Spatial relation extraction

About

Spatial relation extraction is the task of determining whether a given set of spatial elements are in a qualitative spatial relation (QSLINK), an orientation relation (OLINK), or a motion spatial relation (MOVELINK).

• QSLINKs exist between stationary spatial elements with a regional connection (e.g., in "The cup is on the table", the regions of "cup" and "table" are externally connected and hence they are involved in a QSLINK).

• OLINKs exist between stationary spatial elements expressing their relative or absolute orientations (e.g., in the above sentence, "cup" and "table" are involved in an OLINK, which conveys that "cup" is oriented above "table").

• And MOVELINKs, differing from QSLINKs and OLINKs which only relate stationary spatial elements, exist between spatial elements in motion (e.g., in "John walked from Boston to Cambridge", there is a MOVELINK involving mover "John", motion verb "walked", source "Boston", and goal "Cambridge").

Our participating system in SpaceEval 2015, which achieved the best spatial relation extraction results, is described in the following paper.

Jennifer D'Souza and Vincent Ng. June 2015. UTD: Ensemble-Based Spatial Relation Extraction. In Proceedings of SemEval 2015.

Following which, the system was further improved by incorporating tree kernels and multi-pass sieves, enabling the use of an expanding parse tree as a novel structured feature for training MOVELINK classifiers. This system is described in the paper below.

Jennifer D'Souza and Vincent Ng. September 2015. Sieve-Based Spatial Relation Extraction with Expanding Parse Trees. In Proceedings of EMNLP.

Download

This software provides code for two components:

• Learning spatial elements from unlabeled text: Download our Spatial Element Extractor (SpElementEx).

• Learning spatial relations from text labelled with spatial elements: Download our Spatial Relation Extractor (SpatialRelEx).

GitHub: Here are SpElementEx and SpatialRelEx GitHub sites.

Funding Statement

This work was supported in part by NSF Grants IIS-1147644 and IIS-1219142. Any opinions, findings, or conclusions expressed above are those of the authors and do not necessarily reflect the views or official policies of NSF.