved for beginner and intermediate groups.
  • Usability Scores: User groups 2 and 3 (intermediate learners) showed the least difficulty and gained the most advantage from the tool.
  • Qualitative Feedback: Users generally responded positively to the tool's presentation of word segmentation and the informal translator. The graphical representation of word composition was particularly appreciated. Experienced speakers preferred not to use the tool, feeling confident in their existing knowledge.

Limitations and Future Work

  • Limitations:
    • The current system does not support full-fledged machine translation, limiting users to word and phrase analysis rather than free-form translation.
  • The morphological analyzer assumes input in the Ragileo orthography, requiring conversion for other orthographies.
  • Some users (beginners) struggled with technical labels like "VTR" and "VI."
    • Experienced learners did not find the tool necessary.
  • Future Work:
    • Incorporate more recent statistical machine learning components for the orthography converter and morphological analyzer as more data becomes available.
  • Deploy full machine translation systems.
  • Add a text-to-speech component to aid pronunciation teaching.
  • Develop a system for annotating words and phrases to collect data for future tasks.
  • Create an offline version of the tool for areas with low connectivity.
  • Incorporate available Mapuzugun corpora.

Summary updated

BibTeX
@inproceedings{ahumada2022educational,
  author = {Cristian Ahumada and Claudio Gutierrez and Antonios Anastasopoulos},
  title = {{Educational Tools for Mapuzugun}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2022},
}
Figure