Awesome

GF GrammarHelper

GF GrammarHelper is a library designed to facilitate the generation of Grammatical Framework (GF) grammar files tailored for various NLP tasks.

Before diving into the library, we highly recommend familiarizing yourself with the basics of GF by reading the Grammatical Framework Tutorial first to get a basic understanding of GF.

To reproduce the grammars used in the paper, please refer to the Reproduce the grammars needed for the paper section.

Overview

Incorporating the insights presented in this paper, formal grammar can be employed to represent a task-specific language. Leveraging a runtime that supports partial parsing, this formal grammar can incrementally parse sentences, guiding the production of the subsequent permissible tokens.

Using the GF GrammarHelper library, the process of employing formal grammar for constrained language generation is delineated into four steps:

Grammar Definition: Initiate the task-specific formal grammar by constructing a Grammar object (refer to src/grammar.py).
Grammar File Creation: Store the Grammar object into a GF grammar file(.gf) using the Grammar.save() method.
Compilation: Utilize the GF runtime to transform the GF grammar file into a .pgf file.
Constrained Text Generation: Rely on the Python GF wrapper to load the .pgf file and execute constrained text generation. For a practical demonstration, see the xxx repository.

Features:

Offers a abstraction of Grammar and ProductionRule to encapsulate a GF grammar.
Introduces three specialized modules: IE_grammar, ED_grammar, and CP_grammar. These are dedicated to generating GF grammar files for Information Extraction, Entity Disambiguation, and Constituency Parsing tasks, respectively.
Provides a GrammarFactory class to facilitate the creation of Grammar objects in batch.

Requirements

pip install -r requirements.txt

Add the root folder of this repository to PYTHONPATH environment variable.

# in the root folder of this repository
export PYTHONPATH="$(pwd):$PYTHONPATH"  # this is required to run scripts in `example/` folder

Reproduce the grammars needed for the paper

bash generate_gcd_grammars.sh