In the NLP literature, the thematic fit estimation task is defined as the task where, given a verb-specific role (e.g. the agent of to arrest) and a candidate argument (e.g. cop), a system has to predict how likely is the latter to be a good filler for the former (Santus et al., 2017). Despite the good number of contributions recently dedicated to thematic fit modeling, most current systems limit themselves to an evaluation in terms of correlation between their output and the human ratings for isolated verb-filler pairs (Sayeed et al., 2016), also because of the scarcity of benchmark datasets for the task. However, this approach does not account for the dynamic nature of argument expectations: there is robust psycholinguistic evidence that human update their predictions on upcoming arguments during sentence processing, depending on the way other verb arguments are filled (Bicknell et al., 2010; Matsuki et al., 2011). In this paper we introduce DTFit (Dynamic Thematic Fit), a dataset of human ratings for verb-role fillers in a given event context, with the aim of specifically addressing the issue of context-sensitive argument typicality.
@InProceedings{HSU18.6, author = {Chun-hsien Hsu and Chia-ying Lee}, title = {Frequency and Predictability Effects in Natural Reading: Evidence from Co-registration of Eye-movement and Event-related Potentials Measures}, 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