My undergraduate honour’s dissertation was a Natural Language Processing (NLP) research project. It focused on multilingual text generation in under-represented languages. Because existing metrics performed very poorly on evaluating outputs of models trained on the dataset I used to be using, I needed to coach a learned regression metric.
Regression could be useful for a lot of textual tasks, reminiscent of:
- Sentiment evaluation: Predict the strength of positive or negative sentiment as a substitute of straightforward binary classification.
- Writing quality estimation: Predict how high the standard of an article is.
For my use case, I needed the model to attain how good one other model’s prediction was for a given task. My dataset’s rows consisted of the textual input and a label, 0 (bad prediction) or 1 (good prediction).
- Input: Text
- Label: 0 or 1
- The duty: Predict a numerical probability between 0 and 1
But transformer-based models are frequently used for generation tasks. Why would you employ a pre-trained LM for…
