Investigating Training Strategies and Model Robustness of Low-Rank Adaptation for Language Modeling in Speech Recognition