Short Answer

Analyzing Generator Output in Text Corruption

A text-corrupting model is given the masked sentence: 'The chef prepared the [MASK] dish with [MASK] ingredients.' The original, unmasked sentence was: 'The chef prepared the signature dish with fresh ingredients.' The model outputs the following new sentence: 'The chef prepared the delicious dish with local ingredients.' Based on the function of this type of model, explain why the output tokens ('delicious', 'local') can be different from the original tokens ('signature', 'fresh').

0

1

Updated 2025-10-04

Contributors are:

Who are from:

Tags

Ch.1 Pre-training - Foundations of Large Language Models

Foundations of Large Language Models

Foundations of Large Language Models Course

Computing Sciences

Analysis in Bloom's Taxonomy

Cognitive Psychology

Psychology

Social Science

Empirical Science

Science