Elamur Research⌕
☰
home / papers / 2026 / FreeTxt-Vi: A Benchmarked Vietnamese-English Toolkit for Segmentation, Sentiment, and Summarisation

FreeTxt-Vi: A Benchmarked Vietnamese-English Toolkit for Segmentation, Sentiment, and Summarisation

Summary

Vietnamese has over 80 million speakers and solid NLP resources such as VnCoreNLP, PhoBERT and ViT5. Using them, however, usually requires programming skills. Browser-based analysis platforms like Voyant Tools or the original English–Welsh FreeTxt lack Vietnamese support, especially for word segmentation, since Vietnamese spaces mark syllable boundaries rather than word boundaries.

The authors present FreeTxt-Vi, an open-source web application that lets non-programmers build and explore bilingual Vietnamese–English free-text collections, such as survey answers. It combines classic corpus tools (concordances, keywords, collocations, visualisations) with three NLP modules:

  • a segmenter that pairs VnCoreNLP with BPE,
  • a TabularisAI sentiment classifier fine-tuned for both languages,
  • a Qwen2.5 model fine-tuned for abstractive summarisation.

Each module is benchmarked separately against common baselines. The summariser outperforms every compared system on both VNDS and CNN/DailyMail. The authors also report that joint fine-tuning on the two languages helps Vietnamese without hurting English.

The work turns fragmented Vietnamese NLP research into evaluated, freely available infrastructure for education, the social sciences and digital humanities.

Key points

  • Pipeline: hybrid VnCoreNLP + BPE segmentation, fine-tuned TabularisAI (from tabularisai/multilingual-sentiment-analysis) for sentiment, and fine-tuned Qwen2.5 for summarisation. The sentiment module is evaluated on VLSP2016 (Vietnamese) and IMDb (English).
  • Summarisation results: ROUGE-1/2/L of 53.1/27.5/48.4 on VNDS and 52.4/27.0/48.0 on CNN/DailyMail. This is roughly 3–4 ROUGE points above strong baselines such as Flan-T5-base and Gemma-7B. Other baselines include LLaMA2-7B, mBART50, T5-base, BART-large and base Qwen2.5.
  • Speed trade-off: the fine-tuned summariser runs at 112 tok/s on VNDS, somewhat slower than base Qwen2.5 and Gemma-7B (115–120 tok/s) and the smaller models (110–150 tok/s; T5-base 140, Flan-T5-base 150).
  • Cross-lingual gain: the authors report +1.4% for Vietnamese over the multilingual base model, with no loss in English accuracy.
  • Availability: the tool builds on FreeTxt (English–Welsh). Code, fine-tuned models, configurations and sample data are released at github.com/VinNLP/Free-txt-vi. No user data is kept beyond the active session.

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: Linguistics & languages

Cite

The paper itself, not this page. BibTeX from the library's record of the work. DOI 10.63317/58migi9wn3u6

@inproceedings{nguyen2026freetxt,
  title = {{FreeTxt-Vi: A Benchmarked Vietnamese-English Toolkit for Segmentation, Sentiment, and Summarisation}},
  author = {Nguyen, Hung Huy and El-Haj, Mo and Rayson, Paul and Knight, Dawn},
  booktitle = {Proceedings of the Fifteenth Language Resources and Evaluation Conference},
  year = {2026},
  doi = {10.63317/58migi9wn3u6},
  url = {https://doi.org/10.63317/58migi9wn3u6}
}

References

22 references, 2 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet.

  1. Laurence Anthony. 2019. Antconc (version 3.5.8) [computer software]. find
  2. Mauro Cettolo, Jan Niehues, Sebastian Stüker, Luisa Bentivogli, Rolando Cattoni, and Marcello Federico. 2016. The iwslt 2016 evaluation campaign. In Proceedings of the 13th International Conference on Spoken Language Translation. find
  3. Jacob Devlin, Ming-Wei Chang, Kenton Lee, and Kristina Toutanova. 2019. BERT: Pre-training of deep bidirectional transformers for language understanding. In Proceedings of the 2019 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies,… find
  4. Pratik Joshi, Sebastin Santy, Amar Budhiraja, Kalika Bali, and Monojit Choudhury. 2021. The state and fate of linguistic diversity and inclusion in the nlp world. find our card
  5. Nouran Khallaf, Ignatius Ezeani, Dawn Knight, Paul Rayson, Mo El-Haj, John Vidler, James Davies, and Fernando Alva-Manchego. 2025. Freetxt: Analyse and visualise multilingual qualitative survey data for cultural heritage sites. In The Recent Advances in Natural Language Processing Conference (RANLP… find
  6. Adam Kilgarriff et al. 2014. The Sketch Engine: ten years on. In Lexicography, pages 7–36. Springer. find
  7. Dawn Knight, Nouran Khallaf, Paul Rayson, Mahmoud El-Haj, Ignatius Ezeani, and Steve Morris. 2024. FreeTxt: A corpus-based bilingual free-text survey and questionnaire data analysis toolkit. Applied Corpus Linguistics, 4(3):100103. find
  8. Dawn Knight, Steve Morris, Paul Rayson, Jonathan Morris, Mared Williams, and Tess Fitzpatrick. 2018. CorCenCC: The National Corpus of Contemporary Welsh. In Proceedings of the 11th International Conference on Language Resources and Evaluation (LREC 2018). European Language Resources Association (ELR… find
  9. Taku Kudo and John Richardson. 2018. Sentence-Piece: A simple and language independent subword tokenizer and detokenizer for neural text processing. In Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing: System Demonstrations, pages 66– 71, Brussels, Belgium. Asso… find
  10. Rada Mihalcea and Paul Tarau. 2004. TextRank: Bringing order into text. In Proceedings of the 2004 Conference on Empirical Methods in Natural Language Processing, pages 404–411, Barcelona, Spain. Association for Computational Linguistics. find
  11. Quoc Hung Ngo, Werner Winiwarter, and Bartholomäus Wloka. 2013. Evbcorpus-a multi-layer english-vietnamese bilingual corpus for studying tasks in comparative linguistics. In Proceedings of the 11th Workshop on Asian Language Resources, pages 1–9. find
  12. Dat Quoc Nguyen and Anh Tuan Nguyen. 2020. Phobert: Pre-trained language models for vietnamese. arXiv preprint arXiv:2003.00744. find our card
  13. Huyen TM Nguyen, Hung V Nguyen, Quyen T Ngo, Luong X Vu, Vu Mai Tran, Bach X Ngo, and Cuong A Le. 2018. Vlsp shared task: sentiment analysis. Journal of Computer Science and Cybernetics, 34(4):295–310. find
  14. V.-H. Nguyen, T.-C. Nguyen, M.-T. Nguyen, and N. X. Hoai. 2019. Vnds: A vietnamese dataset for summarization. In 2019 6th NAFOSTED Conference on Information and Computer Science (NICS), pages 375–380. IEEE. find
  15. James W Pennebaker, Ryan L Boyd, Kayla Jordan, and Kate Blackburn. 2015. The development and psychometric properties of LIWC2015. University of Texas at Austin. find
  16. Long Phan, Hieu Tran, Hieu Nguyen, and Trieu H Trinh. 2022. Vit5: Pretrained text-to-text transformer for vietnamese language generation. arXiv preprint arXiv:2205.06457. find
  17. Paul Rayson. 2003. Matrix: a statistical method and software tool for linguistic analysis through corpus comparison. Ph.D. thesis, Lancaster University. find
  18. Rico Sennrich, Barry Haddow, and Alexandra Birch. 2015. Neural machine translation of rare words with subword units. arXiv preprint arXiv:1508.07909. find
  19. Stéfan Sinclair and Geoffrey Rockwell. 2016. Voyant tools: A web-based reading and analysis environment. In Digital Humanities 2016. find
  20. Xinying Song, Alex Salcianu, Yang Song, Dave Dopson, and Denny Zhou. 2020. Fast wordpiece tokenization. arXiv preprint arXiv:2012.15524. tabularisai, Samuel Gyamfi, Vadim Borisov, and Richard H. Schreiber. 2025. multilingualsentiment-analysis (revision 69afb83). find
  21. Qwen Team. 2024. Qwen2.5: A party of foundation models. find
  22. Thanh Vu, Dat Quoc Nguyen, Dai Quoc Nguyen, Mark Dras, and Mark Johnson. 2018. Vncorenlp: A vietnamese natural language processing toolkit. arXiv preprint arXiv:1801.01331. find

↑ back to top

This page in Markdown → https://md.research.elamur.ai/free/papers/nguyen-2026-freetxt-vi-benchmarked-vietnamese-english/