Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

GliLem: Leveraging GliNER for Contextualized Lemmatization in Estonian

Aleksei Dorkin, Kairit Sirts

TL;DR

GliLem addresses Estonian lemmatization by combining Vabamorf's rule-based analyses with GliNER's open-vocabulary disambiguation, recasting lemma selection as transformation-rule labeling. Trained on the Estonian UD EDT corpus, GliLem achieves 97.7% lemmatization accuracy ( vs 89.2% for Vabamorf ), approaching oracle performance and outperforming a pattern-based token classifier by about 1–2 points. To assess practical impact, the authors translate English DBpedia-Entity to Estonian and evaluate BM25-based IR across four preprocessing methods, finding that lemmatization generally improves retrieval over stemming, with GliLem yielding small but consistent recall gains at larger top-k values. The work demonstrates that external disambiguation can substantially boost lemmatization quality and offers modest benefits for information retrieval in a hybrid lexical-dense IR setting, while highlighting data quality and computational considerations for future work.

Abstract

We present GliLem -- a novel hybrid lemmatization system for Estonian that enhances the highly accurate rule-based morphological analyzer Vabamorf with an external disambiguation module based on GliNER -- an open vocabulary NER model that is able to match text spans with text labels in natural language. We leverage the flexibility of a pre-trained GliNER model to improve the lemmatization accuracy of Vabamorf by 10% compared to its original disambiguation module and achieve an improvement over the token classification-based baseline. To measure the impact of improvements in lemmatization accuracy on the information retrieval downstream task, we first created an information retrieval dataset for Estonian by automatically translating the DBpedia-Entity dataset from English. We benchmark several token normalization approaches, including lemmatization, on the created dataset using the BM25 algorithm. We observe a substantial improvement in IR metrics when using lemmatization over simplistic stemming. The benefits of improving lemma disambiguation accuracy manifest in small but consistent improvement in the IR recall measure, especially in the setting of high k.

GliLem: Leveraging GliNER for Contextualized Lemmatization in Estonian

TL;DR

GliLem addresses Estonian lemmatization by combining Vabamorf's rule-based analyses with GliNER's open-vocabulary disambiguation, recasting lemma selection as transformation-rule labeling. Trained on the Estonian UD EDT corpus, GliLem achieves 97.7% lemmatization accuracy ( vs 89.2% for Vabamorf ), approaching oracle performance and outperforming a pattern-based token classifier by about 1–2 points. To assess practical impact, the authors translate English DBpedia-Entity to Estonian and evaluate BM25-based IR across four preprocessing methods, finding that lemmatization generally improves retrieval over stemming, with GliLem yielding small but consistent recall gains at larger top-k values. The work demonstrates that external disambiguation can substantially boost lemmatization quality and offers modest benefits for information retrieval in a hybrid lexical-dense IR setting, while highlighting data quality and computational considerations for future work.

Abstract

We present GliLem -- a novel hybrid lemmatization system for Estonian that enhances the highly accurate rule-based morphological analyzer Vabamorf with an external disambiguation module based on GliNER -- an open vocabulary NER model that is able to match text spans with text labels in natural language. We leverage the flexibility of a pre-trained GliNER model to improve the lemmatization accuracy of Vabamorf by 10% compared to its original disambiguation module and achieve an improvement over the token classification-based baseline. To measure the impact of improvements in lemmatization accuracy on the information retrieval downstream task, we first created an information retrieval dataset for Estonian by automatically translating the DBpedia-Entity dataset from English. We benchmark several token normalization approaches, including lemmatization, on the created dataset using the BM25 algorithm. We observe a substantial improvement in IR metrics when using lemmatization over simplistic stemming. The benefits of improving lemma disambiguation accuracy manifest in small but consistent improvement in the IR recall measure, especially in the setting of high k.
Paper Structure (15 sections, 1 figure, 3 tables)

This paper contains 15 sections, 1 figure, 3 tables.

Table of Contents

  1. Introduction
  2. Vabamorf and GliNER
  3. Vabamorf
  4. GliNER
  5. Adapting GliNER for Vabamorf Lemma Disambiguator
  6. Enhancing Disambiguation with GliLem
  7. GliLem training
  8. Token Classification Baseline
  9. Results
  10. Impact on Information Retrieval
  11. Dataset Preparation and Translation
  12. Retrieval Experiments
  13. Evaluation Metrics
  14. Results and Discussion
  15. Conclusion

Figures (1)

  • Figure 1: Schematic representation of the GliNER architecture applied to lemmatization.