pith. sign in

arxiv: 2605.18007 · v1 · pith:WFWVRFI4new · submitted 2026-05-18 · 💻 cs.CL

Semantic Reranking at Inference Time for Hard Examples in Rhetorical Role Labeling

Anas Belfathi , Nicolas Hernandez , Laura Monceaux , Warren Bonnard , Richard Dufour This is my paper

Pith reviewed 2026-05-20 11:23 UTC · model grok-4.3

classification 💻 cs.CL
keywords rhetorical role labelingsemantic rerankinginference-time methodshard exampleslabel semanticslow-confidence predictionsdocument understanding
0
0 comments X

The pith

Semantic reranking of label names at inference time improves accuracy on low-confidence predictions in rhetorical role labeling.

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

Rhetorical role labeling assigns functional roles to sentences in documents used in legal, medical, and scientific work. Language models handle most cases well but falter on hard examples where their prediction confidence is low. The paper tests whether reranking the model's candidate roles by how closely their names match the sentence in a contrastively learned semantic space can fix those errors. The reranking runs only at inference time and leaves the original model unchanged. Experiments across eight domain-specific datasets and seven models show that this step raises performance specifically on the uncertain instances.

Core claim

The central claim is that automatically detecting low-confidence outputs and then reranking label candidates according to similarity between the input sentence and contrastively trained embeddings of the role names themselves recovers accuracy on hard cases. This produces an average gain of 9.15 macro-F1 points on the flagged examples across eight datasets and seven language models of both encoder and causal types, with no retraining required. The work also introduces manual hardness annotations and reports moderate agreement between model and human views of difficulty.

What carries the argument

RISE, the inference-time procedure that flags low-confidence predictions and reranks outputs by semantic similarity to contrastively learned representations of the rhetorical role label names.

If this is right

  • Hard examples identified by low prediction confidence receive consistent accuracy gains from the semantic reranking step.
  • The gains appear across eight domain-specific datasets and seven different language models without any retraining.
  • Treating labels by their semantic content rather than as arbitrary identifiers supplies useful signal for refining uncertain predictions.
  • Model-identified hard examples align moderately with human judgments of difficulty.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same reranking idea could extend to other labeling tasks where class names carry descriptive meaning, such as certain entity or relation classification settings.
  • Deploying this approach might lower the cost of adapting systems to new specialized domains by avoiding full retraining cycles.
  • If low-confidence flags also correlate with human-perceived difficulty, the method could help prioritize data collection or review efforts.

Load-bearing premise

Low model confidence reliably marks the examples where semantic reranking of label names will produce meaningful gains, and the contrastively learned label representations transfer effectively across domains without further adaptation.

What would settle it

Applying the reranking step to low-confidence examples from an additional held-out RRL dataset and observing no increase, or a decrease, in macro-F1 score on those examples would falsify the central performance claim.

Figures

Figures reproduced from arXiv: 2605.18007 by Anas Belfathi, Laura Monceaux, Nicolas Hernandez, Richard Dufour, Warren Bonnard.

Figure 1
Figure 1. Figure 1: Top-k oracle performance reveals prediction ambiguity on the SCOTUSRF dataset [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Overview of the RISE framework. A language model (encoder-based or causal) is first used as a discriminative classifier to produce logits for each input sentence. RISE operates at inference time (gray area) by automatically identifying hard cases based on model confidence. For these instances, label semantics are exploited by reranking logits based on semantic distances derived from contrastively learned t… view at source ↗
Figure 3
Figure 3. Figure 3: Marginal effect of automatic hard-example de [PITH_FULL_IMAGE:figures/full_fig_p007_3.png] view at source ↗
Figure 4
Figure 4. Figure 4: Human vs. Model Hardness: Level-wise correspondence. and efficient solution for resolving semantic ambi￾guity, motivating the design choices behind RISE. 6 Hardness from a Human Perspective: An Empirical Analysis To complement model-centric analyses of predic￾tion difficulty, we examine hardness from a hu￾man perspective through an annotation study on SCOTUSRF, which features strong semantic over￾lap betwe… view at source ↗
Figure 5
Figure 5. Figure 5: Distribution of instance difficulty levels ac [PITH_FULL_IMAGE:figures/full_fig_p014_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Distribution of human-identified explanatory [PITH_FULL_IMAGE:figures/full_fig_p014_6.png] view at source ↗
Figure 8
Figure 8. Figure 8: Financial and time cost comparison. LLM-RR [PITH_FULL_IMAGE:figures/full_fig_p015_8.png] view at source ↗
Figure 9
Figure 9. Figure 9: Data-efficiency comparison between the base [PITH_FULL_IMAGE:figures/full_fig_p017_9.png] view at source ↗
read the original abstract

Rhetorical Role Labeling (RRL) assigns a functional role to each sentence in a document and is widely used in legal, medical, and scientific domains. While language models (LMs) achieve strong average performance, they remain unreliable on hard examples, where prediction confidence is low. Existing approaches typically handle uncertainty implicitly and treat labels as discrete identifiers, overlooking the semantic information encoded in label names. We introduce RISE, an inference-time semantic reranking framework that leverages label semantics to refine predictions on hard instances. RISE automatically identifies low-confidence predictions and reranks model outputs using contrastively learned label representations, without retraining or modifying the underlying model. Experiments on eight domain-specific RRL datasets with seven LMs, including encoder-based and causal architectures, show an average gain of +9.15 macro-F1 points on hard examples. For explainability, we further propose manual hardness annotations to study difficulty from both model and human perspectives, revealing a moderate agreement with Cohen's kappa = 0.40.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 2 minor

Summary. The manuscript introduces RISE, an inference-time semantic reranking framework for Rhetorical Role Labeling (RRL) that identifies low-confidence predictions and refines them using contrastively learned label representations, without retraining the base LM. Experiments across eight domain-specific RRL datasets and seven LMs (encoder and causal) report an average +9.15 macro-F1 gain on hard examples; the work also introduces manual hardness annotations and reports Cohen's kappa = 0.40 between model confidence and human judgments of difficulty.

Significance. If the gains prove robust, the approach offers a practical, training-free route to improving reliability on uncertain instances in specialized RRL tasks. The focus on label-name semantics and inference-time intervention addresses a real deployment pain point in legal, medical, and scientific document processing.

major comments (2)
  1. [Experiments] Experiments section: the reported average +9.15 macro-F1 gain on hard examples is given without per-dataset breakdowns, confidence-threshold definition, statistical significance tests, or an ablation that isolates the contribution of semantic reranking versus simple confidence thresholding.
  2. [Method] Method section: contrastive learning of label representations is described as enabling cross-domain transfer, yet no ablation with a frozen label encoder or explicit cross-domain protocol is provided; this leaves the transfer assumption (critical for the central claim) untested given possible semantic shifts such as “Facts” versus “Results”.
minor comments (2)
  1. [Abstract / Method] The abstract and method should clarify the exact set of seven LMs and any architecture-specific differences in how reranking is applied.
  2. [Explainability / Experiments] The manual hardness annotation protocol is introduced for explainability; additional details on annotator guidelines and full inter-annotator agreement statistics beyond the single kappa value would strengthen the human-model comparison.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for their constructive comments. We address each major comment below and indicate planned revisions to the manuscript.

read point-by-point responses

  1. Referee: [Experiments] Experiments section: the reported average +9.15 macro-F1 gain on hard examples is given without per-dataset breakdowns, confidence-threshold definition, statistical significance tests, or an ablation that isolates the contribution of semantic reranking versus simple confidence thresholding.

    Authors: We agree that these elements would improve clarity and rigor. In the revised manuscript we will add per-dataset macro-F1 breakdowns, explicitly state the confidence threshold used to identify hard examples, report statistical significance tests for the observed gains, and include an ablation comparing semantic reranking against a simple confidence-threshold baseline. This will isolate the contribution of the label-semantics component. revision: yes

  2. Referee: [Method] Method section: contrastive learning of label representations is described as enabling cross-domain transfer, yet no ablation with a frozen label encoder or explicit cross-domain protocol is provided; this leaves the transfer assumption (critical for the central claim) untested given possible semantic shifts such as “Facts” versus “Results”.

    Authors: While the contrastive objective is intended to capture semantic similarities that support transfer, we acknowledge the absence of direct ablations leaves the claim under-tested. We will add an ablation with a frozen label encoder and explicit cross-domain protocols (training the reranker on subsets of datasets and evaluating on held-out domains) to examine robustness to shifts such as “Facts” versus “Results”. revision: yes

Circularity Check

0 steps flagged

No circularity: empirical gains measured on independent datasets

full rationale

The paper introduces an inference-time reranking method (RISE) that identifies low-confidence predictions and refines them using contrastively learned label representations. All reported results consist of empirical macro-F1 improvements (+9.15 average on hard examples) measured across eight distinct domain-specific RRL datasets and seven separate LMs. No equations, fitted parameters, or self-citations are invoked to derive the performance numbers; the gains are obtained by direct evaluation on held-out data rather than by construction from the method's own inputs or prior self-referential claims. The central premise therefore remains externally falsifiable and does not reduce to a renaming or self-definition of the observed outcomes.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Abstract provides insufficient detail to enumerate specific free parameters or axioms; standard assumptions of contrastive learning and confidence thresholding are implicit but not quantified.

pith-pipeline@v0.9.0 · 5715 in / 1043 out tokens · 38893 ms · 2026-05-20T11:23:38.238713+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

159 extracted references · 159 canonical work pages · 11 internal anchors

  1. [1]

    Neural Networks for Joint Sentence Classification in Medical Paper Abstracts

    Dernoncourt, Franck and Lee, Ji Young and Szolovits, Peter. Neural Networks for Joint Sentence Classification in Medical Paper Abstracts. Proceedings of the 15th Conference of the E uropean Chapter of the Association for Computational Linguistics: Volume 2, Short Papers. 2017

    work page 2017

  2. [2]

    BMC bioinformatics , volume=

    Automatic classification of sentences to support evidence based medicine , author=. BMC bioinformatics , volume=. 2011 , organization=

    work page 2011

  3. [3]

    Pretrained Language Models for Sequential Sentence Classification

    Cohan, Arman and Beltagy, Iz and King, Daniel and Dalvi, Bhavana and Weld, Dan. Pretrained Language Models for Sequential Sentence Classification. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019. doi:10.18653/v1/D19-1383

    work page doi:10.18653/v1/d19-1383 2019

  4. [4]

    A deep learning classifier for sentence classification in biomedical and computer science abstracts , year =

    Gon. A deep learning classifier for sentence classification in biomedical and computer science abstracts , year =. Neural Comput. Appl. , month = jun, pages =. doi:10.1007/s00521-019-04334-2 , abstract =

    work page doi:10.1007/s00521-019-04334-2

  5. [5]

    Emerald 110k: A Multidisciplinary Dataset for Abstract Sentence Classification

    Stead, Connor and Smith, Stephen and Busch, Peter and Vatanasakdakul, Savanid. Emerald 110k: A Multidisciplinary Dataset for Abstract Sentence Classification. Proceedings of the 17th Annual Workshop of the Australasian Language Technology Association. 2019

    work page 2019

  6. [6]

    Rhetorical Move Detection in E nglish Abstracts: Multi-label Sentence Classifiers and their Annotated Corpora

    Dayrell, Carmen and Candido Jr., Arnaldo and Lima, Gabriel and Machado Jr., Danilo and Copestake, Ann and Feltrim, Val \'e ria and Tagnin, Stella and Aluisio, Sandra. Rhetorical Move Detection in E nglish Abstracts: Multi-label Sentence Classifiers and their Annotated Corpora. Proceedings of the Eighth International Conference on Language Resources and Ev...

    work page 2012

  7. [7]

    Artificial Intelligence and Law , pages=

    DeepRhole: deep learning for rhetorical role labeling of sentences in legal case documents , author=. Artificial Intelligence and Law , pages=. 2023 , publisher=

    work page 2023

  8. [8]

    Semantic Segmentation of Legal Documents via Rhetorical Roles

    Malik, Vijit and Sanjay, Rishabh and Guha, Shouvik Kumar and Hazarika, Angshuman and Nigam, Shubham Kumar and Bhattacharya, Arnab and Modi, Ashutosh. Semantic Segmentation of Legal Documents via Rhetorical Roles. Proceedings of the Natural Legal Language Processing Workshop 2022. 2022. doi:10.18653/v1/2022.nllp-1.13

    work page doi:10.18653/v1/2022.nllp-1.13 2022

  9. [9]

    Corpus for Automatic Structuring of Legal Documents

    Kalamkar, Prathamesh and Tiwari, Aman and Agarwal, Astha and Karn, Saurabh and Gupta, Smita and Raghavan, Vivek and Modi, Ashutosh. Corpus for Automatic Structuring of Legal Documents. Proceedings of the Thirteenth Language Resources and Evaluation Conference. 2022

    work page 2022

  10. [10]

    International journal of medical informatics , volume=

    Using argumentation to extract key sentences from biomedical abstracts , author=. International journal of medical informatics , volume=. 2007 , publisher=

    work page 2007

  11. [11]

    AMIA annual symposium proceedings , volume=

    Categorization of sentence types in medical abstracts , author=. AMIA annual symposium proceedings , volume=

    work page

  12. [12]

    Generative Content Models for Structural Analysis of Medical Abstracts

    Lin, Jimmy and Karakos, Damianos and Demner-Fushman, Dina and Khudanpur, Sanjeev. Generative Content Models for Structural Analysis of Medical Abstracts. Proceedings of the HLT - NAACL B io NLP Workshop on Linking Natural Language and Biology. 2006

    work page 2006

  13. [13]

    Proceedings of the 22nd ACM/IEEE Joint Conference on Digital Libraries , pages=

    Cross-domain multi-task learning for sequential sentence classification in research papers , author=. Proceedings of the 22nd ACM/IEEE Joint Conference on Digital Libraries , pages=

    work page

  14. [14]

    Hierarchical Neural Networks for Sequential Sentence Classification in Medical Scientific Abstracts

    Jin, Di and Szolovits, Peter. Hierarchical Neural Networks for Sequential Sentence Classification in Medical Scientific Abstracts. Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing. 2018. doi:10.18653/v1/D18-1349

    work page doi:10.18653/v1/d18-1349 2018

  15. [15]

    International Journal on Digital Libraries , volume=

    Sequential sentence classification in research papers using cross-domain multi-task learning , author=. International Journal on Digital Libraries , volume=. 2024 , publisher=

    work page 2024

  16. [16]

    Advances in neural information processing systems , volume=

    Distributed representations of words and phrases and their compositionality , author=. Advances in neural information processing systems , volume=

    work page

  17. [17]

    Understanding of a convolutional neural network , year=

    Albawi, Saad and Mohammed, Tareq Abed and Al-Zawi, Saad , booktitle=. Understanding of a convolutional neural network , year=

    work page

  18. [18]

    A Span-based Dynamic Local Attention Model for Sequential Sentence Classification

    Shang, Xichen and Ma, Qianli and Lin, Zhenxi and Yan, Jiangyue and Chen, Zipeng. A Span-based Dynamic Local Attention Model for Sequential Sentence Classification. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 2: Short Papers). 2021...

    work page doi:10.18653/v1/2021.acl-short.26 2021

  19. [19]

    Sequential Span Classification with Neural Semi- M arkov CRF s for Biomedical Abstracts

    Yamada, Kosuke and Hirao, Tsutomu and Sasano, Ryohei and Takeda, Koichi and Nagata, Masaaki. Sequential Span Classification with Neural Semi- M arkov CRF s for Biomedical Abstracts. Findings of the Association for Computational Linguistics: EMNLP 2020. 2020. doi:10.18653/v1/2020.findings-emnlp.77

    work page doi:10.18653/v1/2020.findings-emnlp.77 2020

  20. [20]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    XProtoNet: diagnosis in chest radiography with global and local explanations , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  21. [21]

    Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    Neural prototype trees for interpretable fine-grained image recognition , author=. Proceedings of the IEEE/CVF conference on computer vision and pattern recognition , pages=

    work page

  22. [22]

    Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

    Multimodal prototypical networks for few-shot learning , author=. Proceedings of the IEEE/CVF winter conference on applications of computer vision , pages=

    work page

  23. [23]

    Joint European Conference on Machine Learning and Knowledge Discovery in Databases , pages=

    Prototypical convolutional neural network for a phrase-based explanation of sentiment classification , author=. Joint European Conference on Machine Learning and Knowledge Discovery in Databases , pages=. 2021 , organization=

    work page 2021

  24. [24]

    arXiv preprint arXiv:2310.15743 , year=

    RAPL: A relation-aware prototype learning approach for few-shot document-level relation extraction , author=. arXiv preprint arXiv:2310.15743 , year=

    work page arXiv

  25. [25]

    SimCSE: Simple Contrastive Learning of Sentence Embeddings

    Simcse: Simple contrastive learning of sentence embeddings , author=. arXiv preprint arXiv:2104.08821 , year=

    work page internal anchor Pith review Pith/arXiv arXiv

  26. [26]

    Mind Your Neighbours: Leveraging Analogous Instances for Rhetorical Role Labeling for Legal Documents

    T.y.s.s., Santosh and Sarwat, Hassan and Abdou, Ahmed Mohamed Abdelaal and Grabmair, Matthias. Mind Your Neighbours: Leveraging Analogous Instances for Rhetorical Role Labeling for Legal Documents. Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-COLING 2024). 2024

    work page 2024

  27. [27]

    BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding

    Devlin, Jacob and Chang, Ming-Wei and Lee, Kenton and Toutanova, Kristina. BERT : Pre-training of Deep Bidirectional Transformers for Language Understanding. Proceedings of the 2019 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technologies, Volume 1 (Long and Short Papers). 2019. doi:10.18653/v...

    work page doi:10.18653/v1/n19-1423 2019

  28. [28]

    Neural computation , volume=

    Long short-term memory , author=. Neural computation , volume=. 1997 , publisher=

    work page 1997

  29. [29]

    Proceedings of the 2016 conference of the North American chapter of the association for computational linguistics: human language technologies , pages=

    Hierarchical attention networks for document classification , author=. Proceedings of the 2016 conference of the North American chapter of the association for computational linguistics: human language technologies , pages=

    work page 2016

  30. [30]

    2024 , eprint=

    Nomic Embed: Training a Reproducible Long Context Text Embedder , author=. 2024 , eprint=

    work page 2024

  31. [31]

    LEGAL - BERT : The Muppets straight out of Law School

    Chalkidis, Ilias and Fergadiotis, Manos and Malakasiotis, Prodromos and Aletras, Nikolaos and Androutsopoulos, Ion. LEGAL - BERT : The Muppets straight out of Law School. Findings of the Association for Computational Linguistics: EMNLP 2020. 2020. doi:10.18653/v1/2020.findings-emnlp.261

    work page doi:10.18653/v1/2020.findings-emnlp.261 2020

  32. [32]

    S ci BERT : A pretrained language model for scientific text

    Beltagy, Iz and Lo, Kyle and Cohan, Arman. S ci BERT : A Pretrained Language Model for Scientific Text. Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (EMNLP-IJCNLP). 2019. doi:10.18653/v1/D19-1371

    work page doi:10.18653/v1/d19-1371 2019

  33. [33]

    Journal of pragmatics , volume=

    Anticipative interlocutive dialogism: sequential patterns and linguistic markers in French , author=. Journal of pragmatics , volume=. 2016 , publisher=

    work page 2016

  34. [34]

    Electronics , volume=

    The k-means algorithm: A comprehensive survey and performance evaluation , author=. Electronics , volume=. 2020 , publisher=

    work page 2020

  35. [35]

    Robust Text Classification: Analyzing Prototype-Based Networks

    Sourati, Zhivar and Deshpande, Darshan Girish and Ilievski, Filip and Gashteovski, Kiril and Saralajew, Sascha. Robust Text Classification: Analyzing Prototype-Based Networks. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.745

    work page doi:10.18653/v1/2024.findings-emnlp.745 2024

  36. [36]

    Journal of english for academic purposes , volume=

    Evaluation of Cohen's kappa and other measures of inter-rater agreement for genre analysis and other nominal data , author=. Journal of english for academic purposes , volume=. 2021 , publisher=

    work page 2021

  37. [37]

    This Patient Looks Like That Patient: Prototypical Networks for Interpretable Diagnosis Prediction from Clinical Text

    van Aken, Betty and Papaioannou, Jens-Michalis and Naik, Marcel and Eleftheriadis, Georgios and Nejdl, Wolfgang and Gers, Felix and Loeser, Alexander. This Patient Looks Like That Patient: Prototypical Networks for Interpretable Diagnosis Prediction from Clinical Text. Proceedings of the 2nd Conference of the Asia-Pacific Chapter of the Association for Co...

    work page doi:10.18653/v1/2022.aacl-main.14 2022

  38. [38]

    Prototypical Networks for Few-shot Learning , url =

    Snell, Jake and Swersky, Kevin and Zemel, Richard , booktitle =. Prototypical Networks for Few-shot Learning , url =

    work page

  39. [39]

    This Looks Like That: Deep Learning for Interpretable Image Recognition , url =

    Chen, Chaofan and Li, Oscar and Tao, Daniel and Barnett, Alina and Rudin, Cynthia and Su, Jonathan K , booktitle =. This Looks Like That: Deep Learning for Interpretable Image Recognition , url =

    work page

  40. [40]

    Hyperspherical Prototype Networks , url =

    Mettes, Pascal and van der Pol, Elise and Snoek, Cees , booktitle =. Hyperspherical Prototype Networks , url =

    work page

  41. [41]

    Robust Classification with Convolutional Prototype Learning , year=

    Yang, Hong-Ming and Zhang, Xu-Yao and Yin, Fei and Liu, Cheng-Lin , booktitle=. Robust Classification with Convolutional Prototype Learning , year=

    work page

  42. [42]

    Robust and explainable identification of logical fallacies in natural language arguments , journal =

    Zhivar Sourati and Vishnu Priya. Robust and explainable identification of logical fallacies in natural language arguments , journal =. 2023 , issn =. doi:https://doi.org/10.1016/j.knosys.2023.110418 , url =

    work page doi:10.1016/j.knosys.2023.110418 2023

  43. [43]

    P roto TE x: Explaining Model Decisions with Prototype Tensors

    Das, Anubrata and Gupta, Chitrank and Kovatchev, Venelin and Lease, Matthew and Li, Junyi Jessy. P roto TE x: Explaining Model Decisions with Prototype Tensors. Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2022. doi:10.18653/v1/2022.acl-long.213

    work page doi:10.18653/v1/2022.acl-long.213 2022

  44. [44]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    Hybrid Attention-Based Prototypical Networks for Noisy Few-Shot Relation Classification , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2019 , month=. doi:10.1609/aaai.v33i01.33016407 , abstractNote=

    work page doi:10.1609/aaai.v33i01.33016407 2019

  45. [45]

    Supervised Contrastive Learning , url =

    Khosla, Prannay and Teterwak, Piotr and Wang, Chen and Sarna, Aaron and Tian, Yonglong and Isola, Phillip and Maschinot, Aaron and Liu, Ce and Krishnan, Dilip , booktitle =. Supervised Contrastive Learning , url =

    work page

  46. [46]

    Improving Few-Shot Relation Classification by Prototypical Representation Learning with Definition Text

    Zhenzhen, Li and Zhang, Yuyang and Nie, Jian-Yun and Li, Dongsheng. Improving Few-Shot Relation Classification by Prototypical Representation Learning with Definition Text. Findings of the Association for Computational Linguistics: NAACL 2022. 2022. doi:10.18653/v1/2022.findings-naacl.34

    work page doi:10.18653/v1/2022.findings-naacl.34 2022

  47. [47]

    Speculative rag: Enhancing retrieval augmented generation through drafting

    Speculative rag: Enhancing retrieval augmented generation through drafting , author=. arXiv preprint arXiv:2407.08223 , year=

    work page arXiv

  48. [48]

    Processing

    Mamakas, Dimitris and Tsotsi, Petros and Androutsopoulos, Ion and Chalkidis, Ilias. Processing Long Legal Documents with Pre-trained Transformers: Modding L egal BERT and Longformer. Proceedings of the Natural Legal Language Processing Workshop 2022. 2022. doi:10.18653/v1/2022.nllp-1.11

    work page doi:10.18653/v1/2022.nllp-1.11 2022

  49. [49]

    Matryoshka-Adaptor: Unsupervised and Supervised Tuning for Smaller Embedding Dimensions

    Yoon, Jinsung and Sinha, Rajarishi and Arik, Sercan O and Pfister, Tomas. Matryoshka-Adaptor: Unsupervised and Supervised Tuning for Smaller Embedding Dimensions. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024. doi:10.18653/v1/2024.emnlp-main.576

    work page doi:10.18653/v1/2024.emnlp-main.576 2024

  50. [50]

    L ux E mbedder: A Cross-Lingual Approach to Enhanced L uxembourgish Sentence Embeddings

    Philippy, Fred and Guo, Siwen and Klein, Jacques and Bissyande, Tegawende. L ux E mbedder: A Cross-Lingual Approach to Enhanced L uxembourgish Sentence Embeddings. Proceedings of the 31st International Conference on Computational Linguistics. 2025

    work page 2025

  51. [51]

    L ong E mbed: Extending Embedding Models for Long Context Retrieval

    Zhu, Dawei and Wang, Liang and Yang, Nan and Song, Yifan and Wu, Wenhao and Wei, Furu and Li, Sujian. L ong E mbed: Extending Embedding Models for Long Context Retrieval. Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing. 2024. doi:10.18653/v1/2024.emnlp-main.47

    work page doi:10.18653/v1/2024.emnlp-main.47 2024

  52. [52]

    N ext L evel BERT : Masked Language Modeling with Higher-Level Representations for Long Documents

    Czinczoll, Tamara and H. N ext L evel BERT : Masked Language Modeling with Higher-Level Representations for Long Documents. Proceedings of the 62nd Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2024. doi:10.18653/v1/2024.acl-long.256

    work page doi:10.18653/v1/2024.acl-long.256 2024

  53. [53]

    Matos-Carvalho and Nuno Fachada , keywords =

    Alina Petukhova and João P. Matos-Carvalho and Nuno Fachada , keywords =. Text clustering with large language model embeddings , journal =. 2025 , issn =. doi:https://doi.org/10.1016/j.ijcce.2024.11.004 , url =

    work page doi:10.1016/j.ijcce.2024.11.004 2025

  54. [54]

    2024 , MONTH = Mar, DOI =

    Lavissi. 2024 , MONTH = Mar, DOI =

    work page 2024

  55. [55]

    , PUBLISHER =

    , AUTHOR =. , PUBLISHER =. Forthcoming , MONTH =. doi:, KEYWORDS =

    work page

  56. [56]

    L egal S eg: Unlocking the Structure of I ndian Legal Judgments Through Rhetorical Role Classification

    Nigam, Shubham Kumar and Dubey, Tanmay and Sharma, Govind and Shallum, Noel and Ghosh, Kripabandhu and Bhattacharya, Arnab. L egal S eg: Unlocking the Structure of I ndian Legal Judgments Through Rhetorical Role Classification. Findings of the Association for Computational Linguistics: NAACL 2025. 2025

    work page 2025

  57. [57]

    Proceedings of the 6th Workshop on Automated Semantic Analysis of Information in Legal Text , year=

    Enhancing Pre-Trained Language Models with Sentence Position Embeddings for Rhetorical Roles Recognition in Legal Opinions , author=. Proceedings of the 6th Workshop on Automated Semantic Analysis of Information in Legal Text , year=

    work page

  58. [58]

    Automatic Rhetorical Roles Classification for Legal Documents using LEGAL-TransformerOverBERT , booktitle =

    Gabriele Marino and Daniele Licari and Praveen Bushipaka and Giovanni Comand. Automatic Rhetorical Roles Classification for Legal Documents using LEGAL-TransformerOverBERT , booktitle =. 2023 , timestamp =

    work page 2023

  59. [59]

    Legal knowledge and information systems , pages=

    Identification of rhetorical roles of sentences in indian legal judgments , author=. Legal knowledge and information systems , pages=. 2019 , publisher=

    work page 2019

  60. [60]

    , author=

    Rhetorical Role Labelling for Legal Judgements Using ROBERTA. , author=. FIRE (Working Notes) , pages=

    work page

  61. [61]

    S em E val-2023 Task 6: L egal E val - Understanding Legal Texts

    Modi, Ashutosh and Kalamkar, Prathamesh and Karn, Saurabh and Tiwari, Aman and Joshi, Abhinav and Tanikella, Sai Kiran and Guha, Shouvik Kumar and Malhan, Sachin and Raghavan, Vivek. S em E val-2023 Task 6: L egal E val - Understanding Legal Texts. Proceedings of the 17th International Workshop on Semantic Evaluation (SemEval-2023). 2023. doi:10.18653/v1/...

    work page doi:10.18653/v1/2023.semeval-1.318 2023

  62. [62]

    Enhancing Pre-Trained Language Representations with Rich Knowledge for Machine Reading Comprehension

    Yang, An and Wang, Quan and Liu, Jing and Liu, Kai and Lyu, Yajuan and Wu, Hua and She, Qiaoqiao and Li, Sujian. Enhancing Pre-Trained Language Representations with Rich Knowledge for Machine Reading Comprehension. Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics. 2019. doi:10.18653/v1/P19-1226

    work page doi:10.18653/v1/p19-1226 2019

  63. [63]

    2025 , MONTH = Jun, KEYWORDS =

    Belfathi, Anas and Gallina, Ygor and Hernandez, Nicolas and Monceaux, Laura and Dufour, Richard , URL =. 2025 , MONTH = Jun, KEYWORDS =

    work page 2025

  64. [64]

    MP roto: Multi-Prototype Network with Denoised Optimal Transport for Distantly Supervised Named Entity Recognition

    Wu, Shuhui and Shen, Yongliang and Tan, Zeqi and Ren, Wenqi and Guo, Jietian and Pu, Shiliang and Lu, Weiming. MP roto: Multi-Prototype Network with Denoised Optimal Transport for Distantly Supervised Named Entity Recognition. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023. doi:10.18653/v1/2023.emnlp-main.145

    work page doi:10.18653/v1/2023.emnlp-main.145 2023

  65. [65]

    PRAM : An End-to-end Prototype-based Representation Alignment Model for Zero-resource Cross-lingual Named Entity Recognition

    Huang, Yucheng and Liu, Wenqiang and Zhang, Xianli and Lang, Jun and Gong, Tieliang and Li, Chen. PRAM : An End-to-end Prototype-based Representation Alignment Model for Zero-resource Cross-lingual Named Entity Recognition. Findings of the Association for Computational Linguistics: ACL 2023. 2023. doi:10.18653/v1/2023.findings-acl.201

    work page doi:10.18653/v1/2023.findings-acl.201 2023

  66. [66]

    Supervised Prototypical Contrastive Learning for Emotion Recognition in Conversation

    Song, Xiaohui and Huang, Longtao and Xue, Hui and Hu, Songlin. Supervised Prototypical Contrastive Learning for Emotion Recognition in Conversation. Proceedings of the 2022 Conference on Empirical Methods in Natural Language Processing. 2022. doi:10.18653/v1/2022.emnlp-main.347

    work page doi:10.18653/v1/2022.emnlp-main.347 2022

  67. [67]

    Consistent Prototype Learning for Few-Shot Continual Relation Extraction

    Chen, Xiudi and Wu, Hui and Shi, Xiaodong. Consistent Prototype Learning for Few-Shot Continual Relation Extraction. Proceedings of the 61st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers). 2023. doi:10.18653/v1/2023.acl-long.409

    work page doi:10.18653/v1/2023.acl-long.409 2023

  68. [68]

    Proceedings of the AAAI Conference on Artificial Intelligence , author=

    ProtGNN: Towards Self-Explaining Graph Neural Networks , volume=. Proceedings of the AAAI Conference on Artificial Intelligence , author=. 2022 , month=. doi:10.1609/aaai.v36i8.20898 , abstractNote=

    work page doi:10.1609/aaai.v36i8.20898 2022

  69. [69]

    Learning from Incomplete and Inaccurate Supervision

    Ming, Yao and Xu, Panpan and Qu, Huamin and Ren, Liu , title =. Proceedings of the 25th ACM SIGKDD International Conference on Knowledge Discovery & Data Mining , pages =. 2019 , isbn =. doi:10.1145/3292500.3330908 , abstract =

    work page doi:10.1145/3292500.3330908 2019

  70. [70]

    Segment-Level and Category-Oriented Network for Knowledge-Based Referring Expression Comprehension

    Bu, Yuqi and Wu, Xin and Li, Liuwu and Cai, Yi and Liu, Qiong and Huang, Qingbao. Segment-Level and Category-Oriented Network for Knowledge-Based Referring Expression Comprehension. Findings of the Association for Computational Linguistics: ACL 2023. 2023. doi:10.18653/v1/2023.findings-acl.557

    work page doi:10.18653/v1/2023.findings-acl.557 2023

  71. [71]

    Enhancing Content Preservation in Text Style Transfer Using Reverse Attention and Conditional Layer Normalization

    Lee, Dongkyu and Tian, Zhiliang and Xue, Lanqing and Zhang, Nevin L. Enhancing Content Preservation in Text Style Transfer Using Reverse Attention and Conditional Layer Normalization. Proceedings of the 59th Annual Meeting of the Association for Computational Linguistics and the 11th International Joint Conference on Natural Language Processing (Volume 1:...

    work page doi:10.18653/v1/2021.acl-long.8 2021

  72. [72]

    A Deeper (Autoregressive) Approach to Non-Convergent Discourse Parsing

    Tsur, Oren and Tulpan, Yoav. A Deeper (Autoregressive) Approach to Non-Convergent Discourse Parsing. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023. doi:10.18653/v1/2023.emnlp-main.796

    work page doi:10.18653/v1/2023.emnlp-main.796 2023

  73. [73]

    Visually Grounded Continual Language Learning with Selective Specialization

    Ahrens, Kyra and Bengtson, Lennart and Hee Lee, Jae and Wermter, Stefan. Visually Grounded Continual Language Learning with Selective Specialization. Findings of the Association for Computational Linguistics: EMNLP 2023. 2023. doi:10.18653/v1/2023.findings-emnlp.469

    work page doi:10.18653/v1/2023.findings-emnlp.469 2023

  74. [74]

    A Coarse-to-Fine Prototype Learning Approach for Multi-Label Few-Shot Intent Detection

    Zhang, Xiaotong and Li, Xinyi and Zhang, Feng and Wei, Zhiyi and Liu, Junfeng and Liu, Han. A Coarse-to-Fine Prototype Learning Approach for Multi-Label Few-Shot Intent Detection. Findings of the Association for Computational Linguistics: EMNLP 2024. 2024. doi:10.18653/v1/2024.findings-emnlp.140

    work page doi:10.18653/v1/2024.findings-emnlp.140 2024

  75. [75]

    Revisiting the Knowledge Injection Frameworks

    Fu, Peng and Zhang, Yiming and Wang, Haobo and Qiu, Weikang and Zhao, Junbo. Revisiting the Knowledge Injection Frameworks. Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing. 2023. doi:10.18653/v1/2023.emnlp-main.677

    work page doi:10.18653/v1/2023.emnlp-main.677 2023

  76. [76]

    2024 5th International Conference on Innovative Trends in Information Technology (ICITIIT) , pages=

    Impact of Rhetorical Roles in Abstractive Legal Document Summarization , author=. 2024 5th International Conference on Innovative Trends in Information Technology (ICITIIT) , pages=. 2024 , organization=

    work page 2024

  77. [77]

    Scientometrics , volume=

    Bibliometric-enhanced information retrieval: a novel deep feature engineering approach for algorithm searching from full-text publications , author=. Scientometrics , volume=. 2019 , publisher=

    work page 2019

  78. [78]

    , author=

    Automatic Classification of Rhetorical Roles for Sentences: Comparing Rule-Based Scripts with Machine Learning. , author=. ASAIL@ ICAIL , volume=

    work page

  79. [79]

    2023 , pages =

    Artificial Intelligence and Law , author =. 2023 , pages =. doi:10.1007/s10506-021-09304-5 , abstract =

    work page doi:10.1007/s10506-021-09304-5 2023

  80. [80]

    and Ravindran, B

    Saravanan, M. and Ravindran, B. and Raman, S. Automatic Identification of Rhetorical Roles using Conditional Random Fields for Legal Document Summarization. Proceedings of the Third International Joint Conference on Natural Language Processing: Volume- I. 2008

    work page 2008

Showing first 80 references.