Elamur Research⌕
☰
home / papers / 2026 / Sequence Labeling for Constituent Parsing: A Comparative Study and Encoding Innovations

Sequence Labeling for Constituent Parsing: A Comparative Study and Encoding Innovations

Summary

Several schemes exist for turning constituent parsing into sequence labeling, where each word gets exactly one label that encodes part of the phrase-structure tree. Each scheme was published with its own architecture, settings, decoding fixes and tree format (binary or n-ary), so their reported scores cannot be fairly compared. The authors do three things. They unify the existing encoding families (common-ancestor/depth-based, directional depth-based and transition-based) under one label format and analyse their theoretical strengths and weaknesses. They add new encodings and variants, including an unbounded encoding with a compact label set, and they reformulate the attach-juxtapose transition parser as a tagger for the first time. They then evaluate all encodings with one setup, the MaChAmp sequence labeler, on the English Penn Treebank and the SPMRL treebanks, nine languages in total. Under these equal conditions, the best encodings average about 92 F1 and parse at roughly 1,300–2,100 words per second on CPU. The study gives practitioners a reliable basis for choosing an encoding, and all implementations are released as open source.

Key points

  • Languages: English from the PTB, plus eight SPMRL languages: Swedish, Polish, Korean, Hungarian, Hebrew, German, French and Basque. The authors note an Indo-European bias, with five of the nine languages being Indo-European.
  • Accuracy by encoding:
    • The highest average F1 across the nine treebanks is 92.33, from the left-directional variant "l-desc br".
    • "dynamic lb" averages 92.20.
    • Attach-juxtapose ("att-jxt") averages 92.13 and has the best PTB score, 95.76.
  • Weak variants: Some 4tag and directional variants drop sharply, to roughly 70–80 average F1. Binarization choices and variant details therefore matter a lot.
  • Speed (CPU-only, i9-13900K, MaChAmp):
    • Absolute, relative and dynamic encodings average about 2,050–2,130 words per second.
    • att-jxt is the slowest, at about 1,346 words per second.
  • Code: All encodings are available in the CoDeLin repository (github.com/Polifack/CoDeLin). The paper is published under CC BY-NC-ND 4.0.

Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.

Fields: Natural language processing

Topics Syntax & grammar formalisms

Institutions

🎓Universidade da CoruñaA Coruña · Spain9 works · 169 citations

Cite

Cite as Diego Roca, David Vilares and Carlos Gómez-Rodríguez. 2026. Sequence Labeling for Constituent Parsing: A Comparative Study and Encoding Innovations. Computational Linguistics.

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.1162/coli.a.603

@article{roca2026sequence,
  title = {{Sequence Labeling for Constituent Parsing: A Comparative Study and Encoding Innovations}},
  author = {Diego Roca, David Vilares and Carlos Gómez-Rodríguez},
  journal = {Computational Linguistics},
  year = {2026},
  doi = {10.1162/coli.a.603},
  url = {https://doi.org/10.1162/coli.a.603}
}

References

14 references, 3 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Abney, Steven P. 1992. Parsing by chunks. Principle-based parsing: Computation and Psycholinguistics, pages 257–278. https://doi.org/10.1007/978-94-011 -3474-3 10 Alonso, Miguel A., Carlos Gómez-Rodrı́guez, and Jesús Vilares. 2021. On the use of parsing for named entity recognition. Applied Scienc… — [doi:10.1007/978-94-011](https://doi.org/10.1007/978-94-011) ↗
  2. Amini, Afra and Ryan Cotterell. 2022. On parsing as tagging. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 8884–8900. https://doi.org/10.18653/v1/2022.emnlp-main.607 Amini, Afra, Tianyu Liu, and Ryan Cotterell. 2023. Hexatagging: Projective depende… ↗
  3. Clark, Stephen and James R. Curran. 2004. The importance of supertagging for wide-coverage CCG parsing. In COLING 2004: Proceedings of the 20th International Conference on Computational Linguistics, pages 282–288. https://doi.org/10 .3115/1220355.1220396 Coavoux, Maximin and Shay B. Cohen. 2019. Dis… find
  4. Li, Junhui, Deyi Xiong, Zhaopeng Tu, Muhua Zhu, Min Zhang, and Guodong Zhou. 2017. Modeling source syntax for neural machine translation. In Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 688–697. https://doi .org/10.18653/v1/P1… find
  5. Lin, Boda, Zijun Yao, Jiaxin Shi, Shulin Cao, Binghao Tang, Si Li, Yong Luo, Juanzi Li, and Lei Hou. 2022. Dependency parsing via sequence generation. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 7339–7353. https://doi.org/10.18653/v1/2022.findings-emnlp.543 Lin,… ↗
  6. Lin, Yongjie, Yi Chern Tan, and Robert Frank. 2019. Open sesame: Getting inside BERT’s linguistic knowledge. In Proceedings of the 2019 ACL Workshop BlackboxNLP: Analyzing and Interpreting Neural Networks for NLP, pages 241–253. https://doi.org/10.18653/v1/W19-4825 Liu, Jiangming and Yue Zhang. 201… ↗ our card
  7. Rotman, Guy and Roi Reichart. 2019. Deep contextualized self-training for low resource dependency parsing. Transactions of the Association for Computational Linguistics, 7:695–713. https://doi.org/10.1162/tacl a 00294 Schlippe, Tim, ThuyLinh Nguyen, and Stephan Vogel. 2008. Diacritization as a mach… ↗ our card
  8. Tenney, Ian, Patrick Xia, Berlin Chen, Alex Wang, Adam Poliak, R. Thomas McCoy, Najoung Kim, Benjamin Van Durme, Samuel R. Bowman, Dipanjan Das, et al. 2018. What do you learn from context? Probing for sentence structure in contextualized word representations. In International Conference on Learning… our card
  9. Tian, Yuanhe, Yan Song, Fei Xia, and Tong Zhang. 2020. Improving constituency parsing with span attention. In Findings of the Association for Computational Linguistics: EMNLP 2020, pages 1691–1703. https://doi.org/10.18653/v1/2020.findings-emnlp.153 Vacareanu, Robert, George Caique Gouveia Barbosa,… ↗
  10. Van Cranenburgh, Andreas, Remko Scha, and Rens Bod. 2016. Data-oriented parsing with discontinuous constituents and function tags. Journal of Language Modelling, 4(1):57–111. https://doi.org/10.15398/jlm.v4i1.100 van der Goot, Rob, Ahmet Üstün, Alan Ramponi, Ibrahim Sharaf, and Barbara Plank. 202… ↗
  11. Vilares, David, Mostafa Abdou, and Anders Søgaard. 2019. Better, faster, stronger sequence tagging constituent parsers. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers),… find
  12. Wang, Bin, Jiangzhou Ju, Yang Fan, Xinyu Dai, Shujian Huang, and Jiajun Chen. 2022. Structure-unified M-tree coding solver for math word problem. In Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing, pages 8122–8132. https://doi.org/10.18653/v1/2022.emnlp -main.… ↗
  13. Yang, Songlin and Kewei Tu. 2023. Don‘t parse, choose spans! Continuous and discontinuous constituency parsing via autoregressive span selection. In Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 8420–8433. https://doi.org/10.18… — [doi:10.18653/v1/2023](https://doi.org/10.18653/v1/2023) find
  14. Yu, Chen and Daniel Gildea. 2022. Sequence-to-sequence AMR parsing with ancestor information. In Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 2: Short Papers), pages 571–577. https://doi.org/10.18653/v1/2022.acl -short.63 Zamaraeva, Olga and Carlos… ↗

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/en/papers/roca-2026-sequence-labeling-constituent-parsing-comparative/