# Proto role labeling (SPRL) with argument-predicate marker embedding
Span based SPRL approach. Given a sentence, a predicate and an argument, the model learns to highlight predicate and argument seperately and predicts up to 18 proto role properties (multi-label or Likert). A simple voter ensemble further improves results.
### Prerequisites
see requirements.yml
## Reproducing the multi-labeling results with Bert embeddings
```
cd src
./run_ensemble.sh
```
when ready
```
python check_test_scores.py
'''
results may slighlty vary from run to run (because of small data, random model inits)
### Running Likert only regression or manipulating/improving the model
General hyperparameters can be set in
```
src/config.py
```
The model can be modified in:
```
src/model.py
```
The training cycle can be improved in:
```
src/proto_model_bert.py
```
The data setup and preprocessing can be modified in
```
src/data_setup.py
```
E.g., for now we average the results of two random annotators in SPR2 data. We could determine trusty annotators and weigh their opinions more. This should improve SPRL for web-texts (i.e. SPR2 data).
## Citation
```
@inproceedings{opitzfrank:2019b,
author = {Juri Opitz and Anette Frank},
title = {{An Argument-Marker Model for Syntax-Agnostic Proto-Role Labeling}},
year = {2019},
publisher = {Association for Computational Linguistics},
booktitle = {Proceedings of The Eighth Joint Conference on Lexical and Computational Semantics (*SEM 2019)},
note = {to appear}
}
```
## SPR data
when using SPR data do not forget to cite the creators. see the readme in src/resources
### For usage, please cite (at least the first two):
Reisinger, D., R. Rudinger, F. Ferraro, C. Harman, K. Rawlins, & B. Van Durme. 2015. Semantic Proto-Roles. Transactions of the Association for Computational Linguistics 3, pp. 475–488.
White, A.S., D. Reisinger, K. Sakaguchi, T. Vieira, S. Zhang, R. Rudinger, K. Rawlins, & B. Van Durme. 2016. Universal Decompositional Semantics on Universal Dependencies. Proceedings of the 2016 Conference on Empirical Methods in Natural Language Processing, pages 1713–1723, Austin, Texas, November 1-5, 2016.
White, A.S., D. Reisinger, R. Rudinger, K. Rawlins, & B. Van Durme. 2016. Computational Linking Theory. arXiv:1610.02544.
White, A. S., K. Rawlins, & B. Van Durme. 2017. The Semantic Proto-Role Linking Model. Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics, pp. 92–98, Valencia, Spain, April 3-7, 2017.
Teichert, A., A. Poliak, B. Van Durme, & M. Gormley. 2017. Semantic Proto-Role Labeling. Proceedings of the Thirty-First AAAI Conference on Artificial Intelligence (AAAI-17).
Rudinger, R., A. Teichert, R. Culkin, S. Zhang, & B. Van Durme. 2018. Neural Davidsonian Semantic Proto-role Labeling. To appear in Proceedings of the 2018 Conference on Empirical Methods in Natural Language Processing, Brussels, Belgium, October 31-November 4, 2018.
