# tosin /pcl_22

## T5Base-PCL

This is a fine-tuned model of T5 (base) on the patronizing and condenscending language (PCL) dataset by Pérez-Almendros et al (2020) used for Task 4 competition of SemEval-2022. It is intended to be used as a classification model for identifying PCL (0 - neg; 1 - pos). The task prefix we used for the T5 model is 'classification: '.

The dataset it's trained on is limited in scope, as it covers only some news texts covering about 20 English-speaking countries. The macro F1 score achieved on the test set, based on the official evaluation, is 0.5452. More information about the original pre-trained model can be found here

• Classification examples:
Prediction Input
0 selective kindness : in europe , some refugees are more equal than others
1 he said their efforts should not stop only at creating many graduates but also extended to students from poor families so that they could break away from the cycle of poverty

### How to use

from transformers import T5ForConditionalGeneration, T5Tokenizer
import torch
model = T5ForConditionalGeneration.from_pretrained("tosin/pcl_22")
tokenizer = T5Tokenizer.from_pretrained("t5-base") # use the source tokenizer because T5 finetuned tokenizer breaks