Deep lexical resources include lexicons for linguistically-precise grammars, template sets for information extraction systems, and ontologies for word sense disambiguation. Such resources are critical for enhancing the performance of systems and for improving their portability between domains. Most deep lexical resources in current use have been developed manually by lexicographers at considerable cost, and yet have limited coverage and require labour-intensive porting to new tasks. Automatic lexical acquisition is a more promising and cost-effective approach to take, and is increasingly viable given recent advances in NLP and machine learning technology, and corpus availability. However, a number of important challenges still need addressing before benefits can be reaped in practical language engineering, such as the (multilingual) acquisition of deep lexical information from corpora and the implementation of accurate, large-scale, portable acquisition techniques.