Improving Verb Phrase Extraction by Targeting Phrasal Verbs based on Valency Frames

In the Gender and Work project (GaW), historians are building a database with information on what men and women did for a living in the Early Modern Swedish society, i.e. approximately 1550–1800 [1]. This information is currently extracted by researchers manually going through large volumes of court records and church documents, searching for relevant text passages describing working activities. In this process, it has been noticed that working activities often are described in the form of verb phrases, such as hugga ved (”chop wood”), sälja fisk (”sell fish”) or tjäna som piga (”serve as a maid”). In previous work I have, in cooperation with the historians, developed a method for automatically extracting verb phrases from historical documents, by use of spelling normalisation, tagging and parsing [4]. With this approach, most of the verbs in historical Swedish texts are correctly extracted. Due to differences in word order and significantly longer sentences than in present-day Swedish texts, it is however still hard for the parser to correctly identify the precise complements associated with a verb in a sentence. The aim of my proposed project is to improve the verb phrase extraction results by providing verb valency information in the extraction step, with a special focus on phrasal verbs. In this project, I refer to phrasal verbs in its broader sense, including particle phrasal verbs such as äga rum (”take place”), prepositional phrasal verbs such as handla med (”trade with”), and the combination of these two as in betala igen till (”pay back to”).