Deep Neural Nets and the Language Instinct

Abstract

In this masters thesis we will conduct a series of translation experiments with sentences that deviate from the principles of universal grammar on human participants and the machine learning models NLLB200 and ChatGPT. Trough this research, we find that while human participants perform well on these sentences, machine learning models have a tougher time translating these sentences. With these finding, we find some evidence that support our hypothesis.