Pith. sign in

REVIEW 4 major objections 6 minor 60 references

DiscoSum: Discourse-aware News Summarization

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

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

Pith's one-line read Discourse-guided beam search can enforce a target summary structure without sacrificing factual alignment.

desk verdict The dataset is the real contribution; the decoding method is a recycled idea with a circular evaluation problem that needs fixing before the empirical claims can be trusted. read the letter →

arxiv 2506.06930 v1 pith:2LKGNNEK submitted 2025-06-07 cs.CL cs.AI

Alexander Spangher , Tenghao Huang , Jialiang Gu , Jiatong Shi , Muhao Chen This is my paper
classification cs.CLcs.AI
keywords newssummarizationdiscoursestructurediscourse-awarebeamsearchcontrollablegenerationsocialmediasummariesdatasetschema
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

DiscoSum claims that news summarization can be made structure-aware by decoding with a sentence-level beam search scored by a discourse labeler. The paper introduces a dataset of 20k news articles paired with more than 100k human-written summaries across Facebook, Instagram, Twitter, and newsletters, plus a five-label discourse schema (Introductory Elements, Contextual Details, Event Narration, Source Attribution, Engagement Directive). On both automatic and human evaluations, the beam-search variant of a fine-tuned language model matches target discourse label sequences better than vanilla or edit-based baselines, while staying factually aligned with the source. The central payoff is that a user-supplied label sequence, mirroring how newsrooms already write for different platforms, can be enforced at generation time without sacrificing content accuracy.

What carries the argument

The mechanism that carries the argument is a sentence-level discourse labeler $C(\cdot)$ trained on the DiscoSum train split, used as a reward inside a beam search: at each position $i$, the generation model proposes $k$ candidate sentences and the decoder keeps the one maximizing $C(s', t_i)$ against the target label $t_i$. The five-label schema is itself induced automatically by asking an LLM to label each summary sentence, embedding those labels with SBERT, and clustering with k-means. This same labeler is later used to compute the structural evaluation metrics, so it is simultaneously the oracle for generation and the ruler for measuring success.

What would settle it

Take summaries from outlets or platforms absent from DiscoSum, have independent annotators label each sentence with the five discourse roles, and measure agreement with the schema; if agreement is near chance or well below the reported inter-annotator $\kappa=0.615$, the labeler is not a trustworthy structural oracle and both the beam-search reward and the structural metrics lose their intended meaning.

Watch

Extended reading notes

Core claim

The paper's central claim is that structural fidelity and factual fidelity can be achieved together in news summarization. Its sentence-level beam search generates several candidate sentences for each position in a user-supplied discourse-label sequence, keeps the candidate whose predicted label matches the target, and thereby produces summaries that align with the designated structure while remaining factually consistent with the source. On automatic evaluation the beam-search variant of a fine-tuned 8-billion-parameter language model reaches a Match Score of 0.72 and the highest AlignScore (a factual-consistency metric) among the compared systems at 0.3890; human annotators give it a structural Match Score of 0.55 and rank it first with a mean reciprocal rank of 0.71. The authors interpret this as evidence that decoding-time guidance by a discourse labeler is a viable alternative to post-hoc editing or fine-tuning for structure-aware summarization.

Load-bearing premise

The load-bearing premise is that the automatically induced five-label discourse schema and the labeler trained on it faithfully capture how news summaries are organized across platforms and cultures, rather than imposing a dataset-specific taxonomy.

Editorial extensions

If this is right

  • If the result holds, newsrooms can supply a platform template (for example, LinkedIn's professional framing versus Instagram's engagement-first order) as a label sequence and the decoder will shape the summary accordingly.
  • Because the beam search improves AlignScore as well as structural Match Score, structure-aware decoding appears to help rather than hurt factual consistency relative to vanilla generation.
  • Larger beams monotonically improve structural alignment (rising LCS, falling Levenshtein distance), suggesting that computational cost can be traded for structural fidelity.
  • The DiscoSum dataset's one-to-many article-summary pairings enable future work on predicting a target structure from the source article and platform rather than requiring the user to supply the label sequence.
  • The approach may transfer to other structured summarization settings, such as legal, medical, or dialogue summarization, wherever a discourse labeler can be trained.

Reading between the lines

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

  • The authors leave the prediction of the target label sequence from the source article for future work; one testable extension is to learn a platform-specific editor model that outputs the label sequence, turning DiscoSum into a full end-to-end structure-aware summarization system.
  • Because the same labeler is used to score candidates during beam search and to evaluate the final summaries, the reported Match Scores partly measure self-consistency; an independent annotation study on final outputs would separate genuine structural control from metric circularity.
  • The dependence on LLM-generated candidates means the beam search inherits whatever stylistic diversity the base model's sampling has; a low-diversity sampling distribution could make all $k$ candidates near-identical and reduce the beam to a single path.
  • The five-label schema was validated by only two journalists, so cross-cultural generalizability is untested; a natural next step is to check whether the same clusters emerge when applying the schema-induction procedure to non-Western outlets.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

4 major / 6 minor

Summary. The paper introduces DiscoSum, a dataset of 20,811 news articles matched with multiple social-media and newsletter summaries, together with a five-label discourse schema automatically induced from LLM-generated labels and k-means clustering. The authors propose a sentence-level beam search decoding method, DiscoSum, that uses a trained discourse labeler C(·) to select sentences matching a user-specified target label sequence T. They report automatic and human evaluations claiming that the beam search variant of LLaMa-3-8B achieves the best structural alignment (automatic Match Score 0.72, human Match Score 0.55) while maintaining competitive factual alignment (AlignScore 0.3890). The paper also provides baseline comparisons with proprietary and open-source models.

Significance. DiscoSum is a potentially valuable resource for studying cross-platform news summarization, and the sentence-level beam search method offers a practical test-time approach to controllable structure. The human evaluation and human validation of the schema provide a partial de-circularization of the evaluation. However, the automatic structural metrics are computed with the same labeler used as the decoding reward, the target label sequences are randomly generated, and the dataset statistics contain internal inconsistencies. If these issues are resolved, the dataset and method could support future work in structure-aware summarization.

major comments (4)
  1. [§5.2 and Algorithm 1, line 8] The automatic structural metrics (Match Score, Levenshtein distance, LCS) are computed using the same labeler C(·) that serves as the beam search reward in Algorithm 1 (line 8, score←C(s′, ti)). Consequently, the beam search's automatic Match Score of 0.72 is substantially an artifact of optimizing the evaluator; it does not independently establish that the generated summaries have recognizable discourse organization. The paper should de-circularize the evaluation by using a separately trained or independently validated labeler for automatic scoring, or by making the human structural scores the primary evidence. In addition, the human evaluation should be expanded (more annotators, more summaries) and inter-annotator agreement should be reported.
  2. [§3.2, Tables 1 and 2] The abstract and Section 1 claim "20k news articles ... multiply paired with over 100k human-written summaries," but Table 2 reports only 45,195 matched article-summary pairs, while Table 1's post counts sum to 103,788. The split description states a 70%/20%/10% split into "14k/4k/2k article-summary pairs," which sums to 20k, not 45k. Please clarify whether the split is at the article level or the pair level, and reconcile these counts, as the current numbers make the dataset statistics unreliable.
  3. [§5.1] For each generation, the authors "randomly generate a list of structural tags" as the target label sequence. This means the evaluation measures how well the model matches an arbitrary label string rather than a discourse structure a reader or editor would consider natural. The paper's claim that beam search "consistently aligns more closely with the designated discourse label sequences" is thus about matching random sequences. Please evaluate on target sequences derived from the actual human-written summaries in the dataset (e.g., label sequences produced by applying the labeler to reference summaries) and report those results, to demonstrate practical relevance.
  4. [§5.4, Table 3] The claim that beam search "maintains competitive performance in surface-level metrics" and "achieves the highest AlignScore (0.3890)" is not well-supported. The beam search's ROUGE-L (42.98) is lower than vanilla LLaMa-3-8B (47.18), and the AlignScore advantage over proprietary models (0.3882–0.3888) is within 0.001, likely not statistically significant. Please report confidence intervals or significance tests, and discuss the tradeoff between structural alignment and content overlap more carefully.
minor comments (6)
  1. [Algorithm 1, line 8] The notation "score←C(s′, ti)" is ambiguous because C is defined as a sentence-level classifier; please clarify whether it is applied to the newly generated candidate sentence c or to the last sentence of s′, and then compared with ti.
  2. [§4.2] The training labels for the discourse labeler are not described. It is unclear whether the classifier is trained on the LLM-generated labels, the cluster assignments from k-means, or the manual annotations. Please specify the annotation protocol for the training data.
  3. [§5.5] The human preference evaluation reports MRR but no inter-annotator agreement or statistical significance. Also, if the same two annotators who validated the schema also performed the preference rankings, that could introduce bias; please clarify the annotator identities.
  4. [§5.2] The text says "we ask annotators to evaluate 100 summaries for each model," but Table 3 lacks human structural scores for O1; please clarify which models were included in the human evaluation and why O1 was excluded.
  5. [§4.1] The schema validity is supported by κ=0.615 between two journalists, which is moderate agreement. The paper should discuss the limitations of this validation, especially because the same schema underpins both the decoding reward and the evaluation metrics.
  6. [Appendix D.4] There are minor typos in the prompts, e.g., "{dis course_la bels}" appears in the definitions prompt; these should be fixed for clarity.

Circularity Check

1 steps flagged · score 6.0 of 10

Automatic structural scores are circular: beam search optimizes the same labeler C used to compute Match Score, so the 0.72 result is partly by construction; human de-circularization is partial and uses the same self-generated schema.

  1. self definitional [Section 4.3.2 (beam search); Section 5.2 (structural evaluation); Section 5.4 (main results)]
    "Section 4.3.2: “At each step i, the LLM generates several candidate sentences (forming a sentence-level “beam”), which are then evaluated by C(·). We choose the candidate that best matches the target label ti.” Section 5.2: “L=Labeler(s 1, s2, . . . , sn), si ∈S” … “where Labeler represents either the human annotator or the automated model designed to identify discourse structures.” Section 5.4: “The beam search variant of LLaMa-3-8B consistently aligns more closely with the designated discourse label sequences, evidenced by its superior Match Score and reduced Levenshtein Distance.”"

    Beam search evaluates each candidate sentence by C(·) and keeps the candidate that best matches ti; Section 5.2 defines the automatic structural metrics from L = Labeler(s1,...,sn), where Labeler is the same automated model. Match Score is exact position-wise agreement between L and T, so choosing sentences to maximize C(s, ti) is exactly choosing sentences to maximize the automatic Match Score. The reported 0.72 vs 0.21 gain therefore demonstrates that the decoder can satisfy its own classifier, not that the summaries have independently verified discourse structure.

full rationale

The automatic structural evaluation is the one load-bearing circular step. Beam search uses C(·) as its scoring function, and the automatic metrics use the same C(·) to produce L; hence the Match Score improvement is largely by construction. Human evaluation is independent of the model but not of the schema: the five labels were induced by LLM+clustering and checked by only two journalists (κ=0.615), and the target label strings are randomly generated, so the human structural scores test adherence to an internally defined taxonomy. The content-accuracy metrics (ROUGE-L, FactCC, AlignScore) are external and not circular, though the AlignScore edge is marginal and ROUGE-L is lower than vanilla; those are correctness not circularity issues. Self-citations in the labeler and matching steps are implementation citations and do not by themselves create a circular derivation. Score is 6 rather than 0 because the paper's headline structural claim rests on optimizing the evaluator, but the presence of human evaluation and external content metrics prevents total circularity.

Assumptions & free parameters 3 free parameters · 4 assumptions · 1 invented entities

The paper's central contribution rests on automatically induced categories: the five-label schema comes from k-means over LLM descriptions, the labeler is trained on those labels, and the target sequences are randomly generated. The dataset construction also leans on LLM-based cleaning and matching with audit-only validation. These choices are reasonable for a pilot, but much of the 'structure' is defined by the same machinery used to evaluate it.

free parameters (3)
  • Number of discourse clusters k = 5
    Selected from candidates k=7,13,23 based on human evaluation trials and inter-annotator agreement (κ=0.615); the schema is the foundation of the labeler and evaluation.
  • Beam size = 16
    Used for both vanilla best-of-16 sampling and beam search; larger beams improve structural scores, so the reported gains depend on this choice.
  • Minimum post length filter = 50 characters
    Posts shorter than 50 characters (about 30% of collected data) were dropped before matching, shaping the dataset composition.
assumptions (4)
  • domain assumption LLM-generated discourse labels reflect real discourse structure.
    Section 4.1 builds the schema from LLM labels on O(100k) sentences, then clusters them with k-means; human validation is limited to two journalists and a 500-sentence annotation set.
  • ad hoc to paper Randomly generated target label sequences are valid user inputs.
    Section 5.1 says the authors 'randomly generate a list of structural tags' to simulate user inputs; if these sequences are unnatural, structural metrics measure compliance to arbitrary templates rather than real platform discourse.
  • domain assumption SBERT plus GPT-4 article-summary matching is correct enough.
    Section 3.2 relies on a two-step rank-and-check matcher with manual audits cited as over 95% accuracy; matching errors propagate to all downstream pairs and experiments.
  • ad hoc to paper The same labeler can serve as both reward and evaluator.
    Algorithm 1 uses C(·) to score candidate sentences, and Section 5.2 uses the same C(·) to compute L versus T; this couples optimization and evaluation.
invented entities (1)
  • Five-label discourse schema (Introductory Elements, Contextual Details, Event Narration, Source Attribution, Engagement Directive) independent evidence
    purpose: Defines the structural roles that generated summary sentences are supposed to satisfy.
    The paper provides human annotation agreement (κ=0.615) and a labeler with 90.9% accuracy, so the schema has falsifiable handles, though it is constructed from LLM labels.

how reviews work

0 comments
Cite this review

Pith. "Pith review of DiscoSum: Discourse-aware News Summarization." pith.science (2026). https://pith.science/paper/2LKGNNEK

@misc{pith2026250606930,
  author       = {Pith},
  title        = {Pith review of: DiscoSum: Discourse-aware News Summarization},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/2LKGNNEK}},
  note         = {Machine review of arXiv:2506.06930}
}
read the original abstract

Recent advances in text summarization have predominantly leveraged large language models to generate concise summaries. However, language models often do not maintain long-term discourse structure, especially in news articles, where organizational flow significantly influences reader engagement. We introduce a novel approach to integrating discourse structure into summarization processes, focusing specifically on news articles across various media. We present a novel summarization dataset where news articles are summarized multiple times in different ways across different social media platforms (e.g. LinkedIn, Facebook, etc.). We develop a novel news discourse schema to describe summarization structures and a novel algorithm, DiscoSum, which employs beam search technique for structure-aware summarization, enabling the transformation of news stories to meet different stylistic and structural demands. Both human and automatic evaluation results demonstrate the efficacy of our approach in maintaining narrative fidelity and meeting structural requirements.

Figures

Figures reproduced from arXiv: 2506.06930 by the authors.

Figure 1
Figure 1. Comparative presentation of the Apollo 11 [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Mean Reciprocal Rank (MRR) scores from human preference evaluations of summary quality across three methods: Vanilla LLaMa-3-8B, Fine-tuned LLaMa-3-8B, and Beam Search LLaMa-3-8B. accuracy and fluency, where the ROUGE-L scores considerably lower than other methods. This de￾cline indicates that while the edit-based approach effectively molds the structure of the summaries, it may deviate significantly from the origin… view at source ↗
Figure 3
Figure 3. Levenshtein Distance and Longest Common Subsequence (LCS) scores as a function of beam size in structured summarization. The graph shows a general decrease in Levenshtein Distance and a gradual increase in LCS scores, indicating improved structural alignment with larger beam sizes. problems in the field of news summarization. Our DiscoSum dataset and corresponding evaluation metrics set the foundation for further ex… view at source ↗
Figures from the paper (1 more)
Figure 4
Figure 4. Figure 4: Confusion Matrix of Discourse Labeler. tags embedded in article URLs. This approach efficiently collects a large volume of summaries but is limited to a single summary per article that was intended for search engine or link preview con￾texts. In contrast, DiscoSum coll…

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

60 extracted references · 39 canonical work pages

  1. [1]

    Regina Barzilay and Kathleen R McKeown. 2005. Sentence fusion for multidocument news summarization. Computational Linguistics, 31(3):297--328

    work page 2005

  2. [2]

    Iz Beltagy, Matthew E Peters, and Arman Cohan. 2020. Longformer: The long-document transformer. arXiv preprint arXiv:2004.05150

    arXiv 2020

  3. [3]

    David Caswell. 2024. Telling every story: Characteristics of systematic reporting. In Journalism and Reporting Synergistic Effects of Climate Change, pages 266--283. Routledge

    work page 2024

  4. [4]

    David Caswell and Konstantin D \"o rr. 2018. Automated journalism 2.0: Event-driven narratives: From simple descriptions to real stories. Journalism practice, 12(4):477--496

    work page 2018

  5. [5]

    Danqi Chen, Jason Bolton, and Christopher D. Manning. 2016. A thorough examination of the cnn/daily mail reading comprehension task. In Association for Computational Linguistics (ACL)

    work page 2016

  6. [6]

    Jiaao Chen and Diyi Yang. 2023. Controllable conversation generation with conversation structures via diffusion models. In Findings of the Association for Computational Linguistics: ACL 2023, pages 7238--7251

    work page 2023

  7. [7]

    Arman Cohan, Franck Dernoncourt, Doo Soon Kim, Trung Bui, Seokhwan Kim, Walter Chang, and Nazli Goharian. 2018. https://doi.org/10.18653/v1/N18-2097 A discourse-aware attention model for abstractive summarization of long documents . In Proceedings of the 2018 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human...

    doi:10.18653/v1/n18-2097 2018

  8. [8]

    Zeyu Dai, Himanshu Taneja, and Ruihong Huang. 2018. Fine-grained structure-based news genre categorization. In Proceedings of the Workshop Events and Stories in the News, pages 17--23

    work page 2018

Show all 60 references
  1. [9]

    Michel De Montaigne. 1580. Essays. Self-published

  2. [10]

    Alexander Fabbri, Irene Li, Tianwei She, Suyi Li, and Dragomir Radev. 2019. https://doi.org/10.18653/v1/P19-1102 Multi-news: A large-scale multi-document summarization dataset and abstractive hierarchical model . In Proceedings of the 57th Annual Meeting of the Association for...

    2019 doi

  3. [11]

    Yaxin Fan, Feng Jiang, Peifeng Li, and Haizhou Li. 2024. Uncovering the potential of chatgpt for discourse analysis in dialogue: An empirical study. In Proceedings of the 2024 Joint International Conference on Computational Linguistics, Language Resources and Evaluation (LREC-...

    2024

  4. [12]

    Tanya Goyal, Junyi Jessy Li, and Greg Durrett. 2022. News summarization and evaluation in the era of gpt-3. arXiv preprint arXiv:2209.12356

    2022 arXiv

  5. [13]

    Matt Grenander, Yue Dong, Jackie Chi Kit Cheung, and Annie Louis. 2019. https://doi.org/10.18653/v1/D19-1620 Countering the effects of lead bias in news summarization via multi-stage training and auxiliary losses . In Proceedings of the 2019 Conference on Empirical Methods in ...

    2019 doi

  6. [14]

    Max Grusky, Mor Naaman, and Yoav Artzi. 2020. https://arxiv.org/abs/1804.11283 Newsroom: A dataset of 1.3 million summaries with diverse extractive strategies . Preprint, arXiv:1804.11283

    2020 arXiv

  7. [15]

    Junxian He, Wojciech Kry \'s ci \'n ski, Bryan McCann, Nazneen Rajani, and Caiming Xiong. 2020. Ctrlsum: Towards generic controllable text summarization. arXiv preprint arXiv:2012.04281

    2020 arXiv

  8. [16]

    Kai Hong, John M Conroy, Benoit Favre, Alex Kulesza, Hui Lin, Ani Nenkova, et al. 2014. A repository of state of the art and competitive baseline summaries for generic news summarization. In LREC, pages 1608--1616

    2014

  9. [17]

    Zhiting Hu, Zichao Yang, Xiaodan Liang, Ruslan Salakhutdinov, and Eric P Xing. 2017. Toward controlled generation of text. In International conference on machine learning, pages 1587--1596. PMLR

    2017

  10. [18]

    Tenghao Huang, Ehsan Qasemi, Bangzheng Li, He Wang, Faeze Brahman, Muhao Chen, and Snigdha Chaturvedi. 2023. https://api.semanticscholar.org/CorpusID:264436541 Affective and dynamic beam search for story generation . ArXiv, abs/2310.15079

    2023 arXiv

  11. [19]

    Feng Jiang, Weihao Liu, Xiaomin Chu, Peifeng Li, Qiaoming Zhu, and Haizhou Li. 2023. Advancing topic segmentation and outline generation in chinese texts: The paragraph-level topic representation, corpus, and benchmark. arXiv preprint arXiv:2305.14790

    2023 arXiv

  12. [20]

    Bente Kalsnes and Anders Olof Larsson. 2018. Understanding news sharing across social media: Detailing distribution on facebook and twitter. Journalism studies, 19(11):1669--1688

    2018

  13. [21]

    Wojciech Kryscinski, Bryan McCann, Caiming Xiong, and Richard Socher. 2020. https://doi.org/10.18653/v1/2020.emnlp-main.750 Evaluating the factual consistency of abstractive text summarization . In Proceedings of the 2020 Conference on Empirical Methods in Natural Language Pro...

    2020 doi

  14. [22]

    Levenshtein

    Vladimir I. Levenshtein. 1965. https://api.semanticscholar.org/CorpusID:60827152 Binary codes capable of correcting deletions, insertions, and reversals . Soviet physics. Doklady, 10:707--710

    1965

  15. [23]

    Haoyuan Li and Snigdha Chaturvedi. 2024. https://doi.org/10.18653/v1/2024.naacl-long.458 Rationale-based opinion summarization . In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (V...

    2024 doi

  16. [24]

    Chin-Yew Lin. 2004. https://aclanthology.org/W04-1013/ ROUGE : A package for automatic evaluation of summaries . In Text Summarization Branches Out, pages 74--81, Barcelona, Spain. Association for Computational Linguistics

    2004

  17. [25]

    Qin Liu, Fei Wang, Nan Xu, Tianyi Yan, Tao Meng, and Muhao Chen. 2024. https://api.semanticscholar.org/CorpusID:268681629 Monotonic paraphrasing improves generalization of language model prompting . ArXiv, abs/2403.16038

    2024 arXiv

  18. [26]

    Bruce T. Lowerre. 1976. https://api.semanticscholar.org/CorpusID:61409851 The harpy speech recognition system

    1976

  19. [27]

    Tao Meng, Sidi Lu, Nanyun Peng, and Kai-Wei Chang. 2022. https://api.semanticscholar.org/CorpusID:249192024 Controllable text generation with neurally-decomposed oracle . ArXiv, abs/2205.14219

    2022 arXiv

  20. [28]

    Does chatgpt measure up to discourse unit segmentation? a comparative analysis utilizing zero-shot custom prompts

    Kota Shamanth Ramanath Nayak. Does chatgpt measure up to discourse unit segmentation? a comparative analysis utilizing zero-shot custom prompts

  21. [29]

    Nguyen Minh Ngoc. 2022. Journalism and social media: The transformation of journalism in the age of social media and online news. European Journal of Social Sciences Studies, 7(6)

    2022

  22. [30]

    Romain Paulus, Caiming Xiong, and Richard Socher. 2017. https://arxiv.org/abs/1705.04304 A deep reinforced model for abstractive summarization . Preprint, arXiv:1705.04304

    2017 arXiv

  23. [31]

    Peper, Wenzhao Qiu, and Lu Wang

    Jospeh J. Peper, Wenzhao Qiu, and Lu Wang. 2024. Pelms: Pre-training for effective low-shot multi-document summarization. In Proceedings of the Conference of the North American Chapter of the Association for Computational Linguistics. Association for Computational Linguistics

    2024

  24. [32]

    Chau Pham, Alexander Hoyle, Simeng Sun, Philip Resnik, and Mohit Iyyer. 2024. Topicgpt: A prompt-based topic modeling framework. In Proceedings of the 2024 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies (V...

    2024

  25. [33]

    N Reimers. 2019. Sentence-bert: Sentence embeddings using siamese bert-networks. arXiv preprint arXiv:1908.10084

    2019 arXiv

  26. [34]

    Liu, and Christopher D

    Abigail See, Peter J. Liu, and Christopher D. Manning. 2017. https://arxiv.org/abs/1704.04368 Get to the point: Summarization with pointer-generator networks . Preprint, arXiv:1704.04368

    2017 arXiv

  27. [35]

    Chenhui Shen, Liying Cheng, Ran Zhou, Lidong Bing, Yang You, and Luo Si. 2022. Mred: A meta-review dataset for structure-controllable text generation. In Findings of the Association for Computational Linguistics: ACL 2022, pages 2521--2535

    2022

  28. [36]

    Tianxiao Shen, Tao Lei, Regina Barzilay, and Tommi Jaakkola. 2017. https://arxiv.org/abs/1705.09655 Style transfer from non-parallel text by cross-alignment . Preprint, arXiv:1705.09655

    2017 arXiv

  29. [37]

    Alexander Spangher, Tenghao Huang, Philippe Laban, and Nanyun Peng. 2025. https://aclanthology.org/2025.naacl-tutorial.1/ Creative planning with language models: Practice, evaluation and applications . In Proceedings of the 2025 Annual Conference of the Nations of the Americas...

    2025

  30. [38]

    Alexander Spangher, Jonathan May, Sz-Rung Shiang, and Lingjia Deng. 2021. Multitask semi-supervised learning for class-imbalanced discourse classification. In Proceedings of the 2021 conference on empirical methods in natural language processing, pages 498--517

    2021

  31. [39]

    Alexander Spangher, Yao Ming, Xinyu Hua, and Nanyun Peng. 2022 a . https://doi.org/10.18653/v1/2022.findings-emnlp.509 Sequentially controlled text generation . In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 6848--6866, Abu Dhabi, United Arab E...

    2022 doi

  32. [40]

    Alexander Spangher, Nanyun Peng, Sebastian Gehrmann, and Mark Dredze. 2024 a . Do llms plan like human writers? comparing journalist coverage of press releases with llms. In Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing, pages 21814--21828

    2024

  33. [41]

    Alexander Spangher, Xiang Ren, Jonathan May, and Nanyun Peng. 2022 b . https://doi.org/10.18653/v1/2022.naacl-main.10 N ews E dits: A news article revision dataset and a novel document-level reasoning challenge . In Proceedings of the 2022 Conference of the North American Chap...

    2022 doi

  34. [42]

    Alexander Spangher, Serdar Tumgoren, Ben Welsh, Nanyun Peng, Emilio Ferrara, and Jonathan May. 2024 b . https://doi.org/10.18653/v1/2024.acl-long.763 Tracking the newsworthiness of public documents . In Proceedings of the 62nd Annual Meeting of the Association for Computationa...

    2024 doi

  35. [43]

    Alexander Spangher, James Youn, Matt DeButts, Nanyun Peng, Emilio Ferrara, and Jonathan May. 2024 c . https://doi.org/10.18653/v1/2024.findings-emnlp.930 Explaining mixtures of sources in news articles . In Findings of the Association for Computational Linguistics: EMNLP 2024,...

    2024 doi

  36. [44]

    Yufei Tian, Tenghao Huang, Miri Liu, Derek Jiang, Alexander Spangher, Muhao Chen, Jonathan May, and Nanyun Peng. 2024. https://doi.org/10.18653/v1/2024.emnlp-main.978 Are large language models capable of generating human-level narratives? In Proceedings of the 2024 Conference ...

    2024 doi

  37. [45]

    Teun A Van Dijk. 1988. News as discourse. Routledge

    1988

  38. [46]

    Danqing Wang, Pengfei Liu, Yining Zheng, Xipeng Qiu, and Xuanjing Huang. 2020. https://doi.org/10.18653/v1/2020.acl-main.553 Heterogeneous graph neural networks for extractive document summarization . In Proceedings of the 58th Annual Meeting of the Association for Computation...

    2020 doi

  39. [47]

    Lu Wang and Claire Cardie. 2013. https://aclanthology.org/P13-1137/ Domain-independent abstract generation for focused meeting summarization . In Proceedings of the 51st Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1395--1405, ...

    2013

  40. [48]

    Lu Wang and Wang Ling. 2016. https://doi.org/10.18653/v1/N16-1007 Neural network-based abstract generation for opinions and arguments . In Proceedings of the 2016 Conference of the North A merican Chapter of the Association for Computational Linguistics: Human Language Technol...

    2016 doi

  41. [49]

    Ben Welsh, Naitian Zhou, Arda Kaz, Michael Vu, and Alexander Spangher. 2024. https://arxiv.org/abs/2501.00004 Newshomepages: Homepage layouts capture information prioritization decisions . Preprint, arXiv:2501.00004

    2024 arXiv

  42. [50]

    Kevin Yang and Dan Klein. 2021. https://doi.org/10.18653/v1/2021.naacl-main.276 FUDGE : Controlled text generation with future discriminators . In Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language ...

    2021 doi

  43. [51]

    Shuai Yang, Zhangyang Wang, Zhaowen Wang, Ning Xu, Jiaying Liu, and Zongming Guo. 2019. Controllable artistic text style transfer via shape-matching gan. In Proceedings of the IEEE/CVF International Conference on Computer Vision, pages 4442--4451

    2019

  44. [52]

    Cohen, and John D

    ChengXiang Zhai, William W. Cohen, and John D. Lafferty. 2003. https://api.semanticscholar.org/CorpusID:52855966 Beyond independent relevance: methods and evaluation metrics for subtopic retrieval . Proceedings of the 26th annual international ACM SIGIR conference on Research ...

    2003

  45. [53]

    Jingqing Zhang, Yao Zhao, Mohammad Saleh, and Peter Liu. 2020. Pegasus: Pre-training with extracted gap-sentences for abstractive summarization. In International conference on machine learning, pages 11328--11339. PMLR

    2020

  46. [54]

    Hashimoto

    Tianyi Zhang, Faisal Ladhak, Esin Durmus, Percy Liang, Kathleen McKeown, and Tatsunori B. Hashimoto. 2024. https://doi.org/10.1162/tacl_a_00632 Benchmarking large language models for news summarization . Transactions of the Association for Computational Linguistics, 12:39--57

    2024 doi

  47. [55]

    Chao Zhao, Faeze Brahman, Tenghao Huang, and Snigdha Chaturvedi. 2022 a . https://api.semanticscholar.org/CorpusID:249097857 Revisiting generative commonsense reasoning: A pre-ordering approach . ArXiv, abs/2205.13183

    2022 arXiv

  48. [56]

    Chao Zhao, Tenghao Huang, Somnath Basu Roy Chowdhury, Muthu Kumar Chandrasekaran, Kathleen McKeown, and Snigdha Chaturvedi. 2022 b . https://doi.org/10.18653/v1/2022.findings-acl.51 Read top news first: A document reordering approach for multi-document news summarization . In ...

    2022 doi

  49. [57]

    Jihao Zhao, Zhiyuan Ji, Yuchen Feng, Pengnian Qi, Simin Niu, Bo Tang, Feiyu Xiong, and Zhiyu Li. 2024. Meta-chunking: Learning efficient text segmentation via logical perception. arXiv preprint arXiv:2410.12788

    2024 arXiv

  50. [58]

    Yang Zhong and Diane Litman. 2023. Strong--structure controllable legal opinion summary generation. In Findings of the Association for Computational Linguistics: IJCNLP-AACL 2023 (Findings), pages 431--448

    2023

  51. [59]

    online" 'onlinestring :=

    ENTRY address archivePrefix author booktitle chapter edition editor eid eprint eprinttype howpublished institution journal key month note number organization pages publisher school series title type volume year doi pubmed url lastchecked label extra.label sort.label short.list...

  52. [60]

    write newline

    " write newline "" before.all 'output.state := FUNCTION n.dashify 't := "" t empty not t #1 #1 substring "-" = t #1 #2 substring "--" = not "--" * t #2 global.max substring 't := t #1 #1 substring "-" = "-" * t #2 global.max substring 't := while if t #1 #1 substring * t #2 gl...

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.