Do Vision and Language Models Share Concepts? A Vector Space Alignment Study
This addresses the debate on whether language models can connect to the world, with implications for multi-modal processing and understanding in AI.
The study investigated whether vision and language models share conceptual representations by evaluating four language model families and three vision model architectures, finding that language models partially converge towards representations isomorphic to those of vision models, influenced by factors like dispersion, polysemy, and frequency.
Large-scale pretrained language models (LMs) are said to ``lack the ability to connect utterances to the world'' (Bender and Koller, 2020), because they do not have ``mental models of the world' '(Mitchell and Krakauer, 2023). If so, one would expect LM representations to be unrelated to representations induced by vision models. We present an empirical evaluation across four families of LMs (BERT, GPT-2, OPT and LLaMA-2) and three vision model architectures (ResNet, SegFormer, and MAE). Our experiments show that LMs partially converge towards representations isomorphic to those of vision models, subject to dispersion, polysemy and frequency. This has important implications for both multi-modal processing and the LM understanding debate (Mitchell and Krakauer, 2023).