Enhancements to the BOUN Treebank Reflecting the Agglutinative Nature of Turkish

By Büsra Marsan, Furkan Akkurt, Muhammet Sen, Merve Gürbüz, Onur Güngör, Şaziye Betül Özateş, Suzan Üsküdarli, Arzucan Özgür, Tunga Güngör, Balkız Öztürk. ALTNLP, 2022

16 citations recorded by the source corpus.

In this study, we aim to offer linguistically motivated solutions to resolve the issues of the lack of representation of null morphemes, highly productive derivational processes, and syncretic morphemes of Turkish in the BOUN Treebank without diverging from the Universal Dependencies framework. In order to tackle these issues, new annotation conventions were introduced by splitting certain lemmas and employing the MISC (miscellaneous) tab in the UD framework to denote derivation. Representational capabilities of the re-annotated treebank were tested on a LSTM-based dependency parser and an updated version of the BoAT Tool is introduced.

Subfields: Language and Linguistics, Natural Language Processing, Computational Linguistics & Semantics

View Paper: https://arxiv.org/abs/2207.11782

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Language and Linguistics

1. Büsra Marsan
2. Furkan Akkurt
3. Nuri Tas
4. Yiğit Bekir Kaya
5. A. C. Tantuğ

In Natural Language Processing

1. Büsra Marsan
2. Furkan Akkurt
3. Arzucan Özgür
4. Nuri Tas
5. Yiğit Bekir Kaya
6. A. C. Tantuğ

In Computational Linguistics & Semantics

1. Büsra Marsan
2. Furkan Akkurt
3. Arzucan Özgür
4. Nuri Tas
5. Yiğit Bekir Kaya
6. A. C. Tantuğ

Citations in the corpus

• Evaluating the Quality of a Corpus Annotation Scheme Using Pretrained Language Models (2024, methodology): “…of the proposed approach, we compare the token-level annotations (specifically lemmas, part-of-speech tags, and morphological features) between two versions of the Turkish BOUN Treebank (Marşan et al., 2022) in the UD framework, the versions 2.8 (Türk et al., 2021) and 2.11 (Marşan et al., 2022).”
• Neural Morphosyntactic Tagging and Dependency Parsing for Uzbek (2026, unknown): “For Turkish, morphology-aware representations improve parsing over purely lexical baselines [12], as well as for Uzbek [13], [14], and similar gains are reported for Kazakh, Uyghur, and Tatar [15].”
• A Comparative Linguistic Analysis of Ottoman and Modern Turkish through UD Treebanks (2026, unknown): “In this formula, O ᵢ denotes the observed frequency of a feature in corpus i (OTA-DUDU or TR-BOUN), and E ᵢ refers to the expected frequency under the null hypothesis of no difference between corpora.”
• The Annotation of Turkic Copula Verbs in UD (2025, unknown): “…- ccomp; PRON - nsubj; VERB - nsubj; VERB - csubj; AUX - aux; VERB - parataxis [22] Tatar-NMCTT × [23] Turkish-Atis × × [24] Turkish-BOUN [25], [26] Turkish-English-BUTR × × VERB - nmod; VERB - ccomp; VERB - acl [27] Turkish-FrameNet × × Turkish-Kenet × × × [32] Turkish-PUD Turkish-TueCL × AUX…”
• POS-tags, lemmatization, and feature frequencies: stylometric analyses in Turkish using stylo (2024, unknown): “Initialpreprocessing,tokenization,lemmatization,andPOS-taggingwasperformed on the corpus using a UDPipe 2 model (Straka, 2023) trained on a Turkish treebank created by researchers at Bo˘gaziçi University (Mar¸san et al., 2022).”
• Strategies for the Annotation of Pronominalised Locatives in Turkic Universal Dependency Treebanks (2024, unknown): “(Taguchi, 2023) 148 2280 0 1264 843 14 28 7 Turkish/Atis (Köse and Yıldız, 2023) 5432 45907 0 2133 995 13 36 7 Turkish/BOUN (Türk et al., 2022; Marşan et al., 2022) (Marşan et al., 2023) 9761 125212 3374 37052 12649 16 46 7 Turkish/FrameNet (Cesur et al., 2023b) 2698 19223 0 8403 3905 15 30 7…”
• RoBERTurk: Adjusting RoBERTa for Turkish (2024, methodology): “Despite being trained on a larger dataset than its competitors, our model achieves competitive scores on BOUN and XTREME datasets but underperforms on IMST.”
• Slovene Morphological and Word Formation Segmentation: A Novel Dataset and Evaluation (2026, unknown): “Explicit ∅ - morpheme annotation is rare, with most datasets lacking this information despite its importance for linguistic analysis (Marşan et al., 2022).”
• BERT2D: Two Dimensional Positional Embeddings for Efficient Turkish NLP (2024, methodology): “We also used the BOUN dataset 5 [38], which includes 9,761 manually annotated sentences from various topics, including biographical texts, national newspapers, instructional texts, articles on popular culture and essays for the POS tagging task.”