Pith. sign in

REVIEW 5 major objections 7 minor 2 cited by

Unpacking Robustness in Inflectional Languages: Adversarial Evaluation and Mechanistic Insights

T0 review · 5 major / 7 minor · reviewed 2026-08-15 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗ Review JSON

Pith's one-line read This paper claims that inflectional morphology in a language model's training data increases its resilience to word-form adversarial attacks, and that this resilience is carried by five specific attention heads in the model's first layer.

desk verdict Good question and a useful dataset, but the causal claim about inflection circuits is not established by the experiment as run. read the letter →

arxiv 2505.07856 v1 pith:WCIARJNH submitted 2025-05-08 cs.CL cs.AI

Pawe{l} Walkowiak , Marek Klonowski , Marcin Oleksy , Arkadiusz Janz This is my paper · ORCID
classification cs.CLcs.AI
keywords adversarialexamplesinflectionalmorphologymechanisticinterpretabilitycircuitdiscoverytextclassificationsentimentanalysisPolishrobustness
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper asks whether the inflectional complexity of a language changes how vulnerable text classifiers are to adversarial examples. It compares Polish, Czech, and English under four attack methods and reports that character-level word-bugging attacks (TextBugger) are the most effective, while synonym-substitution attacks are gentler but less damaging. Its central claim is that models trained on inflectionally rich language become more resilient to word-form attacks, and that this resilience is concentrated in identifiable attention heads in the transformer's first layer. To support this, the authors build a parallel Polish/English dataset of sentiment sentences that differ only in whether the key noun is syncretic or regularly inflected, and use circuit-discovery techniques to isolate the components that react to inflection. A reader should care because adversarial robustness is usually studied on English, and this work suggests morphological variation is a defensive resource, not just a complication.

What carries the argument

The central object is the circuit, a minimal subgraph of a transformer made of attention heads and MLP layers connected by high-importance edges. The method for finding circuits is Edge Attribution Patching with Integrated Gradients (EAP-IG), which scores each directed edge by its contribution to the model's confidence in the correct class when inputs are switched from clean to corrupted. The paper's own instrument is the Inflectional Circuit Detection Dataset, a parallel Polish/English sentiment corpus whose clean and corrupted sentences differ by a single noun: one variant uses a syncretic noun (a single form covering several grammatical cases) and another uses a regularly inflected synonym. Because clean and corrupted inputs keep identical token counts, the circuit-discovery pipeline can attribute the prediction difference to specific edges, exposing the five layer-0 heads that respond to inflection.

What would settle it

Fine-tune one pretrained model on matched English and Polish sentiment data of similar length and difficulty, then run TextBugger and the other three attacks; if English attack accuracy is not below Polish, the inflection-robustness claim is unsupported. Alternatively, patch out the five layer-0 heads a0.h3, a0.h5, a0.h6, a0.h7, a0.h11 from the Polish circuit and check whether adversarial accuracy drops to the syncretic circuit's level; if it does not, those heads are not the mechanism.

Watch

Extended reading notes

Core claim

The central claim is that inflectional morphology can act as a defensive resource in text classification. In the authors' experiments the Polish and Czech sentiment models generally lost less accuracy under attacks than the English model, and the Polish inflectional circuit outperformed both the syncretic Polish circuit and the English circuit in mean accuracy under all four attack methods. Mechanistically, the paper identifies five attention heads in layer 0 of the multilingual model — a0.h3, a0.h5, a0.h6, a0.h7, and a0.h11 — that appear in circuits built from inflected Polish inputs across all circuit sizes and are absent from the English reference circuit. The authors interpret these heads as the components that track inflectional endings, and argue that when an attack modifies part of a word, circuits containing these heads preserve classification confidence better than circuits without them.

Load-bearing premise

The paper's central conclusion rests on treating language as the cause of the robustness gap, even though each language uses a different pretrained model and different datasets; if the gap comes from architecture or data difficulty rather than inflection, the conclusion does not follow.

Editorial extensions

If this is right

  • In sentiment analysis, the Polish and Czech models retained more accuracy under all four attacks than the English model, so the robustness advantage of inflectional languages appears specifically in sentiment tasks, not in topic classification.
  • TextBugger is the most effective attack across languages and tasks, yet its outputs score lowest on statistical similarity, meaning the attack that works best is also the easiest to detect by ROUGE.
  • ROUGE and semantic similarity are stable measures of adversarial example quality, while NLI-based similarity is unreliable because a tail of examples receives near-zero entailment scores.
  • The five layer-0 heads a0.h3, a0.h5, a0.h6, a0.h7, and a0.h11 recur in the Polish inflectional circuit at every circuit size and are absent from the English circuit, localizing inflection processing to a compact set of components.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • If the protective effect is real, then multilingual models may become adversarially more robust as the morphological richness of their pretraining corpus rises; a direct test would fine-tune one architecture on artificially inflected English and measure whether TextBugger attack success falls.
  • Because the inflection heads are found in a multilingual model, the same heads may track case or gender contrasts in other inflection-rich languages such as Czech, German, or Russian; this could be checked by running the same circuit-discovery protocol on Czech parallel data and comparing circuit overlap.
  • The paper's cross-language comparison varies language together with model architecture and dataset, so its strongest claim would need a matched-design study holding architecture and data source fixed while varying only morphological complexity.
  • If the heads are causal rather than merely correlational, then ablating or patching just these five heads from the Polish circuit should remove most of its adversarial advantage; that experiment is a natural next step the paper does not run.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

5 major / 7 minor

Summary. The paper investigates adversarial robustness in inflectional languages (Polish, Czech) versus English across two classification tasks, comparing four attack methods (TextBugger, TextFooler, WordNetTextFooler, BERT-Attack) and three similarity metrics (ROUGE, NLI, Semantic Similarity). It then introduces a parallel 'Inflectional Circuit Detection Dataset' built from MultiEmo and uses EAP-IG to extract circuits for Polish syncretic, Polish inflectional, and English variants of the same sentiment task. The authors report that the Polish inflectional circuit contains additional Layer-0 attention heads (a0.h3, a0.h5, a0.h6, a0.h7, a0.h11) and that this circuit shows the highest average robustness under the four attacks, concluding that incorporating inflection in training data can improve adversarial robustness. The paper also reports a comparison of similarity metrics, finding TextBugger attacks less semantically faithful and NLI similarity unstable.

Significance. If the central claims were established, the paper would make a useful contribution: it targets an understudied question (adversarial robustness in inflectional languages), proposes a new parallel dataset for circuit discovery, and applies mechanistic interpretability methods to a non-English language. The dataset and the EAP-IG protocol are potentially reusable assets. However, the load-bearing conclusions are not currently supported: the cross-language comparison is confounded by different models and datasets, and the circuit-robustness comparison compares entire independently selected subgraphs rather than isolating the identified heads, so the causal attribution to inflection-specific heads is not established. The manuscript would need substantial additional analysis, including causal ablations and statistical testing, before the main claims can be accepted.

major comments (5)
  1. [Section 4.2, Tables 1-2] The cross-language robustness comparison treats language as the independent variable, but each language uses a different pretrained model (BERT, HerBERT, RobeCzech) and different datasets, with Czech baselines at 83-86% accuracy versus 94% and above for Polish and English. Any difference in robustness under attack could be due to model architecture, pretraining corpus, task difficulty, or dataset size rather than inflectional morphology. The conclusion in Section 6.1 that 'incorporating inflexion in language model training corpus can increase its robustness' is therefore not supported by this comparison alone. The authors should either control for these variables, provide a causal identification strategy, or substantially soften the claim.
  2. [Sections 5.3-5.4, Figure 5, Table 5] The claim that the five Layer-0 heads (a0.h3, a0.h5, a0.h6, a0.h7, a0.h11) are causally protective is not established. The Polish Syncretic and Polish Inflectional circuits are separately extracted subgraphs selected by EAP-IG on different corrupted datasets; Figure 5 shows that the circuits differ in many added and removed components, not just the Layer-0 heads. Comparing the adversarial robustness of these whole circuits cannot isolate the effect of the identified heads. A causal test, such as ablating or patching only the candidate heads within a fixed circuit, is needed to support the attribution. Additionally, Table 5 shows the inflectional circuit is not consistently better cell-by-cell (e.g., at sizes 75, 150, 200 for TextFooler, or size 200 for TextBugger), so 'consistently achieved the highest average robustness' overstates the evidence.
  3. [Section 5.1, dataset construction] The Inflectional Circuit Detection Dataset is constructed with a selection step that removes clean-corrupted pairs for which the model's prediction does not change. This filters on the outcome variable and can bias circuit discovery toward pairs that already produce model disagreement, potentially inflating or distorting EAP-IG importance scores. The dataset is also expanded with Llama-3.3-70B-Instruct without any validation of the generated examples, and the three variants use different synonym sets; thus the observed circuit differences may reflect lexical or tokenization properties rather than inflectional morphology per se. The authors should report the number of excluded pairs, validate the LLM-generated examples, and assess sensitivity to the synonym selection.
  4. [Section 5, vs. Table 1] The manuscript states that the MultiEmo-based mechanistic analysis used a training part of 334,960 examples and a test part of 41,710 examples, but Table 1 reports the MultiEmo PL dataset as having 4,319 training and 680 test examples. This is a large discrepancy that affects reproducibility and the interpretation of the circuit experiments. The authors need to clarify exactly which data split was used for the XLM-RoBERTa model in Section 5 and reconcile the numbers.
  5. [Table 5] No statistical significance testing or variance reporting is provided for the circuit robustness comparisons. The differences between the inflectional and syncretic circuits at many individual graph sizes appear small (e.g., 48.05 vs. 48.05 for TextBugger at size 150), and the means are driven substantially by the size-50 row. A paired bootstrap or other appropriate test across the 244 examples should be reported before claiming that the inflectional circuit is more robust.
minor comments (7)
  1. [Throughout] There is inconsistent spelling of 'inflection/inflexion'; the abstract and Section 6.1 use 'inflexion' while the rest of the paper uses 'inflection'. Please standardize.
  2. [Table 1] Several numeric entries use a comma as a decimal separator (e.g., '95,17') while others use a period; please use a consistent format.
  3. [Table 3, Wiki_PL rows] The TextFooler and WordNetTextFooler rows for Wiki_PL report identical ROUGE and Semantic similarity values (96.36 ± 3.56 and 97.52 ± 1.43). This looks like a data-entry or copying error and should be checked.
  4. [Section 5.2] The text says 'Mechanical Interpretability' where 'Mechanistic Interpretability' is meant; also the EAP reference [21] is to a blog post, and the formal EAP citation (Nanda, 2023) may be preferable.
  5. [Figures 2 and 3] The axis label 'Semanctic' is misspelled; it should be 'Semantic'.
  6. [Section 6.1] The sentence 'TextBugger consistently produced statistically lower-quality attacks' is not supported by any reported statistical test; please either add the test or rephrase to 'lower-quality on the reported metrics'.
  7. [General] The paper does not state where the dataset, code, or trained circuits will be released; given the new benchmark is a claimed contribution, a reproducibility statement would be helpful.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: circuit robustness is measured empirically against independent attack methods; observed confounds are validity issues, not definitional reductions.

full rationale

The paper's load-bearing claims are empirical rather than definitional. The EAP-IG pipeline selects top-N edge subgraphs from clean/corrupted pairs (Section 5.3), and the adversarial robustness of those subgraphs is then measured under four attack methods (Section 5.4). No equation in the paper defines robustness as equal to the EAP-IG importance score or to the circuit-selection criterion; the robustness comparison is an observed quantity. The Inflectional Circuit Detection Dataset excludes pairs where the model's prediction is unchanged (Section 5.1), so the identified heads are selected for their role in prediction-changing inflectional substitutions, but that is a dataset-design choice, not a circular reduction of the later robustness conclusion. The only self-citation, reference [8] for WordNetTextFooler, supplies one of four attack baselines; the central derivation does not rest on it, and the method is independently defined by its algorithm, so it does not raise the circularity score. Cross-language robustness comparisons in Section 4.2 are confounded by different pretrained models and datasets (BERT/HerBERT/RobeCzech; accuracies 83-98%), and the circuit comparison in Section 5.4 compares independently selected subgraphs rather than controlled ablations of the five named heads; these are validity threats that would need significance testing and causal ablation, but they are not circularity in the sense of a claim reducing by construction to its inputs. Therefore no specific circular step can be quoted, and the score is 0.

Assumptions & free parameters 2 free parameters · 4 assumptions · 0 invented entities

The mechanistic claim rests on four domain assumptions: EAP-IG faithfulness, the isolation of inflection by synonym substitution, comparability across models and languages, and the validity of corrupted-activation circuit evaluation. No free parameters are fit to produce the headline robustness ordering, but circuit size N and the 90 percent semantic filter are hand-set choices that shape the results.

free parameters (2)
  • Circuit size N = 50, 75, 100, 150, 200, 300
    Hand-chosen number of highest-scoring edges used to define circuits in Section 5.3; the central robustness claim is averaged over these values, and no principled selection criterion is given.
  • Cosine similarity threshold for adversarial candidate filtering = 0.90
    Set by hand in Section 4.2 to filter adversarial examples; affects which inputs enter the quality and robustness analyses.
assumptions (4)
  • domain assumption EAP-IG attribution scores identify causally relevant circuits.
    Invoked in Sections 5.2 and 5.3; the paper relies on prior work but does not independently validate faithfulness for the MultiEmo sentiment task.
  • domain assumption Syncretic versus inflectional synonym substitution isolates inflection as the only linguistic variable.
    Dataset construction in Section 5.1 assumes the synonym pairs differ only in inflectional behavior; confounds such as word frequency or token identity are not controlled.
  • domain assumption Robustness differences across languages can be attributed to inflectional complexity.
    The comparison in Section 4.2 uses different pretrained models and datasets per language, yet the conclusions in Section 6.1 attribute observed differences to morphology.
  • domain assumption Corrupting activations of non-circuit edges is a valid way to measure circuit performance.
    The protocol behind Tables 4 and 5 follows standard EAP practice but is not validated with faithfulness checks in this paper.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Unpacking Robustness in Inflectional Languages: Adversarial Evaluation and Mechanistic Insights." pith.science (2026). https://pith.science/paper/WCIARJNH

@misc{pith2026250507856,
  author       = {Pith},
  title        = {Pith review of: Unpacking Robustness in Inflectional Languages: Adversarial Evaluation and Mechanistic Insights},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/WCIARJNH}},
  note         = {Machine review of arXiv:2505.07856}
}
read the original abstract

Various techniques are used in the generation of adversarial examples, including methods such as TextBugger which introduce minor, hardly visible perturbations to words leading to changes in model behaviour. Another class of techniques involves substituting words with their synonyms in a way that preserves the text's meaning but alters its predicted class, with TextFooler being a prominent example of such attacks. Most adversarial example generation methods are developed and evaluated primarily on non-inflectional languages, typically English. In this work, we evaluate and explain how adversarial attacks perform in inflectional languages. To explain the impact of inflection on model behaviour and its robustness under attack, we designed a novel protocol inspired by mechanistic interpretability, based on Edge Attribution Patching (EAP) method. The proposed evaluation protocol relies on parallel task-specific corpora that include both inflected and syncretic variants of texts in two languages -- Polish and English. To analyse the models and explain the relationship between inflection and adversarial robustness, we create a new benchmark based on task-oriented dataset MultiEmo, enabling the identification of mechanistic inflection-related elements of circuits within the model and analyse their behaviour under attack.

Discussion (0). Continue with ORCID to comment.

Forward citations

Cited by 2 Pith papers

Reviewed papers in the Pith corpus that reference this work. Sorted by Pith novelty score. Full citation record

  1. A Survey on Latent Reasoning

    cs.CL 2025-07 conditional novelty 4.0 of 10

    A survey that organizes latent reasoning methods into vertical recurrence, horizontal recurrence, and infinite-depth diffusion, arguing that silent reasoning can beat explicit chain-of-thought.

  2. The Hitchhiker's Guide to Agentic AI: From Foundations to Systems

    cs.AI 2026-06 unverdicted novelty 2.0 of 10

    A survey-style reference book mapping the full agentic-AI stack from transformer internals to production deployment, with no new research result.

Reference graph

Works this paper leans on

38 extracted references · 26 canonical work pages · cited by 2 Pith papers

  1. [1]

    Camburu, T

    O.-M. Camburu, T. Rocktäschel, T. Lukasiewicz, and P. Blunsom. e- snli: Natural language inference with natural language explanations. In Advances in Neural Information Processing Systems, volume 31, pages 9539–9549, 2018. URL https://papers.nips.cc/paper_files/paper/2018/ hash/4c7a167bb329bd92580a99ce422d6fa6-Abstract.html

    work page 2018

  2. [2]

    N. Carlini. A complete list of all (arxiv) adversarial example papers, 2019-2025. URL https://nicholas.carlini.com/writing/2019/ all-adversarial-example-papers.html

    work page 2019

  3. [3]

    J. Chen, Z. Gong, and W. Liu. A nonparametric model for online topic discovery with word embeddings. Inf. Sci., 504:32–47, 2019. doi: 10. 1016/J.INS.2019.07.048. URL https://doi.org/10.1016/j.ins.2019.07.048

    work page doi:10.1016/j.ins.2019.07.048 2019

  4. [4]

    Conmy, A

    A. Conmy, A. N. Mavor-Parker, A. Lynch, S. Heimersheim, and A. Garriga-Alonso. Towards automated circuit discovery for mecha- nistic interpretability. In A. Oh, T. Naumann, A. Globerson, K. Saenko, M. Hardt, and S. Levine, editors, Advances in Neural Information Pro- cessing Systems 36: Annual Conference on Neural Information Process- ing Systems 2023, Ne...

    work page 2023

  5. [5]

    Conneau, K

    A. Conneau, K. Khandelwal, N. Goyal, V . Chaudhary, G. Wenzek, F. Guzmán, E. Grave, M. Ott, L. Zettlemoyer, and V . Stoyanov. Unsupervised cross-lingual representation learning at scale. CoRR, abs/1911.02116, 2019. URL http://arxiv.org/abs/1911.02116

    arXiv 1911

  6. [6]

    Devlin, M

    J. Devlin, M. Chang, K. Lee, and K. Toutanova. BERT: Pre-training of Deep Bidirectional Transformers for Language Understanding. CoRR, abs/1810.04805, 2018. URL http://arxiv.org/abs/1810.04805

    arXiv 2018

  7. [7]

    Dziob and M

    A. Dziob and M. Piasecki. Implementation of the Verb Model in plWord- Net 4.0. In Proceedings of the 9th Global Wordnet Conference, pages 113–122, Nanyang Technological University (NTU), Singapore, Jan

    work page

  8. [8]

    Gniewkowski, P

    M. Gniewkowski, P. Syga, M. Klonowski, P. Walkowiak, and T. Walkowiak. Do Not Trust Me: Explainability Against Text Clas- sification. In ECAI 2023 - 26th European Conference on Artificial Intelligence, September 30 - October 4, 2023, Kraków, Poland - Includ- ing 12th Conference on Prestigious Applications of Intelligent Systems (PAIS 2023), volume 372 of ...

    work page doi:10.3233/faia230356 2023

Show all 38 references
  1. [9]

    Hanna, S

    M. Hanna, S. Pezzelle, and Y . Belinkov. Have faith in faithfulness: Going beyond circuit overlap when finding model mechanisms. In ICML 2024 Workshop on Mechanistic Interpretability , 2024. URL https://openreview.net/forum?id=grXgesr5dT

    2024

  2. [10]

    P. He, X. Liu, J. Gao, and W. Chen. Deberta: Decoding-enhanced bert with disentangled attention. In International Conference on Learn- ing Representations , 2021. URL https://openreview.net/forum?id= XPZIaotutsD

    2021

  3. [11]

    D. Jin, Z. Jin, J. T. Zhou, and P. Szolovits. Is BERT Really Robust? Natural Language Attack on Text Classification and Entailment. CoRR, abs/1907.11932, 2019. URL http://arxiv.org/abs/1907.11932

    1907 arXiv

  4. [12]

    Koco ´n, P

    J. Koco ´n, P. Miłkowski, and K. Kanclerz. Multiemo: Multilingual, multilevel, multidomain sentiment analysis corpus of consumer reviews. In International Conference on Computational Science, pages 297–312. Springer, 2021

    2021

  5. [13]

    Král and L

    P. Král and L. Lenc. Czech Text Document Corpus v 2.0. CoRR, abs/1710.02365, 2017. URL http://arxiv.org/abs/1710.02365

    2017 arXiv

  6. [14]

    J. Li, S. Ji, T. Du, B. Li, and T. Wang. Textbugger: Generating Adversar- ial Text Against Real-world Applications. CoRR, abs/1812.05271, 2018. URL http://arxiv.org/abs/1812.05271

    2018 arXiv

  7. [15]

    L. Li, R. Ma, Q. Guo, X. Xue, and X. Qiu. BERT-ATTACK: Ad- versarial Attack Against BERT Using BERT. In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Pro- cessing, EMNLP 2020, Online, November 16-20, 2020 , pages 6193–

    2020

  8. [16]

    C.-Y . Lin. ROUGE: A Package for Automatic Evaluation of Summaries. In Text Summarization Branches Out, pages 74–81, Barcelona, Spain, July 2004. Association for Computational Linguistics. URL https://www. aclweb.org/anthology/W04-1013

    2004

  9. [17]

    Lukito, B

    J. Lukito, B. Chen, G. M. Masullo, and N. J. Stroud. Comparing a BERT classifier and a GPT classifier for detecting connective language across multiple social media. In Y . Al-Onaizan, M. Bansal, and Y . Chen, editors, Proceedings of the 2024 Conference on Empirical Methods in...

    2024

  10. [18]

    S. M. Lundberg and S.-I. Lee. A unified approach to interpreting model predictions. Advances in neural information processing systems , 30, 2017

    2017

  11. [19]

    A. L. Maas, R. E. Daly, P. T. Pham, D. Huang, A. Y . Ng, and C. Potts. Learning Word Vectors for Sentiment Analysis. In Proceedings of the 49th Annual Meeting of the Association for Computational Linguistics: Human Language Technologies, pages 142–150, Portland, Oregon, USA, J...

    2011

  12. [20]

    Mroczkowski, P

    R. Mroczkowski, P. Rybak, A. Wróblewska, and I. Gawlik. HerBERT: Efficiently Pretrained Transformer-based Language Model for Polish. In Proceedings of the 8th Workshop on Balto-Slavic Natural Language Processing, pages 1–10, Kiyv, Ukraine, Apr. 2021. Association for Com- putat...

    2021

  13. [21]

    N. Nanda. Attribution patching: Activation patching at industrial scale. 2023. URL https://www.neelnanda.io/mechanistic-interpretability/ attribution-patching

    2023

  14. [22]

    C. Olah, N. Cammarata, L. Schubert, G. Goh, M. Petrov, and S. Carter. Zoom in: An introduction to circuits. Distill, 2020. doi: 10.23915/distill. 00024.001. URL https://distill.pub/2020/circuits/zoom-in/

    2020 doi

  15. [23]

    Gpt-3.5 turbo technical overview

    OpenAI. Gpt-3.5 turbo technical overview. https://platform.openai.com/ docs/models/gpt-3-5, 2023. Accessed: 2025-05-03

    2023

  16. [24]

    Pennington, R

    J. Pennington, R. Socher, and C. Manning. GloVe: Global Vectors for Word Representation. In Proceedings of the 2014 Conference on Empirical Methods in Natural Language Processing (EMNLP), pages 1532–1543, Doha, Qatar, Oct. 2014. Association for Computational Linguistics. doi: ...

    2014 doi

  17. [25]

    Reimers and I

    N. Reimers and I. Gurevych. Sentence-BERT: Sentence Embeddings using Siamese BERT-Networks. In Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing . Association for Computational Linguistics, 11 2019. URL http://arxiv.org/abs/1908. 10084

    2019

  18. [26]

    Saloni, M

    Z. Saloni, M. Woli´nski, R. Wołosz, W. Gruszczy´nski, and D. Skowro´nska. Słownik gramatyczny j˛ ezyka polskiego. 10 2012. ISBN 978-83-927277- 2-9

    2012

  19. [27]

    Saphra and S

    N. Saphra and S. Wiegreffe. Mechanistic? CoRR, abs/2410.09087, 2024. doi: 10.48550/ARXIV .2410.09087. URL https://doi.org/10.48550/arXiv. 2410.09087

    2024 doi arXiv

  20. [28]

    Straka, J

    M. Straka, J. Náplava, J. Straková, and D. Samuel. RobeCzech: Czech RoBERTa, a Monolingual Contextualized Language Representation Model. In Text, Speech, and Dialogue - 24th International Confer- ence, TSD 2021, Olomouc, Czech Republic, September 6-9, 2021, Pro- ceedings, volu...

    2021 doi

  21. [29]

    Sundararajan, A

    M. Sundararajan, A. Taly, and Q. Yan. Axiomatic attribution for deep networks. In D. Precup and Y . W. Teh, editors, Proceedings of the 34th International Conference on Machine Learning, ICML 2017, Syd- ney, NSW, Australia, 6-11 August 2017, volume 70 of Proceedings of Machine...

    2017

  22. [31]

    Szegedy, W

    C. Szegedy, W. Zaremba, I. Sutskever, J. Bruna, D. Erhan, I. J. Good- fellow, and R. Fergus. Intriguing properties of neural networks. In 2nd International Conference on Learning Representations, ICLR 2014, Banff, AB, Canada, April 14-16, 2014, Conference Track Proceedings,

    2014

  23. [32]

    K. S. Tai, R. Socher, and C. D. Manning. Improved Semantic Repre- sentations From Tree-Structured Long Short-Term Memory Networks. CoRR, abs/1503.00075, 2015. URL http://arxiv.org/abs/1503.00075

    2015 arXiv

  24. [33]

    Vaswani, N

    A. Vaswani, N. Shazeer, N. Parmar, J. Uszkoreit, L. Jones, A. N. Gomez, L. Kaiser, and I. Polosukhin. Attention is All you Need. In Advances in Neural Information Processing Systems 30: Annual Conference on Neural Information Processing Systems 2017, December 4-9, 2017, Long B...

    2017

  25. [34]

    Y . Wang, W. Qu, and X. Ye. Selecting between BERT and GPT for text classification in political science research. CoRR, abs/2411.05050, 2024. doi: 10.48550/ARXIV .2411.05050. URL https://doi.org/10.48550/arXiv. 2411.05050

    2024 doi arXiv

  26. [35]

    Woli´nski

    M. Woli´nski. Polimorf, 2018. URL http://hdl.handle.net/11321/577. CLARIN-PL digital repository

    2018

  27. [2014]

    URL http://arxiv.org/abs/1312.6199

    arXiv

  28. [2018]

    URL https://aclanthology.org/2018

    Global Wordnet Association. URL https://aclanthology.org/2018. gwc-1.14

    2018

  29. [2024]

    URL https://aclanthology.org/2024.emnlp-main.1067

    2024

  30. [6202]

    doi: 10.18653/ V1/2020.EMNLP-MAIN.500

    Association for Computational Linguistics, 2020. doi: 10.18653/ V1/2020.EMNLP-MAIN.500. URL https://doi.org/10.18653/v1/2020. emnlp-main.500

    2020 doi

Pith tools

Reviewed August 15, 2026 · model on record in the stance chip above.