Abduction of Mental States with a Formal Theory of Commonsense Psychology (Abstract only) (bibtex)
by Gordon, Andrew S., Hobbs, Jerry R., Ovchinnikova, Hatya, Roemmele, Melissa and Morency, Louis-Philippe
Abstract:
Successful communication and collaboration between humans and intelligent agents of the future will require a robust ability to algorithmically infer the subjective mental states of the human participants. As in human to human interaction, the central concerns of plans, goals, emotions, and beliefs of another must inferred from a mix of explicit and implicit evidence in language, along with contextual and behavioral cues. We propose that this cognitive ability of mental model ascription is best conceived as a process of abduction, where a hypothetical explanation is inferred to account for observable evidence. In this approach, speech and other behavior of a person are observables that require explanation, where the challenge is to find a theoretical explanation that requires the fewest assumptions. Recent advances in abduction-based language processing [1] have led to efficient implementations of Hobbs's conception of weighted-abduction [2], where textual inputs (observations) are explained by searching a knowledgebase of logical axioms for the least-cost proof, with cost incurred when assumptions are asserted.
Reference:
Abduction of Mental States with a Formal Theory of Commonsense Psychology (Abstract only) (Gordon, Andrew S., Hobbs, Jerry R., Ovchinnikova, Hatya, Roemmele, Melissa and Morency, Louis-Philippe), In CogSci 2013, 2013.
Bibtex Entry:
@inproceedings{gordon_abduction_2013,
	address = {Berlin, Germany},
	title = {Abduction of {Mental} {States} with a {Formal} {Theory} of {Commonsense} {Psychology} ({Abstract} only)},
	url = {http://ict.usc.edu/pubs/Abduction%20of%20Mental%20States%20with%20a%20Formal%20Theory%20of%20Commonsense%20Psychology%20(Abstract%20only).pdf},
	abstract = {Successful communication and collaboration between humans and intelligent agents of the future will require a robust ability to algorithmically infer the subjective mental states of the human participants. As in human to human interaction, the central concerns of plans, goals, emotions, and beliefs of another must inferred from a mix of explicit and implicit evidence in language, along with contextual and behavioral cues. We propose that this cognitive ability of mental model ascription is best conceived as a process of abduction, where a hypothetical explanation is inferred to account for observable evidence. In this approach, speech and other behavior of a person are observables that require explanation, where the challenge is to find a theoretical explanation that requires the fewest assumptions. Recent advances in abduction-based language processing [1] have led to efficient implementations of Hobbs's conception of weighted-abduction [2], where textual inputs (observations) are explained by searching a knowledgebase of logical axioms for the least-cost proof, with cost incurred when assumptions are asserted.},
	booktitle = {{CogSci} 2013},
	author = {Gordon, Andrew S. and Hobbs, Jerry R. and Ovchinnikova, Hatya and Roemmele, Melissa and Morency, Louis-Philippe},
	month = jul,
	year = {2013},
	keywords = {The Narrative Group}
}
Powered by bibtexbrowser