Explaining Answers with Entailment Trees
By Bhavana Dalvi, Peter Alexander Jansen, Oyvind Tafjord, Zhengnan Xie, Hannah Smith, Leighanna Pipatanangkura, Peter Clark. 2021
232 citations recorded by the source corpus.
Our goal, in the context of open-domain textual question-answering (QA), is to explain answers by showing the line of reasoning from what is known to the answer, rather than simply showing a fragment of textual evidence (a “rationale”). If this could be done, new opportunities for understanding and debugging the system’s reasoning become possible. Our approach is to generate explanations in the form of entailment trees, namely a tree of multipremise entailment steps from facts that are known, through intermediate conclusions, to the hypothesis of interest (namely the question + answer). To train a model with this skill, we created ENTAILMENTBANK, the first dataset to contain multistep entailment trees. Given a hypothesis (question + answer), we define three increasingly difficult explanation tasks: generate a valid entailment tree given (a) all relevant sentences (b) all relevant and some irrelevant sentences, or (c) a corpus. We show that a strong language model can partially solve these tasks, in particular when the relevant sentences are included in the input (e.g., 35% of trees for (a) are perfect), and with indications of generalization to other domains. This work is significant as it provides a new type of dataset (multistep entailments) and baselines, offering a new avenue for the community to generate richer, more systematic explanations.
Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Retrieval-Augmented Generation, Natural Language Processing
How researchers see this paper
The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.
In Artificial Intelligence
- Yejin Choi
- Mohit Bansal
- Hannaneh Hajishirzi
- Luke S. Zettlemoyer
- Percy Liang
- Noah A. Smith
- Jianfeng Gao
- Christopher D. Manning
In Computer Vision and Pattern Recognition
- Luc Van Gool
- Hannaneh Hajishirzi
- Trevor Darrell
- Yejin Choi
- Xiaogang Wang
- Dacheng Tao
- Ming-Hsuan Yang
- Jitendra Malik
In Retrieval-Augmented Generation
In Natural Language Processing
- Yejin Choi
- Hannaneh Hajishirzi
- Luke S. Zettlemoyer
- Mohit Bansal
- Greg Durrett
- Noah A. Smith
- Peter Clark
- Percy Liang
Citations in the corpus
- Flexible Generation of Natural Language Deductions (2021, methodology): “fact compositions in the validation splits of the QASC and EntailmentBank datasets (Khot et al., 2020; Dalvi et al., 2021).”
- Faithful Reasoning Using Large Language Models (2022, methodology): “We present results on both Proof Writer (PW) (Tafjord et al., 2021) and EntailmentBankQA (EB).”
- Selection-Inference: Exploiting Large Language Models for Interpretable Logical Reasoning (2022, background): “1a); 2) approaches that encourage LLMs to produce reasoning explicitly, but all reasoning steps are produced in one generative step (Cobbe et al., 2021; Dalvi et al., 2021; Jhamtani and Clark, 2020; Nye et al., 2021a; Wei et al., 2022; Zelikman et al., 2022) (e.g. Fig.”
- Learn to Explain: Multimodal Reasoning via Thought Chains for Science Question Answering (2022, background): “To address this issue, other science datasets annotate the explanations, but they are restricted to the textual only modality and limited to small data scales [13, 6, 37] or a small set of topics [20, 14].”
- LAMBADA: Backward Chaining for Automated Reasoning in Natural Language (2022, background): “Explicit Reasoning: Recently, approaches that generate the intermediate reasoning steps such as the chain (or tree) of reasoning (Wei et al., 2022; Nye et al., 2022; Dalvi et al., 2021; Zelikman et al., 2022; Zhang et al., 2022a) have shown massive improvement for many reasoning tasks (Suzgun et al.”
- NaturalProver: Grounded Mathematical Proof Generation with Language Models (2022, methodology): “Retrieved knowledge also turns out to be very helpful, and even comparable to human-provided knowledge in some metrics.”
- A Survey of Large Language Models (2023, unknown): “Complex Reasoning Knowledge Reasoning CSQA [247], StrategyQA [248], ARC [291], BoolQ [315], PIQA [295], SIQA [316], HellaSwag [317], WinoGrande [318], OpenBookQA [302], COPA [319], ScienceQA [301], proScript [320], ProPara [321], ExplaGraphs [322], ProofWriter [323], EntailmentBank [324], ProOntoQA ”
- A Multitask, Multilingual, Multimodal Evaluation of ChatGPT on Reasoning, Hallucination, and Interactivity (2023, unknown): “When we repeat the analysis on advanced tasks, specifically on CLUTRR (Sinha et al., 2019) for induction and En-tailmentBank for deduction (Dalvi et al., 2021), the same conclusion holds based on our experiment.”
- Solving math word problems with process- and outcome-based feedback (2022, unknown): “However, compared to GSM8K, these datasets either contain templated problems (Proof Writer and CLUTTR) or are significantly smaller in size (EntailmentBank and StrategyQA).”
- Honest Students from Untrusted Teachers: Learning an Interpretable Question-Answering Pipeline from a Pretrained Language Model (2022, background): “Another line of work has focused on training language models to perform reasoning by fine-tuning on gold reasoning traces (Bostrom et al., 2022; Creswell and Shanahan, 2022; Dalvi et al., 2021; Tafjord et al., 2021).”
- Shepherd: A Critic for Language Model Generation (2023, unknown): “Specifically, we choose Entailment Bank (deductive reasoning) (Dalvi et al., 2021), Proofwriter (logical reasoning) (Tafjord et al., 2020), GSM8K (arithmetic reasoning) (Cobbe et al., 2021), PIQA (physical reasoning) (Bisk et al., 2020b), CosmosQA (commonsense reasoning) (Huang et al., 2019), ECQA…”
- Natural Language Deduction through Search over Statement Compositions (2022, methodology): “Generative sequence-to-sequence models can be fine-tuned to carry out this task given example trees (Tafjord et al., 2021; Dalvi et al., 2021).”