Characterizing Text Revisions to Better Support Collaborative

Despite advancement in collaborative writing tools, the track changes capability in modern editors remains limited to highlighting syntactic changes, with authors still required to manually read through each of the revisions. We envision a collaborative authoring system where an author could accept all minor edits first and then focus on the substantial changes. To support this, we define the task of significant revision identification as the task of identifying the revisions between two versions of a text according to one of four categories, i.e. formal, meaning preserving, micro- and macro-structure. Micro-structure change corresponds to minor meaning change while macro-structure change corresponds to major meaning change. Our main contribution is to define a computational approach to this task, by framing the task as bi-directional entailment between the original and revised sentences. An existing recognition of textual entailment (RTE) system is applied to evaluate whether the revised texts entails. We evaluate the approach through a novel corpus consisting of multiple versions of drafts of academic papers written by multiple authors, which were annotated with the four revision types by both authors and non-authors of the papers. The proposed bi-directional textual entailment approach performs better than baseline edit distance approaches, which is similar to the current track changes capability built into most word processors.