For the third year in a row, students from the Language Technologies Institute were honored as among the best in the BioASQ Biomedical Semantic Question Answering Challenge (BioASQ 2017).
Khyathi Chandu, Aakanksha Naik, and Aditya Chandrasekar, all Master of Language Technology students enrolled in the Question Answering course during the spring semester, picked up on the work of previous LTI teams to compete in the competition, extending the codebase with the goal of improving performance on “ideal answer” (summarization) questions.
The BioASQ competition, held annually, tasks participating teams with the challenge of combing vast troves of biomedical literature in the PubMed database in order to correctly answer questions. The participants’ systems utilize Information Retrieval, Question Answering, Summarization and other techniques in an attempt to find both the “exact answer” (named entities) and “ideal answer” (paragraph-length answers in natural human language).
The CMU team participated in Test Batches 2, 3, 4, and 5, taking first place for both exact and ideal answer in Batches 4 and 5. The test dataset was released in five batches, each comprising about 100 questions.
This marks the third year in a row that students from the LTI have taken first place in more than one task.
The accomplishment didn’t come easily for this year’s competitors. “After starting out at the bottom of the heap, the team kept analyzing errors and improving their system, and I am happy to report that they made it to the top of the heap,” explained Eric Nyberg, professor in the Language Technologies Institute, who, along with Professor Teruko Mitamura, advised the team.
“I’m very proud of our students’ effort on this task,” Nyberg added.
Full results of this year’s challenge can be found on the BioASQ website.