1. Definition and application of Controlable Text Generation (CTG)
1. What is CTG?
Controllable Text Generation, controllable text generation, is to be able to increase the control of some attributes, styles, key information, etc. of the generated text on the basis of traditional text generation, so that the generated text meets our certain expectations. If a clear definition is given, the definition of another earlier review is quoted in the article:
A Survey of Controllable Text Generation using Transformer-based Pre-trained Language Models[1]
Two simple examples are given in the paper:
The first example is given a story line and requires the model to be generated according to this idea. The second example is about the problems that may arise if the machine generates text without control in traditional human-computer dialogue, such as giving harmful suggestions or even swearing. At this time, we need to control it when the model is generated.
The author summed up such a general framework of Input-Process-Output for CTG, referred to as IPO:
The difference from traditional NLG is that we will add certain control factors to the input, so that the final output can meet certain conditions.
2. Application scenarios of CTG
CTG can be applied to a wide range of scenarios:
- At