This paper presents new hand-corrected deep syntactic annotations for the sentences in two broad-coverage psycholinguistic datasets: the Dundee eye-tracking corpus (Kennedy et al., 2003) and the Natural Stories self-paced reading corpus (Futrell et al., 2017). These texts are more ecologically valid than experiment-specific constructed stimuli, allowing researchers to probe the sentence comprehension process in a naturalistic setting. Deep syntactic annotations such as categorial grammars allow direct access to phenomena like non-local or conjoined semantic argument dependencies which are relevant to many questions about sentence processing but are difficult to compute from common markup frameworks such as Penn Treebank or Universal Dependencies. Previously no gold-standard deep syntactic markups have been available for either Dundee or Natural Stories. The deep syntactic representation used for the proposed annotations (Nguyen et al., 2012) has been shown to (1) facilitate direct extraction of long-distance dependencies as well as many other syntactic constructions of interest, (2) support accurate automatic parsing, and (3) generate surprisal estimates that correlate with measures of processing difficulty (van Schijndel and Schuler, 2015). These annotations can be used for any psycholinguistic inquiry in which predictors must be computed from latent syntax trees.
@InProceedings{SHAIN18.9, author = {Cory Shain and Marten Van Schijndel}, title = {Deep Syntactic Annotations for Broad-Coverage Psycholinguistic Modeling}, booktitle = {Proceedings of the Eleventh International Conference on Language Resources and Evaluation (LREC 2018)}, year = {2018}, month = {may}, date = {7-12}, location = {Miyazaki, Japan}, editor = {Barry Devereux and Ekaterina Shutova and Chu-Ren Huang}, publisher = {European Language Resources Association (ELRA)}, address = {Paris, France}, isbn = {979-10-95546-08-5}, language = {english} }
Summary of the paper