Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Establishing a Foundation for Tetun Ad-Hoc Text Retrieval: Stemming, Indexing, Retrieval, and Ranking

Gabriel de Jesus, Sérgio Nunes

TL;DR

This work addresses the lack of Tetun information retrieval resources by developing a complete baseline pipeline for ad-hoc retrieval in a low-resource language. It introduces Labadain-Stopwords, Labadain-Stemmer (light, moderate, heavy), and Labadain-Avaliadór to enable robust evaluation, following a Cranfield/TREC-style paradigm. Comprehensive experiments reveal that stopword network-based detection yields superior Tetun stopword sets and that hyphen and apostrophe removal substantially improves retrieval, especially for short text, while stemming offers limited gains. The study provides publicly accessible resources and a scalable framework for future Tetun IR research, including potential semantic search and integration of large language models. Overall, the paper establishes Tetun baselines, demonstrates language-specific preprocessing importance, and contributes valuable, openly available evaluation data for low-resource language IR.

Abstract

Searching for information on the internet and digital platforms requires effective retrieval solutions. However, such solutions are not yet available for Tetun, making it difficult to find relevant documents for search queries in this language. To address this gap, we investigate Tetun text retrieval with a focus on the ad-hoc retrieval task. The study begins with the development of essential language resources -- including a list of stopwords, a stemmer, and a test collection -- that serve as a foundation for Tetun text retrieval. Various strategies are evaluated using document titles and content. The results show that retrieving document titles, after removing hyphens and apostrophes but without applying stemming, improves performance compared to the baseline. Efficiency increases by 31.37%, while effectiveness achieves an average relative gains of +9.40% in MAP@10 and +30.35% in NDCG@10 with DFR BM25. Beyond the top-10 cutoff point, Hiemstra LM demonstrates strong performance across multiple retrieval strategies and evaluation metrics. The contributions of this work include the development of Labadain-Stopwords (a list of 160 Tetun stopwords), Labadain-Stemmer (a Tetun stemmer with three variants), and Labadain-Avaliadór (a Tetun test collection comprising 59 topics, 33,550 documents, and 5,900 qrels). These resources are publicly available to support future research in Tetun information retrieval.

Establishing a Foundation for Tetun Ad-Hoc Text Retrieval: Stemming, Indexing, Retrieval, and Ranking

TL;DR

This work addresses the lack of Tetun information retrieval resources by developing a complete baseline pipeline for ad-hoc retrieval in a low-resource language. It introduces Labadain-Stopwords, Labadain-Stemmer (light, moderate, heavy), and Labadain-Avaliadór to enable robust evaluation, following a Cranfield/TREC-style paradigm. Comprehensive experiments reveal that stopword network-based detection yields superior Tetun stopword sets and that hyphen and apostrophe removal substantially improves retrieval, especially for short text, while stemming offers limited gains. The study provides publicly accessible resources and a scalable framework for future Tetun IR research, including potential semantic search and integration of large language models. Overall, the paper establishes Tetun baselines, demonstrates language-specific preprocessing importance, and contributes valuable, openly available evaluation data for low-resource language IR.

Abstract

Searching for information on the internet and digital platforms requires effective retrieval solutions. However, such solutions are not yet available for Tetun, making it difficult to find relevant documents for search queries in this language. To address this gap, we investigate Tetun text retrieval with a focus on the ad-hoc retrieval task. The study begins with the development of essential language resources -- including a list of stopwords, a stemmer, and a test collection -- that serve as a foundation for Tetun text retrieval. Various strategies are evaluated using document titles and content. The results show that retrieving document titles, after removing hyphens and apostrophes but without applying stemming, improves performance compared to the baseline. Efficiency increases by 31.37%, while effectiveness achieves an average relative gains of +9.40% in MAP@10 and +30.35% in NDCG@10 with DFR BM25. Beyond the top-10 cutoff point, Hiemstra LM demonstrates strong performance across multiple retrieval strategies and evaluation metrics. The contributions of this work include the development of Labadain-Stopwords (a list of 160 Tetun stopwords), Labadain-Stemmer (a Tetun stemmer with three variants), and Labadain-Avaliadór (a Tetun test collection comprising 59 topics, 33,550 documents, and 5,900 qrels). These resources are publicly available to support future research in Tetun information retrieval.

Paper Structure

This paper contains 62 sections, 13 figures, 30 tables.

Table of Contents

  1. Introduction
  2. RQ1
  3. RQ2
  4. Background and Related Work
  5. Stopwords
  6. Stemming
  7. Test Collection
  8. Summary
  9. Tetun
  10. Overview
  11. Orthography
  12. Morphology
  13. Portuguese Loanwords
  14. Dataset
  15. Methodology
  16. ...and 47 more sections

Figures (13)

  • Figure 1: Methodology for Establishing Baselines in Tetun Ad-Hoc Text Retrieval.
  • Figure 2: Comparison of the Network-Based Approach Performance at Mid-Range and Higher Cutoff Levels.
  • Figure 3: Process of Constructing a Text Sample for Evaluating Tetun Stemmer's Performance.
  • Figure 4: UI vs. OI Plot Showing ERRT Distances.
  • Figure 5: Sample of Document Formatted Following TREC Guidelines: Original (left) and English translation (right).
  • ...and 8 more figures