Evaluating Morphological Compositional Generalization in Large Language Models
This work addresses the problem of assessing linguistic generalization in LLMs for researchers and developers, revealing limitations in morphological productivity and systematicity, which is incremental as it builds on existing concerns about compositional generalization.
The paper systematically investigates the morphological compositional generalization abilities of large language models (LLMs) in agglutinative languages like Turkish and Finnish, finding that models such as GPT-4 and Gemini struggle with novel word roots and show declining performance as complexity increases, with significant accuracy gaps compared to humans.
Large language models (LLMs) have demonstrated significant progress in various natural language generation and understanding tasks. However, their linguistic generalization capabilities remain questionable, raising doubts about whether these models learn language similarly to humans. While humans exhibit compositional generalization and linguistic creativity in language use, the extent to which LLMs replicate these abilities, particularly in morphology, is under-explored. In this work, we systematically investigate the morphological generalization abilities of LLMs through the lens of compositionality. We define morphemes as compositional primitives and design a novel suite of generative and discriminative tasks to assess morphological productivity and systematicity. Focusing on agglutinative languages such as Turkish and Finnish, we evaluate several state-of-the-art instruction-finetuned multilingual models, including GPT-4 and Gemini. Our analysis shows that LLMs struggle with morphological compositional generalization particularly when applied to novel word roots, with performance declining sharply as morphological complexity increases. While models can identify individual morphological combinations better than chance, their performance lacks systematicity, leading to significant accuracy gaps compared to humans.