Flair JSON-NLP Wrapper

(C) 2019-2020 by Damir Cavar

Contributors to previous versions: Oren Baldinger, Maanvitha Gongalla, Anurag Kumar, Murali Kammili

Brought to you by the NLP-Lab.org!

Introduction

Flair v 4.5 wrapper for JSON-NLP. Flair provides state-of-the-art embeddings, and tagging capabilities, in particular, POS-tagging, NER, shallow syntax chunking, and semantic frame detection.

FlairPipeline

We provide a FlairPipeline class, with the following parameters for ease of use:

• lang: defaults to en. Different languages support different models, see the Flair Docs for details.
• use_ontonotes: defaults to False. Whether or not to use 4-class (True) or 12-class (False) NER tagging.
• fast: defaults to True. Whether or not to use the smaller, faster, but slightly less accurate versions of the models.
• use_embeddings: defaults to ''. Passing default will map to glove,multi-forward,multi-backward, the recommended stacked-embedding configuration.
• char_embeddings: defaults to False. Whether or not to include character-level embeddings.
• bpe_size: defaults to 0. If you want to include Byte-Pair Encodings, set this value to 50, 100, 200, or 300. See more at Flair Embeddings Docs.
• pos: defaults to True. Whether or not to include language-specific part-of-speech tags.
• sentinment: defaults to True. Whether or not to include sentiment analysis, if it is available for the given language.

Tagging and Embedding models are downloaded automatically the first time they are called. This may take a while depending on your internet connection.

Microservice

The JSON-NLP repository provides a Microservice class, with a pre-built implementation of Flask. To run it, execute:

python flairjsonnlp/server.py

Since server.py extends the Flask app, a WSGI file would contain:

from flairjsonnlp.server import app as application

The microservice exposes the following URIs:

• /expressions
• /token_list

These URIs are shortcuts to disable the other components of the parse. In all cases, tokenList will be included in the JSON-NLP output. An example url is:

http://localhost:5000/expressions?text=I am a sentence

Text is provided to the microservice with the text parameter, via either GET or POST. If you pass url as a parameter, the microservice will scrape that url and process the text of the website.

The additional Flair parameters can be passed as parameters as well.

Here is an example GET call:

http://localhost:5000?lang=de&constituents=0&text=Ich bin ein Berliner.