Software requirement specification (SRS) document contains faults due to the inherent ambiguous nature of natural language (NL). These faults are identified and reported (using fault logs) through inspections and are handed back to the requirements author for fixations. This process is very manual, time consuming and a lot of efforts is spent on re-inspection of the SRS document while fault fixations. An automated approach is needed that can map fault logs to faulty requirements and to other similar requirements. The automated approach could enable large fault coverage and can reduce significant manual re-inspection time and efforts. Our proposed approach extracts the key-phrases to identify key problems from fault-logs, and then maps them back to group of similar requirements in an SRS document to inspect requirements that may contain a similar types of faults. Our approach uses key-phrase extraction algorithms, semantic analysis models and clustering approaches to map faults to requirements. We evaluated the mapping of faults to requirements in our approach using two widely used semantic analysis models (i.e., Latent Semantic Analysis and Latent Dirichlet Allocation) with the evaluation performed by the domain expert. Our results have been promising and have showed a large potential to support additional decision making during fault fixations.