IlliniMet: Illinois System for Metaphor Detection

This is IlliMet model for metaphor detection participating in The Second Workshop on Figurative Language Processing. The model takes contextualized representation from the RoBERTa model to predict metaphors. Besides, POS and external linguistic features are explored and incorporated into the model.

Requirements:

1. Python3;

2. Transformers by Huggingface.

• [VARIABLE] in the instructions below refers to the variable names that can be defined by users.

1. Data Preparation

Obtain the datasets (VUA or TOEFL). Extract from the train data the token ids, tokens, POS tags and metaphor labels, and save them to train_ids.txt, train_tokens.txt, train_pos.txt, and train_metaphor.txt respectively. These files are in such format that each row corresponds to one sentence, and each id/token/POS/label is separated by space in a row. Save VUA data to [DATA_DIR]/VUA/, and TOEFL data to [DATA_DIR]/TOEFL/.

Process the test data in the same way as the train data, and save test_ids.txt, test_tokens.txt and test_pos.txt to the folder [DATA_DIR]/VUA/ or [DATA_DIR]/TOEFL/.

2. Feature Preparation

(1) Prepare VUA features

Download and unzip VUA feature files, i.e., naacl_flp_skll_train_datasets.zip and naacl_flp_skll_test_datasets.zip, from Educational Testing Service GitHub repo. Adatpt features to the same format as input sentences.

Process VUA train features

python3 feature_data_helper.py
--feature_dir [DATA_DIR]/VUA/naacl_flp_skll_train_datasets/
--train_type train
--tok_id_fn [DATA_DIR]/VUA/train_ids.txt
--output_dir [DATA_DIR]/VUA/

Process VUA test features

python3 feature_data_helper.py
--feature_dir [DATA_DIR]/VUA/naacl_flp_skll_test_datasets/
--train_type test
--tok_id_fn [DATA_DIR]/VUA/test_ids.txt
--output_dir [DATA_DIR]/VUA/

(2) Prepare TOEFL features:

Download and unzip TOEFL feature files, i.e., toefl_skll_train_features.zip and toefl_skll_test_features_no_labels.zip Educational Testing Service GitHub repo. The

Process TOEFL train features:

python3 feature_data_helper.py
--feature_dir [DATA_DIR]/TOEFL/toefl_skll_train_features/
--train_type train
--tok_id_fn [DATA_DIR]/TOEFL/train_ids.txt
--output_dir [DATA_DIR]/TOEFL/

Process TOEFL test features:

python3 feature_data_helper.py
--feature_dir [DATA_DIR]/TOEFL/toefl_skll_test_features_no_labels/
--train_type test
--tok_id_fn [DATA_DIR]/TOEFL/test_ids.txt
--output_dir [DATA_DIR]/TOEFL/

3. Train Model for Metaphor Detection

(1) Train VUA Model

python3 run_metaphor_detection.py
--data_dir [DATA_DIR]/VUA
--model_type roberta
--model_name_or_path roberta-large
--output_dir [OUTPUT_DIR]/VUA/model/
--dataset VUA
--max_seq_length 256
--do_train
--evaluate_during_training
--do_lower_case
--per_gpu_train_batch_size 6
--per_gpu_eval_batch_size 18
--learning_rate 2e-5
--num_train_epochs 5.0
--warmup_steps 500
--seed [SEED]
--use_pos
--pos_vocab_size 43
--pos_dim [POS_DIM]
--use_features
--feature_dim 696

• DATA_DIR: the directory with data files.

• OUTPUT_DIR: the directory to save models

• SEED: a positive integer as random seed

• --use_pos: whether to use POS as input feature

• --use_features: whether to use external feature for classification

• POS_DIM: dimension for part-of-speech tag embedding

(2) Train TOEFL Model

python3 run_metaphor_detection.py
--data_dir [DATA_DIR]/TOEFL/
--model_type roberta
--model_name_or_path roberta-large
--output_dir [OUTPUT_DIR]/TOEFL/model/
--dataset TOEFL
--max_seq_length 256
--do_train
--evaluate_during_training
--do_lower_case
--per_gpu_train_batch_size 6
--per_gpu_eval_batch_size 18
--learning_rate 2e-5
--num_train_epochs 5.0
--warmup_steps 500
--seed [SEED]
--use_pos
--pos_vocab_size 43
--pos_dim [POS_DIM]
--use_features
--feature_dim 215

• DATA_DIR: the directory with data files.

• OUTPUT_DIR: the directory to save models

• SEED: a positive integer as random seed

• --use_pos: whether to use POS as input feature

• --use_features: whether to use external feature for classification

• POS_DIM: dimension for part-of-speech tag embedding

4. Prediction

VUA prediction from a single model

python3 run_metaphor_detection.py
--data_dir [DATA_DIR]/VUA
--model_type roberta
--model_name_or_path roberta-large
--output_dir [OUTPUT_DIR]/VUA/model/
--dataset VUA
--max_seq_length 256
--do_predict
--do_lower_case
--per_gpu_eval_batch_size 18
--use_pos
--pos_vocab_size 43
--pos_dim [POS_DIM]
--use_features
--feature_dim 696

• The prediction file is saved in [OUTPUT_DIR]/VUA/model/test_labels.txt

TOEFL prediction from a single model

python3 run_metaphor_detection.py
--data_dir [DATA_DIR]/TOEFL/
--model_type roberta
--model_name_or_path roberta-large
--output_dir [OUTPUT_DIR]/TOEFL/model/
--dataset TOEFL
--max_seq_length 256
--do_predict
--do_lower_case
--per_gpu_eval_batch_size 18
--use_pos
--pos_vocab_size 43
--pos_dim [POS_DIM]
--use_features
--feature_dim 215

• The prediction file is saved in [OUTPUT_DIR]/TOEFL/model/test_labels.txt

5. Emsemble Prediction

Repeat step 4 to train multiple models using different random seeds. Ensemble method is adopted to make predictions by taking majority votes among multiple trained models. Put prediction outputs from multiple models into RES_DIR/, and the ensemble method saves final predictions "ensemble_test_labels.txt" in RES_DIR/.

Ensemble VUA prediction

python3 ensemble_test.py 
--data_dir [DATA_DIR]/VUA/
--res_dir [RES_DIR]/VUA/

• DATA_DIR: the directory which contains test_ids.txt

• RES_DIR: the directory to save prediction files

Ensemble VUA prediction

python3 ensemble_test.py 
--data_dir [DATA_DIR]/TOEFL/
--res_dir [RES_DIR]/TOEFL/

• DATA_DIR: the directory to save test_ids.txt

• RES_DIR: the directory to save multiple prediction files

If you have questions, please contact Hongyu Gong ([email protected]).

If you use our code, please cite our work:

Hongyu Gong, Kshitij Gupta, Akriti Jain and Suma Bhat "IlliniMet: Illinois System for Metaphor Detection with Contextual and Linguistic Information", in Proceedings of the Second Workshop on Figurative Language Processing 2020 (pp. 146--153).

@inproceedings{gong-etal-2020-illinimet, title = "{I}llini{M}et: {I}llinois System for Metaphor Detection with Contextual and Linguistic Information", author = "Gong, Hongyu and Gupta, Kshitij and Jain, Akriti and Bhat, Suma", booktitle = "Proceedings of the Second Workshop on Figurative Language Processing", month = jul, year = "2020", address = "Online", publisher = "Association for Computational Linguistics", url = "https://www.aclweb.org/anthology/2020.figlang-1.21", pages = "146--153"}