IteraTeR-PEGASUS-Revision-Generator

IteraTeR PEGASUS model

This model was obtained by fine-tuning google/pegasus-large on IteraTeR-full-sent dataset.

Paper: Understanding Iterative Revision from Human-Written Text <br> Authors: Wanyu Du, Vipul Raheja, Dhruv Kumar, Zae Myung Kim, Melissa Lopez, Dongyeop Kang

Text Revision Task

Given an edit intention and an original sentence, our model can generate a revised sentence.<br> The edit intentions are provided by IteraTeR-full-sent dataset, which are categorized as follows:

<table> <tr> <th>Edit Intention</th> <th>Definition</th> <th>Example</th> </tr> <tr> <td>clarity</td> <td>Make the text more formal, concise, readable and understandable.</td> <td> Original: It's like a house which anyone can enter in it. <br> Revised: It's like a house which anyone can enter. </td> </tr> <tr> <td>fluency</td> <td>Fix grammatical errors in the text.</td> <td> Original: In the same year he became the Fellow of the Royal Society. <br> Revised: In the same year, he became the Fellow of the Royal Society. </td> </tr> <tr> <td>coherence</td> <td>Make the text more cohesive, logically linked and consistent as a whole.</td> <td> Original: Achievements and awards Among his other activities, he founded the Karachi Film Guild and Pakistan Film and TV Academy. <br> Revised: Among his other activities, he founded the Karachi Film Guild and Pakistan Film and TV Academy. </td> </tr> <tr> <td>style</td> <td>Convey the writer’s writing preferences, including emotions, tone, voice, etc..</td> <td> Original: She was last seen on 2005-10-22. <br> Revised: She was last seen on October 22, 2005. </td> </tr> </table>

Usage

from transformers import AutoTokenizer, AutoModelForSeq2SeqLM

tokenizer = AutoTokenizer.from_pretrained("wanyu/IteraTeR-PEGASUS-Revision-Generator")
model = AutoModelForSeq2SeqLM.from_pretrained("wanyu/IteraTeR-PEGASUS-Revision-Generator")
before_input = '<fluency> I likes coffee.'
model_input = tokenizer(before_input, return_tensors='pt')
model_outputs = model.generate(**model_input, num_beams=8, max_length=1024)
after_text = tokenizer.batch_decode(model_outputs, skip_special_tokens=True)[0]