The Preposition Corpus in Sketch Engine

Corpora from the Pattern Dictionary of English Prepositions (PDEP) provide the basis for examining the behavior of 304 English prepositions, with 1040 senses (patterns) describing in 20 fields. The PDEP corpora comprise dependency parses for 81,509 sentences in CoNLL-X format, previously used in several studies, particularly used for disambiguation modeling. We have now put these parse data into Sketch Engine (SE), using its mechanisms for further perspectives of preposition behavior. In the process, we have also annotated each parse with additional information that provides an even richer set of data to examine preposition behavior. For each sentence, Sketch Engine identifies the sense number and the direct link location to the PDEP sense description; these references can be displayed for each concordance line. Sketch Engine data for each sentence also includes the PDEP class, the subclass, and supersense tags for the preposition complements and governors (i.e., semantic relations using the WordNet lexicographer file class). We describe in detail how the corpora were prepared for SE, involving several scripts used in PDEP to access its databases. Some of these scripts provide additional entry points into the PDEP data, particularly for the class and subclass. We describe the use of WordNet noun, verb, adjective, and adverb supersenses to tag the complements and governors, i.e., semantic word sketches for the prepositions. The resultant preposition data within SE provides a perspective different from its usual focus on the main parts of speech, so we describe the unique aspects enabled for the PDEP corpora in considerable detail. We describe several corpus query language (CQL) queries that provide useful perspectives on preposition behavior, particularly showing preposition collocations, (semantic) word sketches, preposition thesauruses, and preposition sketch differences.