Towards hybrid human-machine scientific information extraction

A wealth of valuable research data is locked within the millions of research articles published every year [1]. Extracting pertinent scientific facts (e.g., materials properties, known variants in genomics, population statistics etc.) from those articles has become an unmanageable task for researchers. This problem hinders the advancement of science, making it difficult to build on existing results, avoid unnecessary repetition, and to translate results into applications. Moreover, since these data are often loosely encoded in esoteric scientific articles intended for human consumption, they are, in general, not machine accessible. Thus, it is not often tractable to develop studies that automatically leverage this valuable information.