README.md

---
base_model:
- google-bert/bert-base-uncased
language:
- en
tags:
- semantic-role-labeling
- srl
---

# srl_bert_model

This repository contains a BERT-based model for **Semantic Role Labeling (SRL)**.

## Model Description

We revisit structured SRL modeling with a modernized encoder-based framework that preserves explicit predicate-argument structure while enabling 10× faster inference than AllenNLP.

- **Model name:** `srl_bert_model`
- **Repository:** `yeomtong/srl_bert_model`
- **Architecture:** BERT-based SRL model
- **Framework:** PyTorch
- **Task:** Semantic Role Labeling

## Intended Use

This model is intended for:

- semantic role labeling research
- predicate-argument structure analysis
- downstream NLP tasks that require structured semantic information
- experimentation with role-aware language representations


## How to Use

Example loading code should be adapted to your project setup.

```python
from huggingface_hub import hf_hub_download, snapshot_download

ckpt_path = hf_hub_download(
    repo_id="yeomtong/srl_bert_model",
    filename="best_srl_Sep_29.ckpt")
repo_dir = snapshot_download("yeomtong/srl_bert_model")
sys.path.append(repo_dir)

from predictor import srl_init
from model import PredicateAwareSRL
from visualizer import prediction, prediction_formatted

#load model
srl_init(ckpt_path, bert_name= "bert-base-cased")

test_sentence = "I want to go home"
prediction(test_sentence)
'''
Sentence: I want to go home

————————————————————————————————————————————————————————————
[ARG0: I] [V: want] [ARG1: to go home]

TOKEN: I      want to     go     home  
LABEL: B-ARG0 B-V  B-ARG1 I-ARG1 I-ARG1

————————————————————————————————————————————————————————————
[ARG0: I] want to [V: go] [ARG4: home]

TOKEN: I      want to go  home  
LABEL: B-ARG0 .    .  B-V B-ARG4
'''

prediction_formatted(test_sentence)

'''
{'verbs': [{'verb': 'want',
   'description': '[ARG0: I] [V: want] [ARG1: to go home]',
   'tags': ['B-ARG0', 'B-V', 'B-ARG1', 'I-ARG1', 'I-ARG1']},
  {'verb': 'go',
   'description': '[ARG0: I] want to [V: go] [ARG4: home]',
   'tags': ['B-ARG0', 'O', 'O', 'B-V', 'B-ARG4']}],
 'words': ['I', 'want', 'to', 'go', 'home']}
'''