Skip to main content

BibTeX

@book{wilhelmsson-2010-heuristisk-132135,
	title        = {Heuristisk analys med Diderichsens satsschema – Tillämpningar för svensk text, 2 uppl},
	abstract     = {A heuristic method for parsing Swedish text, heuristic schema parsing, is described and implemented. Focusing on main clause (primary) analysis, a collection of licensing techniques for removing non-primary verb candidates is employed, leaving e.g. the primary verbs, particles and conjunctions (bounded key constituents) that delimit the content of the fields in Diderichsen’s sentence schema. Hereby, the subsequent identification of constituents which do not have an upper bound on their length (subject, object/predicatives and adverbials) can be identified relying to a lesser on extent explicit pattern matching, and more on different heuristic rules. For phrase type identification and delimitation of these constituents, when adjacent to each other, a novel chunking technique, rank-based chunking, is applied. Following this, a series of further rules merge chunks into larger ones, aiming at a final number of nominal chunks compatible with the valency information of the main verb. The aim is to identify full nominal and adverbial constituents, including post-modifiers. The implementation uses the Stockholm Umeå Corpus 2.0, a corpus which is balanced for different genres in published Swedish text. SUC’s tagset is also used unmodified in part-of-speech tagging which enables the program to deal with input text. The functional parsing, which includes no explicit language-defining grammar component is carried out technically using an object-based representation of clause structure.

The thesis work also includes two prototype applications, both requiring high accuracy of the sort of functional syntactic analysis described here. The first one is an implementation of automatic syntactic fronting in the area of text editing for Swedish, where the user is presented with a syntactically analyzed copy of her writing, from which paraphrases easily can be generated. The second application is in the field of natural language query systems and produces questions with answers from an arbitrary declarative input text. This prototype incorporates a text database from Swedish Wikipedia, and investigates primarily generation of WH-questions formed via fronting of unbounded primary constituents. The questions are generated as a text is opened and thus permits users to only ask the available ones, thus aiming at a high precision value.
},
	author       = {Wilhelmsson, Kenneth},
	year         = {2010},
	publisher    = {University of Gothenburg},
	address      = {Göteborg},
	ISBN         = {978-91-977196-9-8},
}