We're Calling an Intervention: Exploring the Fundamental Hurdles in Adapting Language Models to Nonstandard Text (2404.07304v2)

Abstract: We present a suite of experiments that allow us to understand the underlying challenges of LLM adaptation to nonstandard text. We do so by designing interventions that approximate several types of linguistic variation and their interactions with existing biases of LLMs. Applying our interventions during LLM adaptation with varying size and nature of training data, we gain important insights into when knowledge transfer can be successful, as well as the aspects of linguistic variation that are particularly difficult for LLMs to deal with. For instance, on text with character-level variation, performance improves with even a few training examples but approaches a plateau, suggesting that more data is not the solution. In contrast, on text with variation involving new words or meanings, far more data is needed, but it leads to a massive breakthrough in performance. Our findings reveal that existing models lack the necessary infrastructure to handle diverse forms of nonstandard text and linguistic variation, guiding the development of more resilient LLMing techniques for the future. We make the code for our interventions, which can be applied to any English text data, publicly available.