An Evaluation of Large Language Models on Text Summarization Tasks Using Prompt Engineering Techniques

Amr Mohamed AbdelAziz; Taysir Hassan A. Soliman; Walid Mohamed Aly

REVIEW 3 major objections 6 minor 78 references

An Evaluation of Large Language Models on Text Summarization Tasks Using Prompt Engineering Techniques

T0 review · 3 major / 6 minor · reviewed 2026-08-06 · deepseek-v4-flash

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

Pith's one-line read This paper evaluates six open-source large language models on news, dialogue, and scientific summarization using zero-shot and in-context prompts, and claims that a sentence-based chunking strategy measurably improves long…

desk verdict A useful but unpolished benchmark of six open LLMs on four summarization datasets, whose headline chunking claim needs statistical and control fixes before it can be trusted. read the letter →

arxiv 2507.05123 v1 pith:N5BNOXZG submitted 2025-07-07 cs.CL cs.AI

Walid Mohamed Aly , Taysir Hassan A. Soliman , Amr Mohamed AbdelAziz This is my paper

classification cs.CLcs.AI

keywords largelanguagemodelstextsummarizationpromptengineeringzero-shotlearningin-contextlong-documentchunkingROUGEBERTScore

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

The paper asks how far prompt engineering alone, with no task-specific training, can take large language models at text summarization, and answers by evaluating six open-source models—Llama-2 7B, 13B, and 70B chat variants, Mistral-7B, Gemma-7B, and Mixtral-8x7B—on four datasets spanning news (CNN/Daily Mail, NewsRoom), dialogue (SAMSum), and scientific articles (ArXiv). Using zero-shot and in-context learning prompts and scoring with ROUGE (word overlap) and BERTScore (semantic similarity), the authors find that LLMs are competitive summarizers on news and dialogue. Their most central claim is about long scientific documents: a sentence-based chunking strategy, which summarizes each chunk of a paper and then re-summarizes the combined output, measurably raises quality for models whose context windows force heavy truncation, with Llama-2-70B-chat and Mixtral-8x7B showing the clearest gains. If that claim is right, practitioners get a training-free recipe for extending short-context models to long inputs, along with a comparative picture of how model scale, prompt wording, and the number of in-context examples shift summary quality and inference cost.

What carries the argument

The load-bearing mechanism is the paper's sentence-based chunking pipeline. Each ArXiv paper is segmented into sentences, grouped into chunks, each chunk is summarized independently with a zero-shot prompt, and the intermediate summaries are concatenated and reprocessed into one unified summary; this converts a context-window overflow into a series of in-window subproblems, with the second pass recombining partial content. Supporting that mechanism is the evaluation apparatus: dataset-specific zero-shot prompts and in-context learning with one, three, five, and seven demonstrations, with outputs cleaned and scored by ROUGE-1/2/L (lexical overlap) and BERTScore F1 (semantic similarity), and temperature fixed at 0.1 after testing 0.1, 0.5, and 0.9.

What would settle it

Recompute the ArXiv chunking comparison with statistical significance testing: bootstrap confidence intervals over the 2,000-document sample, multiple decoding seeds, or a second reference-summary set. If Llama-2-70B-chat's ROUGE-1 rise from 35.48 to 38.12 and Mixtral-8x7B-Instruct-v0.1's rise from 37.43 to 40.02 (BERTScore 84.73 to 87.16) fall inside the noise band, or if a human panel does not rank the chunked summaries as better, the chunking conclusion collapses. A cheaper test is to check whether the post-chunking summaries contain content from the portions of the papers that the pre-chunking trimmed baseline had cut.

Watch

Extended reading notes

Core claim

The central claim is that prompt-engineered LLMs, without task-specific training, are already credible summarizers on news and dialogue, and that their clearest weakness—long scientific documents—can be substantially repaired by chunking rather than by a larger context window. On the ArXiv dataset, where the baseline condition trims each paper to fit the model's context limit, the sentence-based pipeline raises Llama-2-70B-chat's ROUGE-1 from 35.48 to 38.12 and its BERTScore F1 from 84.26 to 86.19, and Mixtral-8x7B-Instruct-v0.1's from 37.43 to 40.02 and from 84.73 to 87.16, under zero-shot prompting. The paper also reports that the gains are not universal: Llama-2-13b-chat and Gemma-7b-it show no consistent improvement, which the authors attribute to architecture and input sensitivity, and the smallest model tested, Llama-2-7b-chat, posts the highest ArXiv ROUGE-1 both before and after chunking (49.74 and 50.30). Around the chunking result, the paper argues that more in-context demonstrations generally help, that temperature 0.1 is the best setting, and that LLMs trade peak scores against specialized baselines such as PEGASUS and SICK in exchange for working across domains with no training.

Load-bearing premise

The load-bearing premise is that ROUGE-1 and BERTScore F1 against the datasets' reference summaries are adequate stand-ins for summary quality, and that differences of one to three points matter even though no significance testing is reported; the paper itself concedes the metrics' limitations in its Section 5.5 discussion of metric usability.

Editorial extensions

If this is right

On the ArXiv long-document task, the chunking pipeline lifts Llama-2-70B-chat from ROUGE-1 35.48 and BERTScore 84.26 to 38.12 and 86.19, and Mixtral-8x7B-Instruct-v0.1 from 37.43 and 84.73 to 40.02 and 87.16, under zero-shot prompting.
Increasing the number of in-context demonstrations generally improves both metrics, with several models peaking at seven examples; on SAMSum, Mixtral-8x7B-Instruct-v0.1 reaches ROUGE-1 45.86 in the 7-shot condition.
Temperature 0.1 maximizes ROUGE scores on every dataset, so the paper fixes it for all final evaluations, implying deterministic decoding is the preferred setting for summarization.
Inference time grows with parameter count and input length, yet Mixtral-8x7B's sparse mixture-of-experts design reduces latency relative to Llama-2-70B-chat, exposing a concrete quality-versus-cost trade-off.
Against specialized models, the best LLMs trail PEGASUS on news and SICK on dialogue, but require no dataset-specific training, placing their value in broad multi-domain coverage rather than peak score.

Reading between the lines

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

My inference: because the pre-chunking baseline already trims each ArXiv paper to the model's context length, the chunking gain should track how much content the trim discards; a direct test is to split the 2,000-document evaluation by source-paper length and check whether the gains concentrate in the longest papers.
My inference: Mixtral-8x7B-Instruct-v0.1 already has a 32K-token context window, so its large chunking gains (ROUGE-1 +2.59, BERTScore +2.43) suggest the pipeline does more than squeeze input into the window—it hands the model a distilled summary of the paper to attend over, an explanation the paper's numbers support but do not state.
My inference: if these results replicate under human evaluation, the implied deployment recipe is that a small model plus a chunk-then-aggregate loop can substitute for a large-context model on long documents, which matters for hardware-constrained settings the paper does not discuss.

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.

Desk Editor's Note

read the letter

The paper is a workmanlike benchmark study: six open-source LLMs on four summarization datasets, with prompt sweeps, ICL shot counts, inference-time numbers, and a chunking experiment. The descriptive tables are genuinely useful for practitioners who want a rough model-by-dataset comparison for local summarization. It also ships code and generated summaries, which is real reproducibility evidence, and the Section 5.5 discussion of ROUGE/BERTScore limitations is more honest than most papers in this space.

The soft spots are real and concentrated in the chunking claim. First, Table 15's "BERTScore (F-1)" column before chunking appears to report BERTScore Precision from Table 8, not F1. For Llama-2-70b-chat, Table 8 prompt#2 gives F1 of 81.67, but Table 15 lists 84.26, which is the Precision from that same cell. The same pattern holds for gemma-7b-it. Either the column is mislabeled or the values were copied from the wrong row; either way, the before/after comparison is unreliable on that axis. Second, no variance or significance testing is reported anywhere. With 2K ArXiv samples, a 0.4–2.6-point ROUGE shift can be sampling noise, so the abstract's "improves significantly" is not supported. Third, the chunked pipeline uses a different prompt structure and a multi-stage synthesis step, so it is not a controlled comparison of chunking alone. The model-specific pattern (gains for Llama-2-70B and Mixtral, none for Llama-2-13B and gemma) gets a hand-wavy "architecture and input sensitivity" explanation, which reads as post hoc.

There are also smaller issues: temperature 0.1 was chosen because it maximized ROUGE on the test sets, and several headline comparisons use each model's best prompt, which inflates observed differences. The SOTA comparison tables have caption mix-ups (Table 13 says CNN/DM and SAMSum but includes a NEWSROOM row; Table 14 says NEWSROOM and ArXiv but includes SAMSum), and the text has scattered typos. None of these destroy the descriptive value of the benchmark, but they make the paper feel unfinished.

Who is this for? Someone choosing an open LLM for a summarization pipeline and wanting a rough cross-domain map, or a researcher looking for a baseline table to cite. It is not a methodological contribution.

My recommendation: send it to peer review, but require fixes to Table 15 (labeling and ideally variance), a significance test or at least a clear statement that differences are informal, a controlled chunking comparison, and softened claims if the control issue is not resolved. With those, it becomes a citable benchmark. As it stands, the chunking conclusion is not established.

Referee Report

3 major / 6 minor

Summary. The paper evaluates six open-source LLMs (Llama-2-7B/13B/70B-chat, gemma-7b-it, Mistral-7B-Instruct-v0.1, Mixtral-8x7B-Instruct-v0.1) on four summarization datasets (CNN/DailyMail, NewsRoom, SAMSum, ArXiv) using zero-shot and in-context learning prompts, with ROUGE and BERTScore metrics, inference-time measurements, and a sentence-based chunking strategy for long ArXiv documents. The headline claim is that chunking significantly improves long-document summarization, supported by Table 15 where, e.g., Llama-2-70B-chat ROUGE-1 rises from 35.48 to 38.12 and BERTScore from 84.26 to 86.19. The paper also compares LLMs with earlier specialized summarization models and discusses prompt sensitivity, summary lengths, and qualitative examples.

Significance. If the chunking claim were established, the paper would offer a practical, zero-training recipe for long-document summarization with constrained-context open-source models, plus useful cross-domain, cross-model comparisons and efficiency data. The study's strengths are its breadth of datasets and models, the release of prompts and code (GitHub link in Section 4.1.1), the inference-time analysis (Section 5.7), and the candid qualitative error analysis (Section 5.6). However, the central quantitative claim currently rests on post hoc test-set selections and on score differences whose statistical significance is not quantified; the paper's value therefore depends on whether the requested revisions can make the comparisons clean and the gains credible.

major comments (3)

[Section 4.3, Table 15] The claim that chunking 'improves significantly' is not supported by any statistical evidence. The table reports point estimates only: no standard deviations, confidence intervals, or significance tests are provided for the ROUGE-1 and BERTScore differences. With a 2K-sample ArXiv test set, the observed changes (e.g., Llama-2-7B-chat: 49.74 to 50.30 in ROUGE-1; Mistral-7B: 85.61 to 85.80 in BERTScore) are well within plausible sampling noise, and the model-specific pattern (improvements for Llama-2-70B and Mixtral, declines for Llama-2-13B and gemma) is exactly what one would expect from noise. The manuscript should report per-sample score distributions, paired bootstrap or Wilcoxon tests across the 2K documents, and effect sizes. This is load-bearing because Section 5.2 and the abstract both assert a 'significant' improvement based on this table.
[Section 4.3 vs. Section 3.4] The before/after comparison in Table 15 is confounded. The 'before' numbers (e.g., prompt#2 in Table 8) come from the ZSL setting with a trimmed input and a specific prompt ('Summarize the main points and findings of the scientific paper'), while the 'after' numbers use a multi-stage pipeline: sentence-based chunking, independent per-chunk summarization, and a final synthesis step. The method text (Section 3.4) does not specify whether the chunked pipeline used the same base prompt, the same temperature (0.1 per Section 3.6), the same max-new-tokens, the same decoding parameters, or the same post-processing of generated summaries. Without a controlled experiment in which the only change is the input segmentation, the gains in Table 15 cannot be attributed to chunking rather than to prompt structure, context length, or synthesis instructions. The authors should run a clean comparison, or explicitly report all hyperparameters for both conditions.
[Section 3.6 and Section 4.1.1] The evaluation protocol contains post hoc selections on the test data. Temperature 0.1 was chosen because it 'consistently maximizes ROUGE metrics across datasets' (Section 3.6, Table 3) using the same test sets on which all final results are reported. Likewise, for each model and dataset the best-performing prompt (Figure 2, and the 'highest result' selection noted in Footnote 2) is used for subsequent comparisons, including the chunking experiments in Table 15 and the state-of-the-art comparisons in Tables 13 and 14. This test-set-based tuning can inflate reported scores and makes model rankings and the chunking comparison optimistic. The authors should either fix a single prompt and temperature a priori, or use a held-out validation split for prompt/temperature selection and report results on a separate test portion.

minor comments (6)

[Abstract and Section 3.4] The abstract contains an incomplete sentence: 'For Long documents, introduce a sentence-based chunking strategy' lacks a subject; the same phrasing appears in Section 3.4. Please revise to 'we introduce' or similar.
[Table 3] In Table 3, the NewsRoom ROUGE-2 value at temperature 0.9 appears as '80.01', which is almost certainly a typo for '8.01'; please verify all entries for consistency.
[Tables 13 and 14] The captions of Tables 13 and 14 appear to be swapped: Table 13 contains CNN/DM and NEWSROOM rows but its caption mentions 'CNN/DM and SAMSum', and Table 14 contains SAMSum and ArXiv rows but its caption mentions 'NEWSROOM and ArXiv'. Please correct the captions.
[References] The PEGASUS model is cited as [39] in the text, but reference [39] is 'Calibrating Sequence likelihood Improves Conditional Language Generation' by Zhao et al.; the actual PEGASUS paper appears to be reference [40]. Please correct the citation mapping so that PEGASUS is attributed to Zhang et al. (2020).
[Section 2] The manuscript writes 'BLUE' in several places (e.g., 'BLUE[43]' when describing prior work); the standard acronym is BLEU. Please correct these typos.
[Section 3.7 / Table 17] The inference-time analysis in Section 5.7 reports average times but no variance or hardware details beyond 'two NVIDIA A100 GPUs'. Indicating the number of runs, batch size, and whether times include prompt processing would make these numbers more reproducible.

Circularity Check

0 steps flagged · score 0.0 of 10

No significant circularity: the paper is an empirical benchmark with no derivation chain; its post-hoc temperature and prompt selections affect evaluation validity, not circularity.

full rationale

This paper is an empirical evaluation study, not a derivation or prediction framework, so most circularity patterns do not apply. The central claims—that LLMs perform competitively on news and dialogue summarization and that a sentence-chunking strategy improves long-document summarization for some models—are measured outcomes, not quantities derived from fitted inputs. The chunking results in Table 15 compare before/after scores on the ArXiv test set, and the improvement is presented as an empirical observation rather than a forced consequence of a definition. The paper does not define its inputs in terms of its outputs: ROUGE and BERTScore are standard external metrics applied to reference summaries, and the chunking pipeline is not tuned to maximize the reported deltas. The main caveats in the manuscript are evaluation-quality concerns, not circularity: temperature 0.1 was selected because it 'consistently maximizes ROUGE metrics across datasets' (Section 3.6), and some figures show only the best prompt per setting ('only the prompts with the highest result are shown', Section 4.1.1). These choices can inflate or distort reported scores, but they do not make the reported measurements equivalent to their own inputs by construction. Likewise, the absence of statistical significance testing and the confound between chunking and prompt/context changes are threats to the reliability of the causal claim about chunking, but they are not circularity. No load-bearing self-citation or imported uniqueness theorem appears, and the chunking method is explicitly attributed to prior work rather than presented as a novel derivation. Therefore, the circularity score is 0; methodological weaknesses should be assessed under correctness risk, not circularity.

Assumptions & free parameters 3 free parameters · 3 assumptions · 0 invented entities

This is an empirical benchmark with no formal derivation. The main free choices are the temperature (0.1) selected on test data, the best-prompt selection used in headline comparisons, and the undisclosed chunking hyperparameters. The reliability of the central claim depends on the assumptions that ROUGE/BERTScore are valid proxies, that the sampled test documents are representative, and that trimming long documents to context length preserves enough content for summarization.

free parameters (3)

temperature = 0.1
Selected after testing 0.1, 0.5, 0.9 and choosing the value that maximizes ROUGE on the test sets (Table 3); this is a post hoc choice on evaluation data.
prompt choice per model/dataset = best prompt (e.g., prompt#2 for ArXiv)
In figures and SOTA comparisons, only the highest-scoring prompt per model and dataset is reported (Section 4.1.1, note under Figure 2), effectively selecting the best of several hand-written prompts.
chunk size / number of chunks = not reported
The chunking strategy in Section 3.4 segments documents into sentences using NLTK, but no chunk size, overlap, or merging prompt is specified; the choice is being tuned by the authors and not disclosed.

assumptions (3)

domain assumption ROUGE and BERTScore are valid proxies for summary quality
The paper evaluates all models with ROUGE and BERTScore and interprets differences as quality differences (Sections 3.5 and 5.5); if these metrics are unreliable, the rankings could change.
domain assumption Test-set samples are representative
2K random samples per dataset (except SAMSum full test) are used; the paper reports no variance or confidence intervals for these samples (Section 3.1).
domain assumption Trimming long documents to context length preserves the information needed for summarization
For the ArXiv experiments, documents are trimmed to fit context windows; the paper states this may lose contextual nuances (Section 3.4), so results on ArXiv are conditioned on this assumption.

how reviews work

0 comments

Cite this review

Pith. "Pith review of An Evaluation of Large Language Models on Text Summarization Tasks Using Prompt Engineering Techniques." pith.science (2026). https://pith.science/paper/N5BNOXZG

@misc{pith2026250705123,
  author       = {Pith},
  title        = {Pith review of: An Evaluation of Large Language Models on Text Summarization Tasks Using Prompt Engineering Techniques},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/N5BNOXZG}},
  note         = {Machine review of arXiv:2507.05123}
}

read the original abstract

Large Language Models (LLMs) continue to advance natural language processing with their ability to generate human-like text across a range of tasks. Despite the remarkable success of LLMs in Natural Language Processing (NLP), their performance in text summarization across various domains and datasets has not been comprehensively evaluated. At the same time, the ability to summarize text effectively without relying on extensive training data has become a crucial bottleneck. To address these issues, we present a systematic evaluation of six LLMs across four datasets: CNN/Daily Mail and NewsRoom (news), SAMSum (dialog), and ArXiv (scientific). By leveraging prompt engineering techniques including zero-shot and in-context learning, our study evaluates the performance using the ROUGE and BERTScore metrics. In addition, a detailed analysis of inference times is conducted to better understand the trade-off between summarization quality and computational efficiency. For Long documents, introduce a sentence-based chunking strategy that enables LLMs with shorter context windows to summarize extended inputs in multiple stages. The findings reveal that while LLMs perform competitively on news and dialog tasks, their performance on long scientific documents improves significantly when aided by chunking strategies. In addition, notable performance variations were observed based on model parameters, dataset properties, and prompt design. These results offer actionable insights into how different LLMs behave across task types, contributing to ongoing research in efficient, instruction-based NLP systems.

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

78 extracted references · 49 canonical work pages

[1]

Expert Systems with Applications 165, 113679 (2021) https://doi.org/10.1016/j.eswa.2020.113679

El-Kassas, W.S., Salama, C.R., Rafea, A.A., Mohamed, H.K.: Automatic text summarization: A comprehensive survey. Expert Systems with Applications 165, 113679 (2021) https://doi.org/10.1016/j.eswa.2020.113679

work page arXiv 2021
[2]

In: Riezler, S., Goldberg, Y

Nallapati, R., Zhou, B., Santos, C., Gul¸ cehre, C ¸ ., Xiang, B.: Abstractive text summarization using sequence-to-sequence RNNs and beyond. In: Riezler, S., Goldberg, Y. (eds.) Proceedings of the 20th SIGNLL Conference on Computa- tional Natural Language Learning, pp. 280–290. Association for Computational Linguistics, Berlin, Germany (2016). https://do...

doi:10.18653/v1/k16-1028 2016
[3]

Journal of King Saud University - Computer and Information Sciences 34(4), 1029–1046 (2022) https://doi.org/10.1016/j.jksuci.2020.05.006 32

Widyassari, A.P., Rustad, S., Shidik, G.F., Noersasongko, E., Syukur, A., Affandy, A., Setiadi, D.R.I.M.: Review of automatic text summarization tech- niques & methods. Journal of King Saud University - Computer and Information Sciences 34(4), 1029–1046 (2022) https://doi.org/10.1016/j.jksuci.2020.05.006 32

work page doi:10.1016/j.jksuci.2020.05.006 2022
[4]

IOP Conference Series: Materials Science and Engineering 1099(1), 012042 (2021) https://doi.org/10.1088/1757-899X/1099/ 1/012042

Dehru, V., Tiwari, P.K., Aggarwal, G., Joshi, B., Kartik, P.: Text summariza- tion techniques and applications. IOP Conference Series: Materials Science and Engineering 1099(1), 012042 (2021) https://doi.org/10.1088/1757-899X/1099/ 1/012042

work page doi:10.1088/1757-899x/1099/ 2021
[5]

Information Processing & Management 45(1), 20–34 (2009) https://doi.org/10.1016/j.ipm.2008.06.002

Lee, J.-H., Park, S., Ahn, C.-M., Kim, D.: Automatic generic document summa- rization based on non-negative matrix factorization. Information Processing & Management 45(1), 20–34 (2009) https://doi.org/10.1016/j.ipm.2008.06.002

work page doi:10.1016/j.ipm.2008.06.002 2009
[6]

In: 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp

Gigioli, P., Sagar, N., Rao, A., Voyles, J.: Domain-aware abstractive text sum- marization for medical documents. In: 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 2338–2343 (2018). https://doi.org/ 10.1109/BIBM.2018.8621539

work page arXiv 2018
[7]

Neural Computing and Applications 35, 18553–18570 (2023) https://doi.org/10.1007/s00521-023-08680-0

Li, S., Xu, J.: Hiermds: a hierarchical multi-document summarization model with global–local document dependencies. Neural Computing and Applications 35, 18553–18570 (2023) https://doi.org/10.1007/s00521-023-08680-0

work page doi:10.1007/s00521-023-08680-0 2023
[8]

In: Muresan, S., Nakov, P., Villavicencio, A

Xiao, W., Beltagy, I., Carenini, G., Cohan, A.: PRIMERA: Pyramid- based masked sentence pre-training for multi-document summarization. In: Muresan, S., Nakov, P., Villavicencio, A. (eds.) Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 5245–5263. Association for Computational Linguis- ti...

doi:10.18653/v1/2022.acl-long.360 2022

Show all 78 references

[9]

Neural Computing and Applications (2023) https://doi.org/10

Rafi, S., Das, R.: Topic-guided abstractive multimodal summarization with mul- timodal output. Neural Computing and Applications (2023) https://doi.org/10. 1007/s00521-023-08821-5

2023
[10]

In: 2017 International Conference on Computer, Communication and Signal Processing (ICCCSP), pp

Moratanch, N., Chitrakala, S.: A survey on extractive text summarization. In: 2017 International Conference on Computer, Communication and Signal Processing (ICCCSP), pp. 1–6 (2017). https://doi.org/10.1109/ICCCSP.2017. 7944061

2017 doi
[11]

Neural Computing and Applications 32, 11491–11503 (2020) https://doi.org/10.1007/s00521-019-04638-3

Diao, Y., Lin, H., Yang, L., Fan, X., Chu, Y., Wu, D., Zhang, D., Xu, K.: Crhasum: extractive text summarization with contextualized-representation hierarchical- attention summarization network. Neural Computing and Applications 32, 11491–11503 (2020) https://doi.org/10.1007/s...

2020 doi
[12]

In: 2016 International Conference on Circuit, Power and Computing Technologies (ICCPCT), pp

Moratanch, N., Chitrakala, S.: A survey on abstractive text summarization. In: 2016 International Conference on Circuit, Power and Computing Technologies (ICCPCT), pp. 1–7 (2016). https://doi.org/10.1109/ICCPCT.2016.7530193

2016
[13]

Kim, S., Joo, S.J., Chae, H., Kim, C., Hwang, S.-w., Yeo, J.: Mind the gap! injecting commonsense knowledge for abstractive dialogue summarization. In: Calzolari, N., Huang, C.-R., Kim, H., Pustejovsky, J., Wanner, L., Choi, K.-S., 33 Ryu, P.-M., Chen, H.-H., Donatelli, L., Ji...

2022
[14]

In: Latifi, S

Dingare, A., Bein, D., Bein, W., Verma, A.: Abstractive text summarization using machine learning. In: Latifi, S. (ed.) ITNG 2022 19th International Conference on Information Technology-New Generations, pp. 269–276. Springer, Cham (2022)

2022
[16]

Applied Sciences 13(7) (2023) https://doi.org/10.3390/app13074610

Li, Y., Huang, Y., Huang, W., Yu, J., Huang, Z.: An abstractive summariza- tion model based on joint-attention mechanism and a priori knowledge. Applied Sciences 13(7) (2023) https://doi.org/10.3390/app13074610

2023 doi
[17]

In: Webber, B., Cohn, T., He, Y., Liu, Y

Pilault, J., Li, R., Subramanian, S., Pal, C.: On extractive and abstractive neu- ral document summarization with transformer language models. In: Webber, B., Cohn, T., He, Y., Liu, Y. (eds.) Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processin...

2020 doi
[18]

In: 2019 International Joint Conference on Neural Networks (IJCNN), pp

Li, J., Zhang, C., Chen, X., Cao, Y., Liao, P., Zhang, P.: Abstractive text sum- marization with multi-head attention. In: 2019 International Joint Conference on Neural Networks (IJCNN), pp. 1–8 (2019). https://doi.org/10.1109/IJCNN.2019. 8851885

2019 doi
[19]

Tamkin, A., Brundage, M., Clark, J., Ganguli, D.: Understanding the Capabili- ties, Limitations, and Societal Impact of Large Language Models (2021)

2021
[20]

Zhang, T., Kishore, V., Wu, F., Weinberger, K.Q., Artzi, Y.: BERTScore: Evaluating Text Generation with BERT (2020)

2020
[21]

Z Sun: Evaluating large language models on medical evi- dence summarization

L Tang, B.I. Z Sun: Evaluating large language models on medical evi- dence summarization. npj Digit. Med 6, (2023) https://doi.org/10.1038/ s41746-023-00896-7

2023
[22]

arXiv preprint arXiv:2307.11278 (2023)

Abdallah, A., Jatowt, A.: Generator-retriever-generator: A novel approach to open-domain question answering. arXiv preprint arXiv:2307.11278 (2023)

2023 arXiv
[23]

Basyal, L., Sanghvi, M.: Text Summarization Using Large Language Models: A 34 Comparative Study of MPT-7b-instruct, Falcon-7b-instruct, and OpenAI Chat- GPT Models (2023)

2023
[24]

ACM Comput

Liu, P., Yuan, W., Fu, J., Jiang, Z., Hayashi, H., Neubig, G.: Pre-train, prompt, and predict: A systematic survey of prompting methods in natural language processing. ACM Comput. Surv. 55(9) (2023) https://doi.org/10.1145/3560815

2023 doi
[25]

ArXiv abs/2310.14735 (2023)

Chen, B., Zhang, Z., Langren’e, N., Zhu, S.: Unleashing the potential of prompt engineering in large language models: a comprehensive review. ArXiv abs/2310.14735 (2023)

2023 arXiv
[26]

https://arxiv.org/abs/2402.07927

Sahoo, P., Singh, A.K., Saha, S., Jain, V., Mondal, S., Chadha, A.: A System- atic Survey of Prompt Engineering in Large Language Models: Techniques and Applications (2024). https://arxiv.org/abs/2402.07927

2024 arXiv
[27]

ArXiv abs/2401.14423 (2024)

Amatriain, X.: Prompt design and engineering: Introduction and advanced methods. ArXiv abs/2401.14423 (2024)

2024 arXiv
[28]

Accessed: 2024-01-22

The Inverted Pyramid - Purdue OWL - Purdue University. Accessed: 2024-01-22

2024
[29]

Nallapati, R., Zhai, F., Zhou, B.: SummaRuNNer: A Recurrent Neural Network based Sequence Model for Extractive Summarization of Documents (2016)

2016
[30]

Zhou, Q., Yang, N., Wei, F., Huang, S., Zhou, M., Zhao, T.: Neural Document Summarization by Jointly Learning to Score and Select Sentences (2018)

2018
[31]

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A.N., Kaiser, L., Polosukhin, I.: Attention Is All You Need (2023)

2023
[32]

In: Inui, K., Jiang, J., Ng, V., Wan, X

Liu, Y., Lapata, M.: Text summarization with pretrained encoders. In: Inui, K., Jiang, J., Ng, V., Wan, X. (eds.) Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing (E...

2019
[33]

In: Burstein, J., Doran, C., Solorio, T

Devlin, J., Chang, M.-W., Lee, K., Toutanova, K.: BERT: Pre-training of deep bidirectional transformers for language understanding. In: Burstein, J., Doran, C., Solorio, T. (eds.) Proceedings of the 2019 Conference of the North Ameri- can Chapter of the Association for Computa...

2019
[34]

Gambhir, V.G.: Deep learning-based extractive text summarization with word- level attention mechanism

M. Gambhir, V.G.: Deep learning-based extractive text summarization with word- level attention mechanism. Multimed Tools Appl 81, 20829–20852 (2022) https: //doi.org/10.1007/s11042-022-12729-y 35

2022 doi
[35]

Neural Computing and Applications 35(25), 18603–18622 (2023)

Kumar, S., Solanki, A.: An abstractive text summarization technique using trans- former model with self-attention mechanism. Neural Computing and Applications 35(25), 18603–18622 (2023)

2023
[36]

In: Korhonen, A., Traum, D., M` arquez, L

Kouris, P., Alexandridis, G., Stafylopatis, A.: Abstractive text summarization based on deep learning and semantic content generalization. In: Korhonen, A., Traum, D., M` arquez, L. (eds.) Proceedings of the 57th Annual Meeting of the Association for Computational Linguistics,...

2019 doi
[37]

In: Annual Meeting of the Association for Computational Linguistics (2019)

Lewis, M., Liu, Y., Goyal, N., Ghazvininejad, M., Mohamed, A.-r., Levy, O., Stoyanov, V., Zettlemoyer, L.: Bart: Denoising sequence-to-sequence pre- training for natural language generation, translation, and comprehension. In: Annual Meeting of the Association for Computationa...

2019
[38]

In: Durmus, E., Gupta, V., Liu, N., Peng, N., Su, Y

Akiyama, K., Tamura, A., Ninomiya, T.: Hie-BART: Document summarization with hierarchical BART. In: Durmus, E., Gupta, V., Liu, N., Peng, N., Su, Y. (eds.) Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Studen...

2021
[39]

Zhao, Y., Khalman, M., Joshi, R., Narayan, S., Saleh, M., Liu, P.J.: Calibrating Sequence likelihood Improves Conditional Language Generation (2022)

2022
[40]

Zhang, J., Zhao, Y., Saleh, M., Liu, P.J.: PEGASUS: Pre-training with Extracted Gap-sentences for Abstractive Summarization (2020)

2020
[41]

Basyal, L., Sanghvi, M.: Text Summarization Using Large Language Models: A Comparative Study of MPT-7b-instruct, Falcon-7b-instruct, and OpenAI Chat- GPT Models (2023)

2023
[42]

In: Riloff, E., Chiang, D., Hockenmaier, J., Tsujii, J

Narayan, S., Cohen, S.B., Lapata, M.: Don’t give me the details, just the sum- mary! topic-aware convolutional neural networks for extreme summarization. In: Riloff, E., Chiang, D., Hockenmaier, J., Tsujii, J. (eds.) Proceedings of the 2018 Conference on Empirical Methods in N...

2018 doi
[43]

In: Isabelle, P., Charniak, E., Lin, D

Papineni, K., Roukos, S., Ward, T., Zhu, W.-J.: Bleu: a method for automatic evaluation of machine translation. In: Isabelle, P., Charniak, E., Lin, D. (eds.) Proceedings of the 40th Annual Meeting of the Association for Computational Linguistics, pp. 311–318. Association for ...

2002
[44]

In: Koyejo, S., Mohamed, S., Agarwal, A., Belgrave, D., Cho, K., Oh, A

Ouyang, L., Wu, J., Jiang, X., Almeida, D., Wainwright, C., Mishkin, P., Zhang, C., Agarwal, S., Slama, K., Ray, A., Schulman, J., Hilton, J., Kelton, F., Miller, L., Simens, M., Askell, A., Welinder, P., Christiano, P.F., Leike, J., Lowe, R.: Train- ing language models to fol...

2022
[45]

Introducing ChatGPT

OpenAI. Introducing ChatGPT. https://openai.com/blog/chatgpt. Accessed: 2024-01-30

2024
[46]

In: Text Summarization Branches Out, pp

Lin, C.-Y.: ROUGE: A package for automatic evaluation of summaries. In: Text Summarization Branches Out, pp. 74–81. Association for Computational Linguistics, Barcelona, Spain (2004). https://aclanthology.org/W04-1013

2004
[47]

In: Goldstein, J., Lavie, A., Lin, C.-Y., Voss, C

Banerjee, S., Lavie, A.: METEOR: An automatic metric for MT evaluation with improved correlation with human judgments. In: Goldstein, J., Lavie, A., Lin, C.-Y., Voss, C. (eds.) Proceedings of the ACL Workshop on Intrinsic and Extrin- sic Evaluation Measures for Machine Transla...

2005
[48]

Veen, D.V., Uden, C.V., Blankemeier, L., Delbrouck, J.-B., Aali, A., Bluethgen, C., Pareek, A., Polacin, M., Reis, E.P., Seehofnerova, A., Rohatgi, N., Hosamani, P., Collins, W., Ahuja, N., Langlotz, C.P., Hom, J., Gatidis, S., Pauly, J., Chaud- hari, A.S.: Clinical Text Summa...

2023
[49]

In: Rogers, A., Boyd-Graber, J., Okazaki, N

Jain, S., Keshava, V., Mysore Sathyendra, S., Fernandes, P., Liu, P., Neubig, G., Zhou, C.: Multi-dimensional evaluation of text summarization with in-context learning. In: Rogers, A., Boyd-Graber, J., Okazaki, N. (eds.) Findings of the Asso- ciation for Computational Linguist...

2023
[50]

(eds.) Proceedings of the 2022 Conference on Empirical Methods in Natural Language Process- ing, pp

Min, S., Lyu, X., Holtzman, A., Artetxe, M., Lewis, M., Hajishirzi, H., Zettle- moyer, L.: Rethinking the role of demonstrations: What makes in-context learning work? In: Goldberg, Y., Kozareva, Z., Zhang, Y. (eds.) Proceedings of the 2022 Conference on Empirical Methods in Na...

2022 doi
[51]

In: Larochelle, H., Ranzato, M., Hadsell, R., Bal- can, M.F., Lin, H

Brown, T., Mann, B., Ryder, N., Subbiah, M., Kaplan, J.D., Dhariwal, P., Nee- lakantan, A., Shyam, P., Sastry, G., Askell, A., Agarwal, S., Herbert-Voss, A., Krueger, G., Henighan, T., Child, R., Ramesh, A., Ziegler, D., Wu, J., Winter, C., Hesse, C., Chen, M., Sigler, E., Lit...

2020
[52]

Zhang, T., Ladhak, F., Durmus, E., Liang, P., McKeown, K., Hashimoto, T.B.: Benchmarking Large Language Models for News Summarization. Transactions of the Association for Computational Linguistics 12, 39–57 (2024) https://doi.org/10.1162/tacl a 00632 https://direct.mit.edu/tac...

2024 doi
[53]

In: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp

Grusky, M., Naaman, M., Artzi, Y.: Newsroom: A dataset of 1.3 million sum- maries with diverse extractive strategies. In: Proceedings of the 2018 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pp. 708–719...

2018
[54]

In: Proceedings of the 2nd Workshop on New Frontiers in Summarization

Gliwa, B., Mochol, I., Biesek, M., Wawer, A.: Samsum corpus: A human- annotated dialogue dataset for abstractive summarization. In: Proceedings of the 2nd Workshop on New Frontiers in Summarization. Association for Computational Linguistics, ??? (2019). https://doi.org/10.1865...

2019 doi
[55]

In: Walker, M., Ji, H., Stent, A

Cohan, A., Dernoncourt, F., Kim, D.S., Bui, T., Kim, S., Chang, W., Gohar- ian, N.: A discourse-aware attention model for abstractive summarization of long documents. In: Walker, M., Ji, H., Stent, A. (eds.) Proceedings of the 2018 Con- ference of the North American Chapter of...

2018
[56]

ACM Trans

Wang, W., Zheng, V.W., Yu, H., Miao, C.: A survey of zero-shot learning: Set- tings, methods, and applications. ACM Trans. Intell. Syst. Technol. 10(2) (2019) https://doi.org/10.1145/3293318

2019 doi
[57]

ArXiv abs/2211.15661 (2022)

Aky¨ urek, E., Schuurmans, D., Andreas, J., Ma, T., Zhou, D.: What learn- ing algorithm is in-context learning? investigations with linear models. ArXiv abs/2211.15661 (2022)

2022 arXiv
[58]

In: Krause, A., Brunskill, E., Cho, K., Engelhardt, B., Sabato, S., Scarlett, J

Von Oswald, J., Niklasson, E., Randazzo, E., Sacramento, J., Mordvintsev, A., Zhmoginov, A., Vladymyrov, M.: Transformers learn in-context by gradi- ent descent. In: Krause, A., Brunskill, E., Cho, K., Engelhardt, B., Sabato, S., Scarlett, J. (eds.) Proceedings of the 40th Int...

2023
[59]

OpenAI blog 1(8), 9 (2019)

Radford, A., Wu, J., Child, R., Luan, D., Amodei, D., Sutskever, I., et al.: 38 Language models are unsupervised multitask learners. OpenAI blog 1(8), 9 (2019)

2019
[60]

In: Korhonen, A., Titov, I

He, X., Haffari, G., Norouzi, M.: Sequence to sequence mixture model for diverse machine translation. In: Korhonen, A., Titov, I. (eds.) Proceedings of the 22nd Conference on Computational Natural Language Learning, pp. 583–592. Associ- ation for Computational Linguistics, Bru...

2018 doi
[61]

Shi, T., Keneshloo, Y., Ramakrishnan, N., Reddy, C.K.: Neural Abstractive Text Summarization with Sequence-to-Sequence Models (2020)

2020
[62]

Dalal, M., Li, A.C., Taori, R.: Autoregressive models: What are they good for? ArXiv abs/1910.07737 (2019)

2019 arXiv
[63]

Touvron, H., Martin, L., Stone, K., Albert, P., Almahairi, A., Babaei, Y., Bash- lykov, N., Batra, S., Bhargava, P., Bhosale, S., Bikel, D., Blecher, L., Ferrer, C.C., Chen, M., Cucurull, G., Esiobu, D., Fernandes, J., Fu, J., Fu, W., Fuller, B., Gao, C., Goswami, V., Goyal, N...

2023
[64]

In: Barolli, L

Cai, P.-X., Fan, Y.-C., Leu, F.-Y.: Compare encoder-decoder, encoder-only, and decoder-only architectures for text generation on low-resource datasets. In: Barolli, L. (ed.) Advances on Broad-Band Wireless Computing, Communication and Applications, pp. 216–225. Springer, Cham (2022)

2022
[65]

ArXiv abs/2310.06825 (2023)

Jiang, A.Q., Sablayrolles, A., Mensch, A., Bamford, C., Chaplot, D.S., Las Casas, D., Bressand, F., Lengyel, G., Lample, G., Saulnier, L., Lavaud, L.R., Lachaux, M.-A., Stock, P., Scao, T.L., Lavril, T., Wang, T., Lacroix, T., Sayed, W.E.: Mistral 7b. ArXiv abs/2310.06825 (2023)

2023 arXiv
[66]

https://storage.googleapis.com/deepmind-media/gemma/ gemma-report.pdf

Gemma. https://storage.googleapis.com/deepmind-media/gemma/ gemma-report.pdf. Accessed: 2024-02-5

2024
[67]

ArXiv abs/2401.04088 (2024) 39

Jiang, A.Q., Sablayrolles, A., Roux, A., Mensch, A., Savary, B., Bamford, C., Chaplot, D.S., Las Casas, D., Hanna, E.B., Bressand, F., Lengyel, G., Bour, G., Lample, G., Lavaud, L.R., Saulnier, L., Lachaux, M.-A., Stock, P., Subramanian, S., Yang, S., Antoniak, S., Scao, T.L.,...

2024 arXiv
[68]

https://arxiv.org/abs/2407.06204

Cai, W., Jiang, J., Wang, F., Tang, J., Kim, S., Huang, J.: A Survey on Mixture of Experts (2024). https://arxiv.org/abs/2407.06204

2024 arXiv
[69]

https://arxiv.org/abs/ 2005.08056

Gong, H., Shen, Y., Yu, D., Chen, J., Yu, D.: Recurrent Chunking Mechanisms for Long-Text Machine Reading Comprehension (2020). https://arxiv.org/abs/ 2005.08056

2020 arXiv
[70]

IEEE Transactions on Visualization & Computer Graphics 29(01), 1146–1156 (2023) https://doi.org/10.1109/TVCG.2022.3209479

Strobelt, H., Webson, A., Sanh, V., Hoover, B., Beyer, J., Pfister, H., Rush, A.M.: Interactive and visual prompt engineering for ad-hoc task adaptation with large language models. IEEE Transactions on Visualization & Computer Graphics 29(01), 1146–1156 (2023) https://doi.org/...

2023
[71]

ArXiv abs/2304.14670 (2023)

Wang, J., Shi, E., Yu, S., Wu, Z., Ma, C., Dai, H., Yang, Q., Kang, Y., Wu, J., Hu, H., Yue, C., Zhang, H., Liu, Y.-H., Li, X., Ge, B., Zhu, D., Yuan, Y., Shen, D., Liu, T., Zhang, S.: Prompt engineering for healthcare: Methodologies and applications. ArXiv abs/2304.14670 (2023)

2023 arXiv
[72]

Peeperkorn, M., Kouwenhoven, T., Brown, D.G., Jordanous, A.K.: Is tempera- ture the creativity parameter of large language models? ArXiv abs/2405.00492 (2024)

2024 arXiv
[73]

https://arxiv.org/abs/2301.00234

Dong, Q., Li, L., Dai, D., Zheng, C., Ma, J., Li, R., Xia, H., Xu, J., Wu, Z., Chang, B., Sun, X., Li, L., Sui, Z.: A Survey on In-context Learning (2024). https://arxiv.org/abs/2301.00234

2024 arXiv
[74]

IEEE Access 9, 156043–156070 (2021) https://doi.org/10.1109/ACCESS.2021.3129786

Mridha, M.F., Lima, A.A., Nur, K., Das, S.C., Hasan, M., Kabir, M.M.: A survey of automatic text summarization: Progress, process and challenges. IEEE Access 9, 156043–156070 (2021) https://doi.org/10.1109/ACCESS.2021.3129786

2021
[75]

https://arxiv.org/abs/1908.08960

Kry´ sci´ nski, W., Keskar, N.S., McCann, B., Xiong, C., Socher, R.: Neural Text Summarization: A Critical Evaluation (2019). https://arxiv.org/abs/1908.08960

2019 arXiv
[76]

In: Webber, B., Cohn, T., He, Y., Liu, Y

Bhandari, M., Gour, P.N., Ashfaq, A., Liu, P., Neubig, G.: Re-evaluating eval- uation in text summarization. In: Webber, B., Cohn, T., He, Y., Liu, Y. (eds.) Proceedings of the 2020 Conference on Empirical Methods in Natural Language Processing (EMNLP), pp. 9347–9359. Associat...

2020 doi
[165]

https://doi.org/ 10.18653/v1/2021.naacl-srw.20

Association for Computational Linguistics, Online (2021). https://doi.org/ 10.18653/v1/2021.naacl-srw.20 . https://aclanthology.org/2021.naacl-srw.20

2021 doi
[621]

https://doi.org/10.18653/v1/N18-2097

Association for Computational Linguistics, New Orleans, Louisiana (2018). https://doi.org/10.18653/v1/N18-2097 . https://aclanthology.org/N18-2097

2018 doi
[3740]

https://doi.org/10.18653/v1/D19-1387

Association for Computational Linguistics, Hong Kong, China (2019). https://doi.org/10.18653/v1/D19-1387 . https://aclanthology.org/D19-1387

2019 doi

Pith tools

Reviewed August 6, 2026 · model on record in the stance chip above.

[1] [1]

Expert Systems with Applications 165, 113679 (2021) https://doi.org/10.1016/j.eswa.2020.113679

El-Kassas, W.S., Salama, C.R., Rafea, A.A., Mohamed, H.K.: Automatic text summarization: A comprehensive survey. Expert Systems with Applications 165, 113679 (2021) https://doi.org/10.1016/j.eswa.2020.113679

work page arXiv 2021

[2] [2]

In: Riezler, S., Goldberg, Y

Nallapati, R., Zhou, B., Santos, C., Gul¸ cehre, C ¸ ., Xiang, B.: Abstractive text summarization using sequence-to-sequence RNNs and beyond. In: Riezler, S., Goldberg, Y. (eds.) Proceedings of the 20th SIGNLL Conference on Computa- tional Natural Language Learning, pp. 280–290. Association for Computational Linguistics, Berlin, Germany (2016). https://do...

doi:10.18653/v1/k16-1028 2016

[3] [3]

Journal of King Saud University - Computer and Information Sciences 34(4), 1029–1046 (2022) https://doi.org/10.1016/j.jksuci.2020.05.006 32

Widyassari, A.P., Rustad, S., Shidik, G.F., Noersasongko, E., Syukur, A., Affandy, A., Setiadi, D.R.I.M.: Review of automatic text summarization tech- niques & methods. Journal of King Saud University - Computer and Information Sciences 34(4), 1029–1046 (2022) https://doi.org/10.1016/j.jksuci.2020.05.006 32

work page doi:10.1016/j.jksuci.2020.05.006 2022

[4] [4]

IOP Conference Series: Materials Science and Engineering 1099(1), 012042 (2021) https://doi.org/10.1088/1757-899X/1099/ 1/012042

Dehru, V., Tiwari, P.K., Aggarwal, G., Joshi, B., Kartik, P.: Text summariza- tion techniques and applications. IOP Conference Series: Materials Science and Engineering 1099(1), 012042 (2021) https://doi.org/10.1088/1757-899X/1099/ 1/012042

work page doi:10.1088/1757-899x/1099/ 2021

[5] [5]

Information Processing & Management 45(1), 20–34 (2009) https://doi.org/10.1016/j.ipm.2008.06.002

Lee, J.-H., Park, S., Ahn, C.-M., Kim, D.: Automatic generic document summa- rization based on non-negative matrix factorization. Information Processing & Management 45(1), 20–34 (2009) https://doi.org/10.1016/j.ipm.2008.06.002

work page doi:10.1016/j.ipm.2008.06.002 2009

[6] [6]

In: 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp

Gigioli, P., Sagar, N., Rao, A., Voyles, J.: Domain-aware abstractive text sum- marization for medical documents. In: 2018 IEEE International Conference on Bioinformatics and Biomedicine (BIBM), pp. 2338–2343 (2018). https://doi.org/ 10.1109/BIBM.2018.8621539

work page arXiv 2018

[7] [7]

Neural Computing and Applications 35, 18553–18570 (2023) https://doi.org/10.1007/s00521-023-08680-0

Li, S., Xu, J.: Hiermds: a hierarchical multi-document summarization model with global–local document dependencies. Neural Computing and Applications 35, 18553–18570 (2023) https://doi.org/10.1007/s00521-023-08680-0

work page doi:10.1007/s00521-023-08680-0 2023

[8] [8]

In: Muresan, S., Nakov, P., Villavicencio, A

Xiao, W., Beltagy, I., Carenini, G., Cohan, A.: PRIMERA: Pyramid- based masked sentence pre-training for multi-document summarization. In: Muresan, S., Nakov, P., Villavicencio, A. (eds.) Proceedings of the 60th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pp. 5245–5263. Association for Computational Linguis- ti...

doi:10.18653/v1/2022.acl-long.360 2022