|Author||Marc Luder, Simon Clematide|
|Title||Constructing a Constructional MWE Lexicon for Psycho-Conceptual Annotation: An Evaluation of CPA and DUELME for Lexicographic Description|
|Abstract||The German JAKOB lexicon provides a basis for the coding of patient narratives and is currently extended in the direction of a phraseological and construction-grammar resource. For this purpose, we will compare two formalisms for the representation of multiword expressions (MWE): The Dutch Electronic Lexicon of Multiword Expressions (DuELME, Grégoire 2009) and the verb patterns from Corpus Pattern Analysis (CPA, Hanks 2008). We are looking for a representation format which is human-readable, and equally adapted for natural language processing (NLP). The JAKOB lexicon is implemented in the OLIF format and currently contains 7000 entries. The MWEs investigated are verbal phraseologisms and originate from the corpora of three different clients, consisting of a total of more than 400 transcribed sessions.|
The narrative analysis method JAKOB is a tool for investigating everyday stories from psychotherapy transcripts (Boothe 2004). Stories are annotated on the basis of our predefined psycho-conceptual coding system represented in the lexicon. JAKOB allows formulating hypotheses about the client’s conflicts, the analysis of the discourse being one component thereof.
DuELME is an NLP lexicon project which encodes MWE descriptions in a theory- and implementation-independent way. Every MWE is an instance of a construction class with elements including morpho-syntactic parameters. CPA patterns represent semantic properties for the elements of a (verbal) construction, whereas syntactic properties are represented in the JAKOB lexicon by the subcategorization frames (Satzmuster) of Wahrig (2007). We are implementing an additional lexicon property ‘bauplan’ which is formally constructed as a combination of the DuELME component list, the Wahrig subcategorization frame and semantic information out of the CPA-pattern. Because this structure is difficult to read for the lexicographer, it is generated automatically and can be hidden from the user, but is available for NLP tasks.
|Session||Computational Lexicography and Lexicology|