FrameNet for Modeling Extraction from Coordinate Structures
- Mai Al-Khatib, Center for Cognitive Sciences, University of Minnesota, Minneapolis, Minnesota, United States
AbstractA non-probabilistic model of speakers’ competence regarding extraction from a coordinate structure, which was argued to be sensitive to how the conjuncts are connected to each other in discourse (e.g., Lakoff, 1986; Kehler 2002) is presented. The model makes use of Lakoff’s (1986) account of the acceptability of extraction from coordinate structures by adopting the Frame Semantics framework. Lakoff argues that acceptability of extraction is affected by the belonging of the conjuncts to certain scenarios (e.g. a natural sequence of events), something that is measurable in this framework. An algorithm that measures the degree of relatedness between two conjuncts by consulting FrameNet (the framework implementation) and quantifying the common frames they belong to is proposed and tested on sentences used in an acceptability judgement survey on extraction from coordinate structures (Harris, 2009). The model’s outcomes interact with the experimental conditions in predicting human judgements, providing initial support for the proposal.