NLP Progress in Indigenous Latin American Languages

Atnafu Lambebo Tonja, Fazlourrahman Balouchzahi, Sabur Butt, Olga Kolesnikova, Hector Ceballos, Alexander Gelbukh, Thamar Solorio

TL;DR

This paper surveys the current state of Natural Language Processing (NLP) for indigenous Latin American languages, highlighting a significant gap between rapid technological progress and community needs. It advocates for a more inclusive, collaborative research paradigm that moves beyond machine translation to address the diverse linguistic and cultural requirements of these marginalized communities.

What Problem Does This Paper Address?

The "digital divide" isn't just about hardware; it's about language. While AI and NLP advance at breakneck speed for dominant languages, over 88% of the world's languages are neglected by current technology. In Latin America, hundreds of indigenous languages face the threat of extinction, a process accelerated by a lack of technological support that restricts their visibility and utility in the modern world.

The authors address the severe underrepresentation of these languages in NLP research. They explore the disconnect between what researchers are building—often driven by data availability—and what indigenous communities actually need to preserve their cultural heritage and participate in the global digital landscape.

Methods

The authors employed a two-pronged approach to map the current landscape:

  1. Meta-Analysis: They conducted a systematic review of the ACL Anthology to quantify NLP research efforts. They tracked the number of languages represented per country and the specific NLP tasks addressed (e.g., translation, speech recognition).
  2. Qualitative Survey: To understand the human element, they surveyed two distinct groups: 27 researchers/students active in the field and 350 members of indigenous communities across Latin America. This allowed them to compare academic priorities with the lived experiences and desires of the speakers themselves.

Datasets

The paper is a survey and meta-analysis, so it does not introduce a new computational dataset for model training. Instead, it uses the ACL Anthology as its primary source for research statistics. It also highlights the critical role of the AmericasNLP shared tasks in fostering the creation of new datasets for machine translation in various indigenous languages.

Key Insights and Ideas

  • The Machine Translation (MT) Bias: Research is heavily skewed toward MT, which accounts for approximately 40% of all publications. While useful, this focus often comes at the expense of fundamental tools like morphological analyzers or word embeddings that are essential for building robust systems in low-resource settings.
  • Divergent Perspectives on Challenges: There is a notable "perception gap." Researchers primarily cite a "lack of resources" as their biggest hurdle. In contrast, community members emphasize a "lack of inclusion" in the research process and "little government interest." This suggests that the barrier to progress is as much social and political as it is technical.
  • The "AmericasNLP" Effect: The establishment of dedicated venues like the AmericasNLP workshop in 2021 led to a dramatic spike in publications. This proves that institutional visibility and dedicated tracks are key drivers for research in underrepresented areas.
  • Linguistic Erasure through Classification: The paper notes that some governments relegate indigenous languages to the status of "mere dialects." This isn't just a label; it's a policy that denies these languages the recognition and resources needed for survival in the digital age.
Figure 2: Number of indigenous languages present in NLP research per country

Key Results

  • Geographic Disparity: Out of 33 countries in the Americas, only 13 have indigenous languages represented in NLP research. Mexico leads the way with 22 languages represented, followed by Brazil (15) and Peru (12).
  • Language Leaders: Quechua is by far the most researched language, followed by Shipibo-Konibo, Nahuatl, Aymara, and Guarani. As seen in Figure 3, these languages cover a wider variety of NLP tasks compared to others.
  • Task Scarcity: Beyond MT, there is a "notable scarcity" of tools. For example, word embeddings—a cornerstone of modern NLP—are almost entirely absent for these languages, with the sole exception of Guarani.
Figure 3: Number of publications per language vs tasks. We did not include publications, tasks, and languages from shared tasks like AmericasNLP in these statistics.

Limitations and Future Work

The authors acknowledge that their study is limited to papers indexed in the ACL Anthology, which may miss work published in regional or non-NLP-specific venues. Furthermore, the survey results, while broad, may not generalize to every indigenous community.

Future Directions: * Inclusive Data Collection: Researchers must move toward involving native speakers in the creation of "gold standard" datasets to ensure cultural and linguistic accuracy. * Diversifying Tasks: There is an urgent need to focus on neglected tasks like Automatic Speech Recognition (ASR), Named Entity Recognition (NER), and Part-of-Speech (POS) tagging. * Policy and Ethics: The community needs to advocate for government support and implement ethical audits to prevent cultural misappropriation and ensure that technological advancements truly serve the speakers' interests.

Summary updated

Abstract

The paper focuses on the marginalization of indigenous language communities in the face of rapid technological advancements. We highlight the cultural richness of these languages and the risk they face of being overlooked in the realm of Natural Language Processing (NLP). We aim to bridge the gap between these communities and researchers, emphasizing the need for inclusive technological advancements that respect indigenous community perspectives. We show the NLP progress of indigenous Latin American languages and the survey that covers the status of indigenous languages in Latin America, their representation in NLP, and the challenges and innovations required for their preservation and development. The paper contributes to the current literature in understanding the need and progress of NLP for indigenous communities of Latin America, specifically low-resource and indigenous communities in general.
BibTeX
@inproceedings{tonja2024nlp,
  author = {Atnafu Lambebo Tonja and Fazlourrahman Balouchzahi and Sabur Butt and Olga Kolesnikova and Hector Ceballos and Alexander Gelbukh and Thamar Solorio},
  title = {{NLP Progress in Indigenous Latin American Languages}},
  booktitle = {Conference of the North American Chapter of the Association for Computational Linguistics},
  year = {2024},
  doi = {10.18653/v1/2024.naacl-long.385},
}
Figure