Transfer of Structural Knowledge from Synthetic Languages
This work addresses the challenge of efficient natural language understanding for less powerful models, though it appears incremental as it builds on prior synthetic language research.
The paper tackles the problem of improving transfer learning to English by fine-tuning models on synthetic languages, introducing a new synthetic language that enhances transfer and a new benchmark, Tiny-Cloze, which shows better performance on various tasks.
This work explores transfer learning from several synthetic languages to English. We investigate the structure of the embeddings in the fine-tuned models, the information they contain, and the capabilities of the fine-tuned models on simple linguistic tasks. We also introduce a new synthetic language that leads to better transfer to English than the languages used in previous research. Finally, we introduce Tiny-Cloze Benchmark - a new synthetic benchmark for natural language understanding that is more informative for less powerful models. We use Tiny-Cloze Benchmark to evaluate fine-tuned models in several domains demonstrating that fine-tuning on a new synthetic language allows for better performance on a variety of tasks.