Synthetic child adult talk does not match natural speech patterns
Synthetic Child-Adult Interactions Are Not Developmentally Plausible
Computation and Language
Summary
Many researchers use computer-generated conversations between children and adults to train language models. This paper shows that these synthetic dialogues sound very different from real conversations, especially because they miss the messy and broken flow of how people actually talk. The authors suggest a new way to improve synthetic dialogue by training models with real conversations and show a small model can become more realistic with this method.
What this means in practice
- •For language technology developers: Improve child-adult dialogue modeling by fine-tuning models on real conversation data to better mimic natural speech patterns.
- •For dialogue system engineers: Create more realistic synthetic dialogues for training conversational agents by incorporating the fragmented style of spoken interaction.
Authors
Bastian Bunzeck, Francesca Padovani
Abstract
Synthetic data has become a common resource for training language models, and this trend has spilled over into research on developmentally plausible language modeling. However, it remains unclear whether LM-generated child-adult interactions actually resemble natural ones. We compare the synthetic TinyDialogues dataset to natural dialogue across a broad range of linguistic measures. While synthetic data matches natural data on the lexical level, it drastically underestimates the messy, fragmented nature of spoken interaction. To mitigate this, we propose instruction-tuning with data created from real dialogues, and show in a pilot study that a small model approximates the statistical tendencies of natural data after moderate parameter-efficient fine-tuning.