Follow
Area: Cross-Document NLP
Static, isolated, short texts have been the main focus of NLP research to date. However, many real-world tasks require humans to simultaneously work with multiple, connected, potentially long documents that change over time: from collaborative writing to fake news detection, and from peer review to social media management. While isolated, application-specific approaches to cross-document discourse modeling exist, the general NLP methodology for cross-document analysis is yet to be established.
Instead of treating each application scenario separately, the InterText initiative develops a joint, unified framework for cross-document modeling. Inspired by the theoretical works in literary and discourse studies, we propose a typology of general cross-document relations that might differ by their type, granularity and explicitness, aiming to cover a wide range of cross-document discourse phenomena. We use this typology to model cross-document discourse in diverse application scenarios.
Publications
Datasets and Code
The first openly licensed cross-document corpus in multi-domain, open, journal-style peer review. The corpus consists of peer review reports, manuscripts and their revisions in a wide range of domains. We provide multiple annotation layers that cover one full revise-and-resubmit cycle of academic publishing: pragmatic tagging determines the intention of reviewers' comments, linking connects peer reviews to their manuscripts, version alignment maps manuscripts to their next revision triggered by the peer review.
A pre-release of our general-purpose library for cross-document NLP modelling and analysis. Current version of the library provides converters from several document formats into a uniform data model, as well as an API for common graph operations that facilitate cross-document analysis on varying granularity levels. The library is constantly extended to cover more document formats and cross-document relation types, star the repo to stay up-to-date with the new releases!
