Advancing Automated Spatio-Semantic Analysis in Picture Description Using Language Models