rknaebel/discopy repository overview

⁠Discopy: Python Shallow Discourse Parsing

Discopy⁠ aims to provide implementations for feature-based and neural-based discourse parsers. It consists of a parser pipeline architecture that stacks individual parser components to continuously add discourse information. The focus is currently on explicit relations that were handled first in most pipelines. Further, the remaining sentence pairs without explicit sense relation are processed with the non-explicit component. The current implementation follows the Conll2016 data and implementation guidelines.

⁠Example Usage

Currently, the standard Lin et al. parser for the PDTB version 2 Conll2016 is supported via docker image lin-1.0. Incoming data (via REST API) is pre-processed with Stanza (1.5.0) to generate linguistic features such as POS tags, dependencies, and constituency parse trees.

• docker pull rknaebel/discopy:lin-1.0
• docker run -it --rm -p 8080:8080 rknaebel/discopy:lin-1.0

The container provides a REST API via fastapi. Checkout the documentation: http://localhost:8080/docs⁠

Similarly, the docker image 1.0 provides access to a neural discourse parser without relying on linguistic feature extraction.