One of the steps involved in the process of sign language generation is generating a sequence of poses that represent the signs. This paper presents a method for using textual information to improve the translation of signs in HamNoSys format into sequences of poses. The method comprises a description generator that translates HamNoSys into a textual description, an LLM fine-tuned to the task of predicting a pose sequence from a HamNoSys description, and a VQ-VAE network that encodes and decodes pose sequences as a list of discrete symbols. Our experiments found that even using simple dictionary descriptions of HamNoSys, it is possible to improve the predictions of pose sequences by leveraging the information from a pretrained LLM.