Variation in nominal and adjectival property concepts
Time: 05:45pm - 07:15pm
Place: Ballantine Hall 344
Andrew Koontz-Garboden (University of Manchester)A common belief, both traditionally and in the functional-typological literature, is that lexical categoryhood has some kind of meaning, with e.g., verbs (prototypically) naming (transient) actions, nouns naming (time-stable) things, etc. (see e.g., Givon 1984, Langacker 1987, etc.). Such ideas often come under criticism, however (e.g., Newmeyer 1998, Baker 2003), in part due to lack of formal articulation of key notions, which can make it difficult to identify falsifiable predictions. It might be expected that this is a kind of problem that the model-theoretic semantics literature could shed some light on, given that one of its goals is articulating in a formally precise fashion specific meanings for the constituents of semantic composition. There has been, however, little discussion in this literature on the semantic typology of lexical categories. In this talk, I report on preliminary work aimed at addressing this question.
In category: Morphosyntax and semantics
Evaluating parse error detection across varied conditions
Time: 03:00pm - 04:00pm
Place: Ballantine 015
Amber Smith and Markus DickinsonWe investigate how parse error detection methods work under real-world conditions, outlining and testing different variables for parse error detection evaluation and pointing to useful experimental conditions and evaluation metrics. In particular, we focus on four different conversion methods, ten different training data sizes, two parsers, and two error detection methods. By comparing a set number of tokens across conditions, we are able to use error detection precision and revised labeled attachment scores, in order to see the effect of each of the variables. We demonstrate the overwhelming importance of accounting for training data size (cf. parser quality) and to some extent conversion scheme. Most importantly, we provide a useful framework for evaluating error detection and thus helping build very large annotated corpora.
In category: Computational linguistics
JEvents v3.0.9 Stable
Copyright © 2006-2013
Powered by Joomla!®
Copyright © 2008 The Trustees of Indiana University | Copyright Complaints