Knowledge-based spatial reasoning for scene generation from text descriptions
AAAI'08 Proceedings of the 23rd national conference on Artificial intelligence - Volume 3
Spatial role labeling: Towards extraction of spatial relations from natural language
ACM Transactions on Speech and Language Processing (TSLP)
Relational learning for spatial relation extraction from natural language
ILP'11 Proceedings of the 21st international conference on Inductive Logic Programming
SemEval-2012 task 3: spatial role labeling
SemEval '12 Proceedings of the First Joint Conference on Lexical and Computational Semantics - Volume 1: Proceedings of the main conference and the shared task, and Volume 2: Proceedings of the Sixth International Workshop on Semantic Evaluation
Extracting Spatial Information From Place Descriptions
Proceedings of The First ACM SIGSPATIAL International Workshop on Computational Models of Place
Hi-index | 0.00 |
Understanding text is a trivial task for literate humans. For computers, however, it is extremely difficult due to (among other reasons) a lack of knowledge about language and the world, as well as an intelligent reasoning mechanism to process such resources. As a result, computational approaches to text understanding generally lack common sense and suffer from poor performance. This work presents a system that addresses a set of critical cognitive, linguistic, and computational issues. On the cognitive level, it considers the role of spatial characteristics of the external world. On the linguistic level, it considers the roles of underspecification, vagueness, uncertainty, context, and frame of reference in how humans communicate about space. On the computational level, it implements a constraint-based, declarative knowledge representation for qualitative spatial reasoning over the dimensions, positions, and orientations of representative objects (primarily animals and plants) in a simulated microworld of a zoo environment. This system extracts into a semantic network the explicit information in rudimentary text descriptions of static, spatial scenes, integrates it with implicit, background information from an object-oriented, commonsense knowledge base, reasons over the combined representation, and renders a set of corresponding graphical interpretations. From these depictions, it extracts new information that iteratively feeds back into the original description to augment the understanding further. As part of a Monte Carlo simulation, the architecture supports a multidimensional test-and-evaluation framework to investigate a variety of related issues that apply to many applications in artificial intelligence.