Verb Physics: Relative Physical Knowledge of Actions and Objects

By Maxwell Forbes, Yejin Choi. 2017

87 citations recorded by the source corpus.

Learning commonsense knowledge from natural language text is nontrivial due to reporting bias: people rarely state the obvious, e.g., “My house is bigger than me.” However, while rarely stated explicitly, this trivial everyday knowledge does influence the way people talk about the world, which provides indirect clues to reason about the world. For example, a statement like, “Tyler entered his house” implies that his house is bigger than Tyler. In this paper, we present an approach to infer relative physical knowledge of actions and objects along five dimensions (e.g., size, weight, and strength) from unstructured natural language text. We frame knowledge acquisition as joint inference over two closely related problems: learning (1) relative physical knowledge of object pairs and (2) physical implications of actions when applied to those object pairs. Empirical results demonstrate that it is possible to extract knowledge of actions and objects from language and that joint inference over different types of knowledge improves performance.

Subfields: Artificial Intelligence, Computer Vision and Pattern Recognition, Multimodal Learning

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

In Computer Vision and Pattern Recognition

In Multimodal Learning

Citations in the corpus

SWAG: A Large-Scale Adversarial Dataset for Grounded Commonsense Inference (2018, background): “Physical knowledge Prior work has studied learning grounded knowledge about objects and verbs: from knowledge bases (Li et al., 2016), syntax parses (Forbes and Choi, 2017), word embed-dings (Lucy and Gauthier, 2017), and images and dictionary deﬁnitions (Zellers and Choi, 2017).”
DREAM: A Challenge Data Set and Models for Dialogue-Based Reading Comprehension (2019, background): “This might be due to the nature of dialogues: for efﬁcient oral communication, people rarely state obvious explicit world knowledge (Forbes and Choi, 2017) such as “Christmas Day is celebrated on December 25th” .”
“Going on a vacation” takes longer than “Going for a walk”: A Study of Temporal Commonsense Understanding (2019, background): “…in recent years and existing NLP works have mainly investigated the acquisition and evaluation of commonsense in the physical world, including but not limited to, size, weight, and strength (Forbes and Choi, 2017), roundness and deliciousness (Yang et al., 2018), and intensity (Cocos et al., 2018).”
Does it Make Sense? And Why? A Pilot Study for Sense Making and Explanation (2019, background): “Note that there has been dataset which focus on non-linguistic world knowledge plausibility (Wang et al., 2018) or only limited attributes or actions of physical knowledge like verbphysics (Forbes and Choi, 2017).”
Do NLP Models Know Numbers? Probing Numeracy in Embeddings (2019, background): “For instance, embeddings may capture the size of objects (Forbes and Choi, 2017), speed of vehicles, and many other “commonsense” phenomena (Yang et al.”
Birds Have Four Legs?! NumerSense: Probing Numerical Commonsense Knowledge of Pre-trained Language Models (2020, background): “Forbes and Choi (2017) and Goel et al. (2019) studied commonsense comparisons between two physical objects (e.g., a house is usually bigger than a person) in pre-trained word embeddings.”
G-DAug: Generative Data Augmentation for Commonsense Reasoning (2020, background): “notions that are seldom stated explicitly (Gordon and Van Durme, 2013; Forbes and Choi, 2017), and authoring such questions can require creativity.”
Extracting Commonsense Properties from Embeddings with Limited Human Guidance (2018, methodology): “Methods for learning this kind of common sense have been developed previously (e.g. Forbes and Choi, 2017), but the best-performing methods in that previous work requires dozens of manually-annotated frames for each comparison property, to connect the property to how it is indirectly reﬂected in…”
PROST: Physical Reasoning about Objects through Space and Time (2021, background): “Other datasets (Forbes and Choi, 2017; Elazar et al., 2019; Goel et al., 2019) focus on object– attribute comparison.”
Representing Numbers in NLP: a Survey and a Vision (2021, background): “VerbPhysics (Forbes and Choi, 2017) is a benchmark of binary comparisons between physical attributes of various objects, e.g., ball size tiger.”
ESPRIT: Explaining Solutions to Physical Reasoning Tasks (2020, background): “, humans rarely state events that are obvious (Forbes and Choi, 2017).”
Language to Action: Towards Interactive Task Learning with Physical Agents (2018, background): “However, except for a few works that look into physical properties of verbs [Forbes and Choi, 2017; Zellers and Choi, 2017], how verbs and their corresponding actions affect the state of the physical world is largely under-explored.”
CREAK: A Dataset for Commonsense Reasoning over Entity Knowledge (2021, background): “Our work relates to judging plausibility of events [Forbes and Choi, 2017, Wang et al., 2018], closely tied to inferences accessible from feature norms [McRae et al., 2005], but again these past efforts do not focus on judgments around speciﬁc entities.”
Temporal Common Sense Acquisition with Minimal Supervision (2020, background): “These works include but are not limited to size, weight, and strength (Bagherinezhad et al., 2016; Forbes and Choi, 2017; Elazar et al., 2019), roundness and deli-ciousness (Yang et al., 2018), and intensity (Cocos et al., 2018).”
How Large Are Lions? Inducing Distributions over Quantitative Attributes (2019, unknown): “There is a line of work (Forbes and Choi, 2017; Yang et al., 2018) to determine the quantitative relation between two nouns on a speciﬁc scale.”
SemEval-2020 Task 4: Commonsense Validation and Explanation (2020, background): “Some datasets focus on physical knowledge validation (Wang et al., 2018; Porada et al., 2019), or only limited attributes or actions of world knowledge (Forbes and Choi, 2017).”
Do Language Embeddings capture Scales? (2020, result): “We also show improvements by zero-shot transfer from our probes to 2 related tasks: relative comparisons (Forbes and Choi, 2017) and product price prediction (Jianmo Ni, 2019), indicating that our results are robust across datasets.”
Machine Common Sense Concept Paper (2018, background): “[38]) and Language (VERB PHYSICS [39]) [55], ‘mental’ simulation [56], and multi‐step reasoning [57].”
Do Language Embeddings capture Scales? (2020, methodology): “We also show improvements by zero-shot transfer from our probes to 2 related tasks: relative comparisons (Forbes and Choi, 2017) and product price prediction (Jianmo Ni, 2019), indicating that our results are robust across datasets.”