Papers about Commonsense Causality

This is the paper list for the literature summarized in our survey published in EMNLP 2024: The Odyssey of Commonsense Causality: From Foundational Benchmarks to Cutting-Edge Reasoning

Citation 🫡

If you find our survey useful for your research on commonsense causality, please support us by citing our work as follows:

@inproceedings{cui-etal-2024-odyssey,
    title = "The Odyssey of Commonsense Causality: From Foundational Benchmarks to Cutting-Edge Reasoning",
    author = {Cui, Shaobo  and
      Jin, Zhijing  and
      Sch{\"o}lkopf, Bernhard  and
      Faltings, Boi},
    editor = "Al-Onaizan, Yaser  and
      Bansal, Mohit  and
      Chen, Yun-Nung",
    booktitle = "Proceedings of the 2024 Conference on Empirical Methods in Natural Language Processing",
    month = nov,
    year = "2024",
    address = "Miami, Florida, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://aclanthology.org/2024.emnlp-main.932",
    pages = "16722--16763",
}

Table of Contents

• Main Part of This Paper List
  • Handbook for researchers interested in commonsense causality
  • Benchmarks
    • Classification by Commonsense Types
    • Overview of Causal Datasets
  • Acquisition Methods over Commonsense Causality
    • Extractive Methods
    • Generative Methods
    • Manual Annotation Methods
    • Comparison of Methods
  • Reasoning Methods over Commonsense Causality
    • Qualitative Causal Reasoning
    • Quantitative Causal Reasoning
• If You Want to Know More Details

Main Part of This Paper List

Handbook for researchers interested in commonsense causality

Benchmarks

Classification by Commonsense Types

According to the commonsense types (see Appendix for more background on commonsense types), causality can be roughly classified into four categories:

1. Physical Causality
   Physical causality refers to the cause-effect relationships grounded in the physical world. It typically covers domains such as physics, chemistry, and environmental science. Example datasets include: CRAFT (Ates et al., 2022), e-CARE (Du et al., 2022).

2. Social Causality
   Social causality involves understanding social norms, cultures, human behavior, intents, and reactions. For instance, criticism (cause) can lead to depression (effect) in a social context. It covers domains like law, culture, education, and psychology. Example datasets include: ATOMIC (Sap et al., 2019), GLUCOSE (Mostafazadeh et al., 2020), IfQA (Yu et al., 2023), etc.

3. Biological Causality
   Biological causality relates to cause-effect pairs that govern biological processes and phenomena, such as how a healthy diet contributes to longevity. Example datasets include: BioCause (Mihuailua et al., 2013), CBND (Boue et al., 2015), etc.

4. Temporal Causality
   Temporal causality involves the sequential understanding that a cause must precede an effect in time. Example datasets include: Temporal-Causal (Bethard et al., 2008), CausalTimeBank (Mirza et al., 2014), CaTeRs (Mostafazadeh et al., 2016), etc.

Overview of Causal Datasets

Dataset	Annotation Unit	#Overall	#Causal	C.F.	Commonsense Types	Brief Introduction	License
First-Principle Causality
CauseEffectPairs Mooij et al., 2016	Variable	108	108	-	General	108 different cause-effect pairs selected from 37 datasets covering domains like meteorology, economy, medicine, engineering, biology. Focuses on the causal discovery problem (deciding whether X causes Y or Y causes X).	FreeBSD
IHDP Shalit et al., 2017	Variable	2,000	2,000	½	Biological	IHDP is the Infant Health and Development Program dataset, focusing on the effect of home visits on cognitive test scores for infants.	Custom Dataset Terms
CRAFT Ates et al., 2022	Video	58,000	-	Full	Physical	A video question-answering dataset requiring comprehension of physical forces and object interactions. Contains descriptive and counterfactual questions.	MIT
Commonsense Causality in Text Format
Temporal-Causal Bethard et al., 2008	Clause	1,000	271	-	Temporal	A corpus of 1,000 event pairs covering both temporal and causal relations.	Missing
CW Ferguson & Sanford, 2008	Clause	128	128	Full	General	CW is collected from psycholinguistic experiments and includes counterfactual examples.	Missing
SemEval07-T4 Girju et al., 2007	Phrase	220	114	-	General	Focuses on semantic analysis and automatic recognition of relations between word pairs, including causal relations.	Missing
SemEval10-T8 Hendrickx et al., 2010	Phrase	10,717	1,331	-	General	Similar to SemEval07-T4, focuses on classification of semantic relations between pairs of nominals, including cause-effect relations.	CC BY 3.0 Unported
COPA Roemmele et al., 2011	Sentence	2,000	1,000	-	General	Each question has a premise and two plausible causes/effects, with the correct one being more plausible.	BSD 2-Clause
EventCausality Do et al., 2011	Clause	583	583	-	General	A causality corpus built by detecting causality between events using discourse connectives.	Missing
BioCause Mihuailua et al., 2013	Clause	851	851	-	Biological	Contains 851 causal relations from 19 biomedical journal articles in infectious diseases.	Creative Commons
CausalTimeBank Mirza et al., 2014	Sentence	318	318	-	Temporal	Timebank corpus with causal samples taken from TempEval-3 corpus.	CC BY-NC-SA 3.0
CaTeRs Mostafazadeh et al., 2016	Sentence	2,502	308	-	Temporal	Causal and temporal relations annotated from ROCStories corpus.	Missing
AltLex Hidey & McKeown, 2016	Clause	44,240	4,595	-	General	An open class of markers that contains causality.	Missing
BECauSE 2.0 Dunietz et al., 2017	Sentence	729	554	-	General	Focuses on causal relations and other co-existing relations.	MIT
ESL Caselli & Vossen, 2017	Sentence	2,608	2,608	-	Temporal	A corpus for detecting causal and temporal relations.	CC BY 3.0 Unported
PDTB Webber et al., 2019	Clause	7,991	7,991	-	General	Marks discourse relations, including causation, grounded in explicit words or phrases.	LDC User Agreement
TimeTravel Qin et al., 2019	Sentence	109,964	29,849	½	General	Contains original stories, counterfactual facts, and new storylines compatible with the counterfactual facts.	MIT
GLUCOSE	Clause	670K	670K	-	Social	Annotates 10 dimensions of causal explanation from short stories, focusing on implicit causes and effects.	Creative Commons Attribution-NonCommercial 4.0
XCOPA Ponti et al., 2020	Sentence	11,000	11,000	-	General	Multilingual version of the COPA dataset, spanning 11 languages.	CC BY 4.0
SemEval20-T5 Yang et al., 2020	Clause	25,501	25,501	Full	General	Dataset for determining counterfactual statements and extracting antecedents and consequents.	Missing
CausalBank Li et al., 2021	Clause	314M	314M	-	General	Cause-effect statements collected from the Common Crawl corpus using causal lexical patterns.	-
e-CARE Du et al., 2022	Sentence	21,324	21,324	-	Physical	Cause-effect pairs and conceptual explanations for causation.	MIT
CoSIm Kim et al., 2022	Image + Text	3,500	3,500	Full	General	A multimodal counterfactual reasoning dataset for commonsense scene imagination, with both text and image components.	MIT
CRASS Frohberg & Binder, 2022	Sentence	274	274	Full	General	Focuses on counterfactual reasoning in a question-answering format.	Apache 2.0
IfQA Yu et al., 2023	Sentence	3,800	3,800	Full	Social	Open-domain counterfactual question-answering dataset.	Missing
CW-extended Li et al., 2023	Sentence	10,848	10,848	Full	General	Augmentation of CW dataset through word replacements, focusing on counterfactual statements.	Missing
CausalQuest Ceraolo et al., 2024	Sentence	13,500	13,500	½	General	A dataset of natural causal questions collected from social networks, search engines, and AI assistants.	Apache 2.0
δ-CAUSAL Cui et al., 2024	Sentence	11,245	11,245	½	General	Causal dataset exploring defeasibility and uncertainty in commonsense causality.	MIT
Commonsense Causality in Knowledge Graph Format
CausalNet Luo et al., 2016	Word	11M	11M	-	General	Vast collection of causal relationships from Bing web pages.	Missing
ConceptNet Speer et al., 2017	Phrase	473,000	-	-	General	Knowledge graph version of the Open Mind Common Sense project, including causal relations.	CC BY-SA 4.0
Event2Mind Rashkin et al., 2018	Phrase	25,000	-	-	Social	Annotates intent and reactions to given events, including causal relations.	MIT
ATOMIC Sap et al., 2019	Sentence	877K	-	½	Social	Collects commonsense knowledge in the form of "if-then" relations.	CC BY 4.0
ASER Zhang et al., 2020	Sentence	64M	494K	-	General	Eventuality knowledge graph extracted from large textual data.	MIT
CauseNet Heindorf et al., 2020	Word	11M	11M	-	General	Causal relations extracted from web sources with 83% precision.	CC BY 4.0
CEGraph Li et al., 2021	Phrase	89.1M	89.1M	-	General	Large lexical causal knowledge graph associated with CausalBank.	Missing

Acquisition Methods over Commonsense Causality

Extractive Methods

1. Sakaji et al. (2008): "Extracting Causal Knowledge Using Clue Phrases and Syntactic Patterns", paper.
2. Cole et al. (2006): "A lightweight tool for automatically extracting causal relationships from text, paper.
3. Girju (2003): "Automatic Detection of Causal Relations for Question Answering", paper.
4. Blanco et al. (2008): "Causal Relation Extraction", paper.
5. Zhao et al. (2016): "Event Extraction Using Causal Connectives", paper.

Generative Methods

1. Rashkin et al. (2018): "Event2Mind: Commonsense Inference on Events, Intents, and Reactions", paper.
2. Choudhry (2020): "Narrative Generation to Support Causal Exploration of Directed Graphs", paper.
3. Li et al. (2021): "Guided Generation of Cause and Effect", paper.
4. Kim et al. (2023): "SODA: Million-scale Dialogue Distillation with Social Commonsense Contextualization", paper.

Manual Annotation Methods

1. Palmer et al. (2005): "The Proposition Bank: An Annotated Corpus of Semantic Roles", paper.
2. Mihaila et al. (2013): "What causes a causal relation? Detecting Causal Triggers in Biomedical Scientific Discourse", paper.
3. Dunietz (2018): "Annotating and automatically tagging constructions of causal language", thesis.
4. Mostafazadeh et al. (2016): "CaTeRS: Causal and Temporal Relation Scheme for Semantic Annotation of Event Structures", paper.
5. Mirza et al. (2014): "Annotating Causality in the TempEval-3 Corpus", paper.

Comparison of Methods

Method	Accuracy	Cost	Coverage	Explainability
Extractive	★★★★☆	★★★★★	★★★★☆	★★★★☆
Generative	★★★☆☆	★★★★☆	★★★☆☆	★☆☆☆☆
Manual Annotation	★★★★★	★★☆☆☆	★★★★★	★★★★★

Reasoning Methods over Commonsense Causality

Qualitative Causal Reasoning

Qualitative causal reasoning focuses on classifying cause-effect relationships in a binary fashion, often bypassing uncertainty through simplification.

1. Jin et al. (2023): "CLadder: Assessing Causal Reasoning in Language Models", paper.
2. Zhang et al. (2022): "ROCK : Causal Inference Principles for Reasoning about Commonsense Causality", paper.
3. Ning et al. (2018): "Joint Reasoning for Temporal and Causal Relations", [paper] (https://arxiv.org/abs/1808.09506).
4. Zhang and Foo (2001): "Embedding Logic Rules into Causal Reasoning Mechanisms."
   • Link to paper

Quantitative Causal Reasoning

Quantitative causal reasoning provides numerical estimates for causal effects, accounting for uncertainty and variability in causal relationships.

1. Good (1961): "Log-Likelihood Metric for Causality Strength."
   • Link to paper
2. Suppes (1973): "A Probabilistic Theory of Causality."
   • Link to paper
3. Eells (1991): "Probabilistic Causality and Its Applications."
   • Link to paper
4. Pearl (2009): "Causality: Models, Reasoning, and Inference."
   • Link to paper
5. Luo et al. (2016): "CEQ: A Word-Level Causal Estimation Metric."
   • AAAI paper
6. Cui et al. (2024): "CESAR: A Weighted Approach to Measuring Causal Strength in Text."
   • ACL Findings paper

If You Want to Know More Details