Learning Natural Language Generation from Scratch
This approach addresses the dependency on labeled datasets and reduces biases in language generation for tasks like visual question generation, though it is incremental in applying RL to language models.
The paper tackles the problem of training conditional language models without labeled data by introducing TrufLL, which uses reinforcement learning with dynamic vocabulary truncation, and reports positive results on visual question generation tasks, including performance and language metrics validated by human evaluation.
This paper introduces TRUncated ReinForcement Learning for Language (TrufLL), an original ap-proach to train conditional language models from scratch by only using reinforcement learning (RL). AsRL methods unsuccessfully scale to large action spaces, we dynamically truncate the vocabulary spaceusing a generic language model. TrufLL thus enables to train a language agent by solely interacting withits environment without any task-specific prior knowledge; it is only guided with a task-agnostic languagemodel. Interestingly, this approach avoids the dependency to labelled datasets and inherently reduces pre-trained policy flaws such as language or exposure biases. We evaluate TrufLL on two visual questiongeneration tasks, for which we report positive results over performance and language metrics, which wethen corroborate with a human evaluation. To our knowledge, it is the first approach that successfullylearns a language generation policy (almost) from scratch.