Skip to content

Code for Learning idiolectal style variation in online register

Notifications You must be signed in to change notification settings

lingjzhu/idiolect

Repository files navigation

Idiosyncratic but not Arbitrary: Learning Idiolects in Online Registers Reveals Distinctive yet Consistent Individual Styles

To appear in EMNLP 2021 main conference. arXiv

Requirements

python >= 3.7
transformers >= 4.0
torch >= 1.7

Data & Pre-trained weights

The data and pre-trained models used in the study can be found here.

Training a model

python model_roberta_self_attention.py --training_data ./authorship/Amazon_train_contrastive --develop_data ./authorship/Amazon_dev_contrastive --train --tau_low 0.4 --tau_high 0.6 --loss modified_anchor --mask_prob 0.1 --test_data ./authorship/Amazon_test_contrastive --test --save_test_data --distance cosine --epochs 6 --alpha 30

Evaluating style embeddings

See analysis/analogy.ipynb

About

Code for Learning idiolectal style variation in online register

Resources

Stars

Watchers

Forks

Releases

No releases published

Packages

No packages published