This project focuses on the development of a framework for interpreting linguistic descriptions of places and locations as well as objects in motion found in natural language texts. The resulting static and dynamic descriptions are represented in a spatiotemporal markup language called STML, which will be incorporated into a proposed international standard for spatial annotation called ISO-Space. The STML output then enables for grounding within a metric representation such as Google Earth, through an automatic conversion to KML. A Dynamic Interval Temporal Logic (DITL) is also developed, which is consistent with the STML output and which provides the semantics for STML for subsequent reasoning about the text.
In order to automatically capture locations, paths, and motion constructions in the text, spatial processing algorithms are created. These algorithms build on earlier work on temporal processing as well as new and continuing work on identifying places, analyzing the internal structure of events with the Event Structure Lexicon, and adding paths with the PathFinder algorithm. Manual STML annotation is performed on a corpus of travel descriptions in order to evaluate these algorithms.
This work is directly relevant to current efforts to standardize semantic annotations while also creating interoperable resources. One major impact of this research is the engagement of several diverse communities and their resources into a new dialogue and sharing of ideas. These include the areas of computational semantics and linguistics, qualitative reasoning, and dynamic approaches to logic and reasoning.
The research conducted under this funding involved the development of a general computational framework for capturing how motion isexpressed in natural language. This entailed four specific subtasks: The development of a formal framework for modeling the dynamicproperties of objects, called Dynamic Interval Temporal Logic(DITL). This integrates a first order interval temporal logic with arestricted subset of dynamic logic. This language distinguishesthe various ways in which language refers to motion . The design and implementation of a specification language formovement, called Spatiotemporal Markup Language (STML), that capturesthe representations developed in (1) above. This was later folded intoa newly approved ISO standard for spatial annotation in language,called ISOspace. The creation of a corpus annotated with spatial and spatiotemporalmarkup, to use as a gold standard dataset for training machinelearning algorithms for identifying locations, paths, and motion intext. The initial corpus developed, MotionBank, was recently foldedinto SpaceBank, and has formed the basis for training and testing forSemEval 2015 Task 8 . The development of spatial processing algorithms that identifyplaces, locations of objects, paths, and qualitative relations betweenobjects. The baseline classifiers developed in the context of Semeval2015, SpaceFinder, have been enhanced to the status of robust spatialrecognition algorithms in their own right. One of the major accomplishments of this research is the developmentof a novel computational theory of motion as expressed in language.The theory synthesizes well known linguistic observations regardinghow languages express motion with models of qualitative spatiotemporalreasoning that have been largely overlooked by the computationallinguistics community. The product of this integration is The dynamicinterval temporal logic, DITL, allows for direct and transparentinferences relating to spatial and spatiotemporal relations inlanguage. This is important for both the lexical semantic encoding ofpredicates (for interpretationa and translation), as well as fortextual entailment and inference tasks. The broader impact of this research is due to the fact that this newrepresentation has been encoded as a markup language for creatingannotated datasets for machine learning and other algorithm use. TheISOspace specification language (now an approved ISO standard) bringsthe community closer to standardizing semantic annotations while alsocreating interoperable resources. One major impact of the proposedwork is the engagement of several diverse communities and theirresources into a new dialogue and sharing of ideas. These include theareas of computational semantics and linguistics, qualitativereasoning, and dynamic approaches to logic and reasoning.
