%0 Conference Paper
%A Tran, Mai Vu
%A Le, Hoang Quynh
%A Phi, Van Thuy
%A Pham, Thanh Binh
%A Nigel, Collier
%B SW4PHD: the 2016 Scientific Workshop for PhD Students
%C Hanoi
%D 2016
%F SisLab:1550
%T Exploring a Probabilistic Earley Parser for Event Composition in Biomedical Texts
%U https://eprints.uet.vnu.edu.vn/eprints/id/eprint/1550/
%X We describe a high precision system for extracting events of biomedical significance that was developed during the BioNLP shared task 2013 and tested on the Cancer Genetics data set. Our system explored a multi-stage approach including trigger detection, edge detection and event composition. After trigger edge detection is finished we are left with a semantic graph from which we must select the optimal subset that is consistent with the semantic frames for each event type.  The system achieved an F-score on the development data of 73.67 but was ranked 5th out of six with an F-score of 29.94 on the test data. How-ever, precision was the second highest ranked on the task at 62.73. Analysis suggests the need to continue to improve our system for complex events particularly taking into account cross-domain differences in argument distributions.