Pith. sign in

REVIEW 6 major objections 7 minor 35 references

Reasoning-Oriented and Analogy-Based Methods for Locating and Editing in Zero-Shot Event-Relational Reasoning

T0 review · 6 major / 7 minor · reviewed 2026-08-10 · deepseek-v4-flash

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

Pith's one-line read The paper claims that locating one reasoning module and editing its weights transfers event-relation skills between tasks, beating prefix tuning and large closed models on most of ten zero-shot benchmarks.

desk verdict The SOTA claim is inflated by per-dataset cherry-picking across four ABLE variants, but the cheap editing-transfer idea and the localization findings are worth a serious look. read the letter →

arxiv 2501.00803 v1 pith:QAKPVESQ submitted 2025-01-01 cs.CL cs.AI

Jingyao Tang , Lishuang Li , Liteng Mi , Haiming Wu , Hongbin Lu This is my paper · ORCID
classification cs.CLcs.AI
keywords zero-shotlearningevent-relationalreasoningknowledgeeditingmodelinterpretabilityprefixtuninganalogy-basedtransfercausalrelationextractionlanguage
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

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

The reading

This paper tries to show that zero-shot event-relational reasoning—deciding whether events stand in causal, temporal, or sub-event relations—can be improved by locating and editing a small set of weights inside a large language model instead of training task-specific prefixes. The authors' ROLE method finds the single module (the encoder's MLP or the decoder's cross-attention) whose output most affects the yes/no answer, then edits its output projection weights with a constrained objective that suppresses the model's tendency to answer 'Yes' too often. Their ABLE method then treats the locating and editing information of three tasks as the corners of a parallelogram and computes the fourth task's edit by vector subtraction, claiming that this transfers reasoning knowledge across tasks. If the claim is right, state-of-the-art zero-shot results on most of ten event-relational benchmarks can be obtained in a fraction of a second of training, with an interpretable account of where the model stores relational cues.

What carries the argument

The load-bearing object is the edit vector $\Delta W$ for a located module $H_{\langle T,L \rangle}$—the change to the output-projection weights of the encoder's MLP or the decoder's cross-attention. It is computed with a constrained objective (Equation 5, in the style of knowledge editing) using 500 positive and 500 negative MAVEN samples per task. The transfer identity is the parallelogram rule $\Delta W_D = \Delta W_C - \alpha(\Delta W_A - \Delta W_B)$, with the location of D obtained by the same subtraction (Equation 8); the value of $\alpha$ is not reported. The average indirect effect, Pearl's measure of how much a module changes the answer probability when its hidden state is perturbed, is what makes locating empirical: the key module is selected by its measured causal effect rather than by training.

What would settle it

Reproduce the ABLE transfer for a target task D without touching D's labels: compute $\Delta W_C$, $\Delta W_A$, and $\Delta W_B$ from their MAVEN locating samples, sweep $\alpha$ over a fine grid, and apply Equation 9 to the located module. If no single $\alpha$ brings the transferred edit within one or two F1 points of the direct ROLE edit on a held-out subset of the same MAVEN family—or if the best $\alpha$ changes sharply across the four task-quadruples in Table 6—the parallelogram premise fails. Also check whether any of the 500+500 locating samples appear in the MAVEN-intra evaluation subsets; if they do, the zero-shot numbers are inflated by memorization.

Watch

Extended reading notes

Core claim

The paper's central claim is that a language model's ability to reason about event relations can be improved by locating one type of module—the MLP in the encoder and the cross-attention in the decoder of Flan-T5-large—and editing the weights of that module, rather than by training soft prompts. On the locating side, the paper uses the average indirect effect of each module on the yes/no answer to identify the token and layer that matter most, and it reports that the encoder MLP stores relational cues such as relation-type words and question words while the decoder cross-attention integrates those cues at the start token. On the transfer side, ABLE treats four tasks as the corners of a parallelogram: given located edits for three tasks A, B, C, the edit for a fourth task D is $\Delta W_D = \Delta W_C - \alpha(\Delta W_A - \Delta W_B)$ and its location is obtained analogously. The paper reports state-of-the-art zero-shot F1 on most of ten datasets for causal and sub-event relation tasks, with the largest gains on sparse targets such as CTB-uni and MAVEN-intra sub-event extraction.

Load-bearing premise

The load-bearing premise is that edit vectors compose by parallelogram arithmetic in weight space with a stable $\alpha$; the paper also never states that the 500+500 MAVEN locating samples are disjoint from the MAVEN-intra test subsets.

Editorial extensions

If this is right

  • Zero-shot event-relational reasoning can be improved without training prefixes at all: editing the output weights of one located module in a frozen Flan-T5-large suffices.
  • Because the transfer rule is arithmetic on edit vectors, a new task D costs only the cheap computations of the A, B, C edits; the paper reports ABLE training times of 0.09 seconds (encoder) and 0.03 seconds (decoder) on CTB-uni.
  • The located modules give a mechanism story: the encoder MLP handles relation-type and question words, while the decoder cross-attention routes relational information from the start token, which also explains the model's 'Yes' bias and where to correct it.
  • The analogy is not uniform: causal/temporal and causal/sub-event pairs transfer well, while temporal/sub-event transfer is weak, so the method's gains should be expected on tasks near one of those axes.
  • On the weakest baselines the margins are large (for example, CTB-uni 13.64 F1 versus 8.95 for UniEvent and 2.49 for GPT-4), indicating that the edit also fixes a systematic over-prediction of 'Yes'.

Reading between the lines

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

  • A reader should verify whether the 500+500 MAVEN locating samples are disjoint from the MAVEN-intra evaluation subsets; the paper never states this, and overlap would inflate the reported zero-shot gains through memorization.
  • If the parallelogram rule holds beyond this setting, the same vector arithmetic could be tested on other binary relational heads of a frozen transformer—contradiction, coreference, or sentiment-pair classification—where locating samples are abundant and target labels are scarce.
  • The paper's story that the edit mainly counters a 'Yes' bias suggests a sharper test: apply the same $\Delta W$ to an unrelated binary task that also suffers from yes-hallucination; if it transfers, the edit vector is a bias correction, and if not, it is relation-specific.
  • The unreported $\alpha$ is worth probing: reporting the sensitivity curve of F1 versus $\alpha$ for each task quadruple would show whether the analogy has one universal strength or needs per-pair tuning.
Share X Bluesky LinkedIn Reddit HN

Signed reviews

No signed human review yet.

Editorial analysis

A structured set of objections, weighed in public.

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

Referee Report

6 major / 7 minor

Summary. The paper proposes two model-editing methods for zero-shot event-relational reasoning with Flan-T5-large. ROLE locates key MLP/cross-attention modules via average indirect effects and computes parameter edits to improve reasoning; ABLE transfers edits between tasks using a parallelogram analogy in weight space (Eq. 8-9). The methods are evaluated on 10 datasets covering causal relation extraction/classification and sub-event extraction, with claims of state-of-the-art results and large computational savings.

Significance. If the claims were reproducible and internally consistent, the work would be valuable: it brings interpretable locate-and-edit techniques to reasoning tasks, reports a detailed module-level analysis, and demonstrates potential efficiency gains over prefix tuning. The paper also includes a thoughtful discussion of reasoning mechanisms and a novel analogy-based transfer idea. However, the central SOTA claim is currently undermined by the issues detailed below, so the significance is conditional on substantial revision.

major comments (6)
  1. [Section 4.3, Table 2 vs Tables 3-5] The 'ABLE' row in Table 2 is not a single method. Cross-referencing the per-dataset F1 values with Tables 3-5 shows that the reported score is in each case the maximum over the four ABLE variants (ABLE1Enc, ABLE2Enc, ABLE1Dec, ABLE2Dec); for example, SCI-uni uses ABLE2Dec, ESL-uni uses ABLE1Enc, CTB-uni uses ABLE2Dec, and MAVEN-intra-subevent uses ABLE1Enc. The paper never states which variant is reported or how it was selected. Since the benchmark is presented as zero-shot, there is no validation split on which such a selection could legitimately be made. If the row is an oracle over four configurations, the 'SOTA on most datasets' claim is inflated by multiple comparisons. Report a fixed configuration or a principled selection rule, and adjust the SOTA claims accordingly.
  2. [Table 5, ABLE2Enc row on HiEve] The ABLE2Enc entry on HiEve reports P=59.00, R=98.82, F1=11.14. These three numbers are inconsistent: the harmonic mean of 59.00 and 98.82 is approximately 73.89, not 11.14. This indicates a transcription or computation error and casts doubt on the reliability of the other values in Table 5. The authors must correct the entry and re-verify all reported F1 scores.
  3. [Section 3.2, Eq. (9)] The hyperparameter α in Eq. (9) is never reported, and the text does not state how it is chosen. The composed edit ΔW_D depends directly on α; without reporting its value or selection procedure, the method is not reproducible, and if α is tuned on target-task data, the zero-shot transfer claim would be compromised. Provide the α values used for each dataset and describe how they were selected.
  4. [Section 4.6, Table 7] The analogicality analysis of editing magnitude is circular. Table 7 computes similarities between differences of edit vectors (e.g., ΔW_AB vs ΔW_CD) that are the same quantities used to motivate Eq. (9). This does not test whether the composed edit ΔW_D = ΔW_C − α(ΔW_A − ΔW_B) improves a held-out target. The only direct test is in the ablation tables, but those rely on the post-hoc oracle selection criticized above. A held-out evaluation of the composed edit is needed to validate the analogy.
  5. [Section 4.4, last paragraph] The claim that 'ABLE 1Enc, ABLE 2Enc, ABLE 1Dec, and ABLE 2Dec improve the F1 score of all tasks' is false as stated: in Table 5, ABLE2Enc on HiEve has F1=11.14, which is lower than the w/oAll baseline of 11.18. This factual error should be corrected to reflect the per-dataset results.
  6. [Appendix A and Table 2] The paper uses 500 positive and 500 negative MAVEN samples for reasoning-oriented locating but never states that these samples are disjoint from the MAVEN-intra-causal and MAVEN-intra-subevent evaluation sets. Without this assurance, the zero-shot status of those two datasets in Table 2 is unclear. Specify the exact data split (e.g., train versus test) and confirm disjointness.
minor comments (7)
  1. [Table 1 and Appendix B] The token names are inconsistent: Table 1 uses '</s>' for the decoder cross-attention token, while Table 11's caption refers to the '<s>' token; please standardize the notation.
  2. [Section 3.1.1, Eqs. (1)-(2)] The conditioning notation in Eqs. (1) and (2) is ambiguous: it is unclear what x*, h*, and the patched module condition refer to, and why Eq. (1) conditions on x* in both terms while Eq. (2) does not. Clarify these definitions.
  3. [Section 4.4, first paragraph] When discussing why ROLE performs poorly, the paper does not name the task; specify that this concerns causal relation classification and identify which datasets are affected.
  4. [Table 3] The column header 'MAVEN-intra' should be 'MAVEN-intra-causal' for consistency with Table 2 and Section 4.1.
  5. [Section 4.2 and Table 2] The list of baselines includes text-davinci-002 and text-davinci-003, but Table 2 reports only GPT-3.5 and GPT-4; clarify which models were actually evaluated on which datasets.
  6. [Section 4.7, Table 8] The meanings of 'Params (M)' and 'Training time (s)' for ROLE and ABLE are not defined; clarify whether these refer to the number of edited parameters and the locating-plus-editing time, respectively.
  7. [Appendix C, Algorithm 1] The pseudocode has a formatting typo in the print statement, 'print i, count[i])', and needs a base case for i=0; also consider defining the eigenvalue threshold criterion more precisely.

Circularity Check

2 steps flagged · score 6.0 of 10

The ABLE transfer rule is validated by a tautological similarity check, and the SOTA table is a per-dataset best-of-four selection.

  1. self definitional [Section 3.2, Eq. (9) and Section 4.6, Table 7]
    "∆WD = ∆WC −α· (∆WA −∆WB), where, α is a hyperparameter that regulates the degree of being analogized. ... Let ∆W X denote the editing magnitude for a module parameter of task X, and let ∆W XY = ∆W X − ∆W Y ... Table 7 shows the similarity between different ∆W XY ."

    By Eq. (9), the transferred edit for task D is defined as ∆WC − α(∆WA − ∆WB). Therefore the difference ∆WCD = ∆WC − ∆WD equals α(∆WA − ∆WB) = α∆WAB identically, for any positive α. The main-eigenvector cosine similarity sim(∆WAB, ∆WCD) reported in Table 7 is thus exactly 1 by construction, regardless of whether the tasks are actually analogous. The same identity applies to any pair whose transfer target is produced by Eq. (9) from the same A and B. Section 4.6 presents this as evidence that 'verifies the analogicality of editing magnitude', but it is a restatement of the defining equation, not an independent empirical check. The hyperparameter α is unreported, so the claimed transfer rule is also not reproducible as stated.

  2. fitted input called prediction [Table 2; Tables 3, 4, and 5; Section 4.4]
    "ABLE 83.48 72.42 13.64 38.48 21.63 37.43 69.90 68.42 12.59 17.69 ... We construct four forms of ABLE, including ABLE 1 Enc, ABLE 2 Enc, ABLE 1 Dec, and ABLE 2 Dec ... based on the best results of ABLE in each table, we observe that strong analogies are shown..."

    Each F1 in the 'ABLE' row of Table 2 coincides with the maximum of the four ABLE variants for that dataset in Tables 3–5 (e.g., SCI-uni 83.48 = ABLE2Dec, ESL-uni 72.42 = ABLE1Enc, CTB-uni 13.64 = ABLE2Dec, CNC 69.90 = ABLE2Dec). The paper never states which fixed configuration produced Table 2, and Section 4.4 says the analogies are read 'based on the best results', i.e., after seeing test outcomes. In a zero-shot setting there is no validation split on which such a choice can legitimately be made, so the reported SOTA row is effectively the maximum over four configurations selected per dataset using target labels. The headline prediction is therefore partly an oracle selection over the method's own outputs rather than the result of a single fixed zero-shot transfer rule.

full rationale

The central ABLE transfer formula, Eq. (9), does have independent content for the eight non-MAVEN benchmarks: no task-D labels are used to compute ΔW_D provided α is fixed and the variant is fixed. However, the paper's own validation of the editing analogy is circular. Section 4.6 defines ΔW_XY = ΔW_X − ΔW_Y and then reports high similarity between ΔW_AB and ΔW_CD, but ΔW_D is defined in Eq. (9) as ΔW_C − α(ΔW_A − ΔW_B), making ΔW_CD = αΔW_AB identically; the similarity is a tautology. This is a concrete equation-level reduction, not a vague feeling. In addition, the SOTA claim in Table 2 is presented for a single method named 'ABLE', yet the numbers show that the row is the per-dataset maximum of four distinct ABLE configurations, with the selection made after inspecting 'the best results'. That makes the headline comparison partly a multiple-comparisons artifact rather than a fixed zero-shot prediction. The MAVEN-locating to MAVEN-intra evaluation overlap is a data-leakage concern but is not itself a circular derivation; the paper does not state disjointness. No load-bearing self-citation chain was found: the MEMIT-style editing formula is imported from external work and applied here to new reasoning tasks. Overall, the transfer concept is not entirely vacuous, but two load-bearing validation/summary steps reduce to their own construction, warranting a score of 6.

Assumptions & free parameters 4 free parameters · 5 assumptions · 0 invented entities

The central claim rests on borrowed ROME/MEMIT assumptions, an unvalidated additive weight-space transfer rule, and several unreported hyperparameters (alpha, lambda, seed, top-k rules). These shift the burden of proof to the empirical evaluation, which is only partially clean. No new particles, forces, or theoretical entities are introduced; the reasoning-mechanism hypothesis is an interpretation of existing modules, not a new entity.

free parameters (4)
  • alpha (analogy interpolation weight, Eq. 9) = not reported
    Controls the magnitude of the transferred edit between tasks in ABLE; the paper never gives its value or selection procedure.
  • lambda (regularizer weight in C0, Eq. 5) = not reported
    Controls the constraint strength in the ROME/MEMIT-style closed-form update; the value is never reported.
  • Layer selection rule for ROLE (top-k choice) = top-3 encoder, top-1 decoder, with an ad hoc exception for sub-event extraction
    The number of edited layers and the exception rule (Section 4.5, Table 6) are chosen by hand without sensitivity analysis.
  • MAVEN locating sample selection (500 positive + 500 negative per task) = random, seed not reported
    Appendix A states random selection but gives no seed, so the locating samples are not reproducible and may overlap the MAVEN-intra evaluation subsets.
assumptions (5)
  • domain assumption Causal tracing via average indirect effect (Eqs. 1-2) reliably identifies the modules whose intervention changes model behavior in this task family.
    Borrowed from ROME (Meng et al. 2022a) and applied to T5 without re-validation against ground-truth reasoning circuits.
  • domain assumption The closed-form update DeltaW = R K1^T (C0 + K1 K1^T)^(-1) (Eq. 5) installs the desired behavior at module H<T,L>.
    Directly imported from MEMIT (Meng et al. 2022b); assumes the linearization assumptions of ROME hold for T5's MLP and cross-attention modules.
  • ad hoc to paper Task edits form an approximate parallelogram: DeltaW_D = DeltaW_C - alpha*(DeltaW_A - DeltaW_B) (Eq. 9).
    The paper's central transfer postulate; validated only on the same edit vectors used to build it (Table 7), with alpha unreported.
  • domain assumption Flan-T5-large's pretraining makes it biased toward answering 'Yes' (memory hallucination), justifying the objective in Eq. 4.
    Invoked to construct the target output M1; the yes-bias is cited to Mckenna et al. 2023 rather than measured on these datasets.
  • ad hoc to paper The SVD-based prompt segmentation (Algorithm 1) yields meaningful token groups for interpreting heatmaps.
    Heuristic eigenvalue-threshold method with no formal criterion or evaluation of the resulting segmentation.

how reviews work

0 comments
Cite this review

Pith. "Pith review of Reasoning-Oriented and Analogy-Based Methods for Locating and Editing in Zero-Shot Event-Relational Reasoning." pith.science (2026). https://pith.science/paper/QAKPVESQ

@misc{pith2026250100803,
  author       = {Pith},
  title        = {Pith review of: Reasoning-Oriented and Analogy-Based Methods for Locating and Editing in Zero-Shot Event-Relational Reasoning},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/QAKPVESQ}},
  note         = {Machine review of arXiv:2501.00803}
}
read the original abstract

Zero-shot event-relational reasoning is an important task in natural language processing, and existing methods jointly learn a variety of event-relational prefixes and inference-form prefixes to achieve such tasks. However, training prefixes consumes large computational resources and lacks interpretability. Additionally, learning various relational and inferential knowledge inefficiently exploits the connections between tasks. Therefore, we first propose a method for Reasoning-Oriented Locating and Editing (ROLE), which locates and edits the key modules of the language model for reasoning about event relations, enhancing interpretability and also resource-efficiently optimizing the reasoning ability. Subsequently, we propose a method for Analogy-Based Locating and Editing (ABLE), which efficiently exploits the similarities and differences between tasks to optimize the zero-shot reasoning capability. Experimental results show that ROLE improves interpretability and reasoning performance with reduced computational cost. ABLE achieves SOTA results in zero-shot reasoning.

Figures

Figures reproduced from arXiv: 2501.00803 by the authors.

Figure 1
Figure 1. Comparison of knowledge transfer between [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Overview of ROLE and ABLE. The left side shows the application of ROLE and ABLE in the Flan [PITH_FULL_IMAGE:figures/full_fig_p003_2.png] view at source ↗
Figure 3
Figure 3. Heatmaps of the effect of Transformer, MLP, and Self-Attention modules on each token for each layer [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (4 more)
Figure 4
Figure 4. Figure 4: Line plots of the effect of the Transformer, MLP, Self-Attention and Cross-Attention modules on the [PITH_FULL_IMAGE:figures/full_fig_p005_4.png]
Figure 5
Figure 5. Figure 5: Reasoning mechanism for Flan-T5-large in [PITH_FULL_IMAGE:figures/full_fig_p005_5.png]
Figure 6
Figure 6. Figure 6: Line plots of the effect difference between classification and extraction for temporal, causal, and subevent [PITH_FULL_IMAGE:figures/full_fig_p008_6.png]
Figure 7
Figure 7. Figure 7: A graphical illustration of the segmentation results of the SVD-based prompt segmentation algorithm [PITH_FULL_IMAGE:figures/full_fig_p013_7.png]

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

35 extracted references · 24 canonical work pages

  1. [1]

    Tommaso Caselli and Piek Vossen. 2017. The event storyline corpus: A new benchmark for causal and temporal relation extraction. In Proceedings of the Events and Stories in the News Workshop, pages 77--86

    work page 2017

  2. [2]

    Meiqi Chen, Yixin Cao, Kunquan Deng, Mukai Li, Kun Wang, Jing Shao, and Yan Zhang. 2022. Ergo: Event relational graph transformer for document-level event causality identification. In Proceedings of the 29th International Conference on Computational Linguistics, pages 2118--2128

    work page 2022

  3. [3]

    Mingliang Chen, Wenzhong Yang, Fuyuan Wei, Qicai Dai, Mingjie Qiu, Chenghao Fu, and Mo Sha. 2024. Event causality identification via structure optimization and reinforcement learning. Knowledge-Based Systems, 284:111256

    work page 2024

  4. [4]

    Qingxiu Dong, Damai Dai, Yifan Song, Jingjing Xu, Zhifang Sui, and Lei Li. 2022. Calibrating factual knowledge in pretrained language models. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 5937--5947

    work page 2022

  5. [5]

    Giacomo Frisoni, Gianluca Moro, and Lorenzo Balzani. 2022. Text-to-text extraction and verbalization of biomedical event graphs. In Proceedings of the 29th International Conference on Computational Linguistics, pages 2692--2710

    work page 2022

  6. [6]

    Jinglong Gao, Xiao Ding, Bing Qin, and Ting Liu. 2023. Is chatgpt a good causal reasoner? a comprehensive evaluation. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 11111--11126

    work page 2023

  7. [7]

    Goran Glava s , Jan S najder, Marie Francine Moens, and Parisa Kordjamshidi. 2014. Hieve: A corpus for extracting event hierarchies from news stories. In Proceedings of the Ninth International Conference on Language Resources and Evaluation (LREC'14), pages 3678--3683

    work page 2014

  8. [8]

    Christopher Hidey and Kathleen McKeown. 2016. Identifying causal relations using parallel wikipedia articles. In Proceedings of the 54th Annual Meeting of the Association for Computational Linguistics (Volume 1: Long Papers), pages 1424--1433

    2016

Show all 35 references
  1. [9]

    Zeyu Huang, Yikang Shen, Xiaofeng Zhang, Jie Zhou, Wenge Rong, and Zhang Xiong. 2023. Transformer-patcher: One mistake worth one neuron. arXiv preprint arXiv:2301.09785

    2023 arXiv

  2. [10]

    Viet Lai, Hieu Man, Linh Ngo, Franck Dernoncourt, and Thien Nguyen. 2022. Multilingual subevent relation extraction: A novel dataset and structure induction method. In Findings of the Association for Computational Linguistics: EMNLP 2022, pages 5559--5570

    2022

  3. [11]

    Changsheng Lv, Shuai Zhang, Yapeng Tian, Mengshi Qi, and Huadong Ma. 2024. Disentangled counterfactual learning for physical audiovisual commonsense reasoning. Advances in Neural Information Processing Systems, 36

    2024

  4. [12]

    Arjun Majumdar, Anurag Ajay, Xiaohan Zhang, Pranav Putta, Sriram Yenamandra, Mikael Henaff, Sneha Silwal, Paul Mcvay, Oleksandr Maksymets, Sergio Arnaud, et al. 2024. Openeqa: Embodied question answering in the era of foundation models. In Proceedings of the IEEE/CVF Conferenc...

    2024

  5. [13]

    Hieu Man, Franck Dernoncourt, and Thien Huu Nguyen. 2024 a . Mastering context-to-label representation transformation for event causality identification with diffusion models. In Proceedings of the AAAI Conference on Artificial Intelligence, volume 38, pages 18760--18768

    2024

  6. [14]

    Hieu Man, Minh Van Nguyen, and Thien Huu Nguyen. 2022. Event causality identification via generation of important context words. In Proceedings of the 11th Joint Conference on Lexical and Computational Semantics (* SEM) at NAACL 2022

    2022

  7. [15]

    Hieu Man, Chien Van Nguyen, Nghia Trung Ngo, Linh Ngo, Franck Dernoncourt, and Thien Huu Nguyen. 2024 b . Hierarchical selection of important context for generative event causality identification with optimal transports. In Proceedings of the 2024 Joint International Conferenc...

    2024

  8. [16]

    Nick Mckenna, Tianyi Li, Liang Cheng, Mohammad Hosseini, Mark Johnson, and Mark Steedman. 2023. Sources of hallucination by large language models on inference tasks. In Findings of the Association for Computational Linguistics: EMNLP 2023, pages 2758--2774

    2023

  9. [17]

    Kevin Meng, David Bau, Alex Andonian, and Yonatan Belinkov. 2022 a . Locating and editing factual associations in gpt. Advances in Neural Information Processing Systems, 35:17359--17372

    2022

  10. [18]

    Kevin Meng, Arnab Sen Sharma, Alex Andonian, Yonatan Belinkov, and David Bau. 2022 b . Mass-editing memory in a transformer. arXiv preprint arXiv:2210.07229

    2022 arXiv

  11. [19]

    Paramita Mirza, Rachele Sprugnoli, Sara Tonelli, and Manuela Speranza. 2014. Annotating causality in the tempeval-3 corpus. In Proceedings of the EACL 2014 Workshop on Computational Approaches to Causality in Language (CAtoCL), pages 10--19

    2014

  12. [20]

    Eric Mitchell, Charles Lin, Antoine Bosselut, Chelsea Finn, and Christopher D Manning. 2021. Fast model editing at scale. arXiv preprint arXiv:2110.11309

    2021 arXiv

  13. [21]

    Jingcheng Niu, Saifei Liao, Victoria Ng, Simon De Montigny, and Gerald Penn. 2024. Contempo: A unified temporally contrastive framework for temporal relation extraction. In Findings of the Association for Computational Linguistics ACL 2024, pages 1521--1533

    2024

  14. [22]

    Judea Pearl. 2022. Direct and indirect effects. Probabilistic and Causal Inference: The Works of Judea Pearl, page 373

    2022

  15. [23]

    Colin Raffel, Noam Shazeer, Adam Roberts, Katherine Lee, Sharan Narang, Michael Matena, Yanqi Zhou, Wei Li, and Peter J Liu. 2020. Exploring the limits of transfer learning with a unified text-to-text transformer. Journal of machine learning research, 21(140):1--67

    2020

  16. [24]

    Victor Sanh, Albert Webson, Colin Raffel, Stephen H Bach, Lintang Sutawika, Zaid Alyafeai, Antoine Chaffin, Arnaud Stiegler, Teven Le Scao, Arun Raja, et al. 2022. Multitask prompted training enables zero-shot task generalization. In ICLR 2022-Tenth International Conference on...

    2022

  17. [25]

    Xiaoming Shi, Siqiao Xue, Kangrui Wang, Fan Zhou, James Zhang, Jun Zhou, Chenhao Tan, and Hongyuan Mei. 2024. Language models can improve event prediction by few-shot abductive reasoning. Advances in Neural Information Processing Systems, 36

    2024

  18. [26]

    Fiona Anting Tan, Ali H \"u rriyeto g lu, Tommaso Caselli, Nelleke Oostdijk, Tadashi Nomoto, Hansi Hettiarachchi, Iqra Ameer, Onur Uca, Farhana Ferdousi Liza, and Tiancheng Hu. 2022. The causal news corpus: Annotating causal relations in event sentences from news. In Proceedin...

    2022

  19. [27]

    Zhengwei Tao, Zhi Jin, Yifan Zhang, Xiancai Chen, Xiaoying Bai, Yue Fang, Haiyan Zhao, Jia Li, and Chongyang Tao. 2024. A comprehensive evaluation on event reasoning of large language models. arXiv preprint arXiv:2404.17513

    2024 arXiv

  20. [28]

    Zhengwei Tao, Zhi Jin, Haiyan Zhao, Chengfeng Dou, Yongqiang Zhao, Tao Shen, and Chongyang Tao. 2023. Unievent: Unified generative model with multi-dimensional prefix for zero-shot event-relational reasoning. In Proceedings of the 61st Annual Meeting of the Association for Com...

    2023

  21. [29]

    Jin Wang, JinFei Wang, Shuying Dai, Jiqiang Yu, and Keqin Li. 2024. Research on emotionally intelligent dialogue generation based on automatic dialogue system. Journal of Computer Technology and Applied Mathematics, 1(1):1--5

    2024

  22. [30]

    Xiaozhi Wang, Yulin Chen, Ning Ding, Hao Peng, Zimu Wang, Yankai Lin, Xu Han, Lei Hou, Juanzi Li, Zhiyuan Liu, et al. 2022. Maven-ere: A unified large-scale dataset for event coreference, temporal, causal, and subevent relation extraction. In Proceedings of the 2022 Conference...

    2022

  23. [31]

    Jing Yang, Yu Zhao, Yang Linyao, Xiao Wang, and Fei-Yue Wang. 2024. Temprompt: Multi-task prompt learning for temporal relation extraction in rag-based crowdsourcing systems. arXiv preprint arXiv:2406.14825

    2024 arXiv

  24. [32]

    Yunzhi Yao, Peng Wang, Bozhong Tian, Siyuan Cheng, Zhoubo Li, Shumin Deng, Huajun Chen, and Ningyu Zhang. 2023. Editing large language models: Problems, methods, and opportunities. In Proceedings of the 2023 Conference on Empirical Methods in Natural Language Processing, pages...

    2023

  25. [33]

    Chenhan Yuan, Qianqian Xie, and Sophia Ananiadou. 2023. Zero-shot temporal relation extraction with chatgpt. In The 22nd Workshop on Biomedical Natural Language Processing and BioNLP Shared Tasks, pages 92--102

    2023

  26. [34]

    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...

  27. [35]

    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 10, 2026 · model on record in the stance chip above.