/2020grammarly_ws

Workshop on Shallow Discourse Parsing at Grammarly, Kyiv, 2020

Primary LanguageJupyter Notebook

Workshop on Shallow Discourse Parsing at Grammarly, Kyiv, 2020

Workshop announcement

Schedule

10:15-13:30 - Part 1: The theory of shallow discourse structure

13:30-14:30 - lunch

14:30-17:45 - Part 2: Shallow Discourse parsing

Preparation

Please download or clone this repository.

Please download/clone and install the baseline PDTB parser discopy.

The scripts in this repository (and in the parser) use Python3, and need the NLTK package installed. We will be working in a Jupyter notebook.

Finally, we will be using the following dataset in two formats (passwords are distributed separately). Please note that the data must be deleted after completion of the workshop.

Part 1: The theory of shallow discourse structure

Theory of discourse coherence

slides

Discourse Annotations

In-class exercise: What is a connective? Identifying connectives in Ukranian

PDTB exploration

Part 2: Shallow discourse parsing

Introduction to shallow discourse parsing

Error analysis on existing parsers

Implementing a discourse parser

Modify/play around with a connective detection module