Protein function prediction from sequence, structure, gene expression profiles, and published literature are needed to understand all biological processes. Natural language processing of biological text and large language model (LLM)-based encoding of sequence and structure opens powerful paths to rapid function annotation and novel training models. In this survey, we take a look at the available models for function prediction, especially the NLP- and LLM-based models. The survey highlights the major advances made and the ground that still needs to be covered to automate the process of function prediction from two major sources namely protein sequences and published research documents.