Spatial Modeling (Triple Extraction - Part 6)
CraigTrim 110000G799 Visits (3596)
Spatial Relations through Prepositions
What is a preposition?
Prepositions are function words and characteristically express spatial relations. Like any function word, prepositions are important for the structure they bring to a sentence.
A function word has little semantic content of its own and chiefly indicates a grammatical relationship. The extraction of semantics (that is, meaning) from English text is chiefly served by examination of content words. Function words serve an important role in determining the structure of the sentence, and demonstrating the relationships between content words. Function words can be considered the "glue" that holds the sentence together.
Prepositions that are expressive of spatial relationships contain important information that should form part of any triple extraction toolkit. There are about 150 prepositions in English.
The following table contains common prepositions that express spatial relationships. Italicized text is not generally indicative of spatial relationships. Other text like "past" can be ambiguous. And others, like "including", could be better used in partonomous modeling, rather than spatial relationships. However, semantic overlap should always be allowed for, since the ambiguity of English does not permit a great deal of precision.
Italicized prepositions are not representative of spatial relations.
These examples will be used through out this document for performing triple extraction when prepositions are present:
I can think of at least three options for dealing with prepositions in English.
The first option would be to construct a separate predicate for each predicate form:
This approach has two downsides: It creates an explosion of predicates and does not give us semantically useful information about the relationship. What does ":in" mean? That the drill bit is in the borehole? The drill bit can be in the bore hole? Might be?
The second option would be to cluster predicates into general categories, such as in (inside, into, within, through), on (upon, over, onto), under (underneath, below, beneath), etc.
Rather than construct a separate predicate for each preposition, or to merge prepositions into similar clusters, I recommend using a single predicate type and reifying with the normalized preposition form.
Each preposition can then be typed in a hierarchy (if needed). This would similar to the second approach where "within" might be part of the "in" cluster. However, this last step is optional.