Kirill Milintsevich

Postdoc at INA (Institut national de l'audiovisuel)

Paris, France

I am a postdoctoral researcher at the Institut national de l’audiovisuel (INA) in Paris, working on automatic coreference resolution and robust evaluation of NLP models for French. My research interests span NLP and clinical AI, with a focus on external knowledge and dataset quality.

During my PhD, I developed neural models for depression symptom estimation from clinical text, integrating domain-specific lexicons and working closely with mental health professionals. This work highlighted the critical importance of dataset validity and careful annotation practices in building reliable models for high-stakes applications.

My broader research has explored diverse NLP topics including low-resource languages, morphological analysis, neural lemmatization, and most recently, coreference resolution in spoken French. Two core themes consistently guide my work:

When working with low-resource or noisy data, the incorporation of external knowledge (lexicons, morphological analyzers, domain-specific indicators) significantly improves robustness;
Data quality matters. Careful annotation, understanding dataset biases, and rigorous validation are essential for developing models that actually work in practice.

Current focus: Advancing robust coreference resolution for French, with emphasis on evaluating model performance across diverse domains and improving handling of ASR-transcribed speech.

selected publications

LREC-COLING
Analyzing Symptom-based Depression Level Estimation through the Prism of Psychiatric Expertise

Navneet* Agarwal, Kirill* Milintsevich, Lucie Metivier, and 3 more authors

In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024), May 2024

Abs Bib

The ever-growing number of people suffering from mental distress has motivated significant research initiatives towards automated depression estimation. Despite the multidisciplinary nature of the task, very few of these approaches include medical professionals in their research process, thus ignoring a vital source of domain knowledge. In this paper, we propose to bring the domain experts back into the loop and incorporate their knowledge within the gold-standard DAIC-WOZ dataset. In particular, we define a novel transformer-based architecture and analyse its performance in light of our expert annotations. Overall findings demonstrate a strong correlation between the psychological tendencies of medical professionals and the behavior of the proposed model, which additionally provides new state-of-the-art results.
@inproceedings{agarwal-etal-2024-analyzing, title = {Analyzing Symptom-based Depression Level Estimation through the Prism of Psychiatric Expertise}, author = {Agarwal, Navneet and Milintsevich, Kirill and Metivier, Lucie and Rotharmel, Maud and Dias, Ga{\"e}l and Dollfus, Sonia}, editor = {Calzolari, Nicoletta and Kan, Min-Yen and Hoste, Veronique and Lenci, Alessandro and Sakti, Sakriani and Xue, Nianwen}, booktitle = {Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024)}, month = may, year = {2024}, address = {Torino, Italia}, publisher = {ELRA and ICCL}, url = {https://aclanthology.org/2024.lrec-main.87}, pages = {974--983} }

Brain Informatics

Towards automatic text-based estimation of depression through symptom prediction

Kirill Milintsevich, Kairit Sirts, and Gaël Dias

Brain Informatics, 2023

Bib Code

@article{milintsevich2023towards,
  title = {Towards automatic text-based estimation of depression through symptom prediction},
  author = {Milintsevich, Kirill and Sirts, Kairit and Dias, Ga{\"e}l},
  journal = {Brain Informatics},
  volume = {10},
  number = {1},
  pages = {1--14},
  year = {2023},
  publisher = {SpringerOpen}
}

EACL
Enhancing Sequence-to-Sequence Neural Lemmatization with External Resources

Kirill Milintsevich and Kairit Sirts

In Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume, Apr 2021

Abs DOI Bib Code

We propose a novel hybrid approach to lemmatization that enhances the seq2seq neural model with additional lemmas extracted from an external lexicon or a rule-based system. During training, the enhanced lemmatizer learns both to generate lemmas via a sequential decoder and copy the lemma characters from the external candidates supplied during run-time. Our lemmatizer enhanced with candidates extracted from the Apertium morphological analyzer achieves statistically significant improvements compared to baseline models not utilizing additional lemma information, achieves an average accuracy of 97.25% on a set of 23 UD languages, which is 0.55% higher than obtained with the Stanford Stanza model on the same set of languages. We also compare with other methods of integrating external data into lemmatization and show that our enhanced system performs considerably better than a simple lexicon extension method based on the Stanza system, and it achieves complementary improvements w.r.t. the data augmentation method.
@inproceedings{milintsevich-sirts-2021-enhancing, title = {Enhancing Sequence-to-Sequence Neural Lemmatization with External Resources}, author = {Milintsevich, Kirill and Sirts, Kairit}, editor = {Merlo, Paola and Tiedemann, Jorg and Tsarfaty, Reut}, booktitle = {Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics: Main Volume}, month = apr, year = {2021}, address = {Online}, publisher = {Association for Computational Linguistics}, url = {https://aclanthology.org/2021.eacl-main.272}, doi = {10.18653/v1/2021.eacl-main.272}, pages = {3112--3122} }

Automated Word Stress Detection in Russian

Maria Ponomareva, Kirill Milintsevich, Ekaterina Chernyak, and 1 more author

In Proceedings of the First Workshop on Subword and Character Level Models in NLP, Sep 2017

Abs DOI Bib Code

In this study we address the problem of automated word stress detection in Russian using character level models and no part-speech-taggers. We use a simple bidirectional RNN with LSTM nodes and achieve accuracy of 90% or higher. We experiment with two training datasets and show that using the data from an annotated corpus is much more efficient than using only a dictionary, since it allows to retain the context of the word and its morphological features.

@inproceedings{ponomareva-etal-2017-automated,
  title = {Automated Word Stress Detection in {R}ussian},
  author = {Ponomareva, Maria and Milintsevich, Kirill and Chernyak, Ekaterina and Starostin, Anatoly},
  editor = {Faruqui, Manaal and Schuetze, Hinrich and Trancoso, Isabel and Yaghoobzadeh, Yadollah},
  booktitle = {Proceedings of the First Workshop on Subword and Character Level Models in {NLP}},
  month = sep,
  year = {2017},
  address = {Copenhagen, Denmark},
  publisher = {Association for Computational Linguistics},
  url = {https://aclanthology.org/W17-4104},
  doi = {10.18653/v1/W17-4104},
  pages = {31--35}
}