Naming latencies as evidence for two modes of lexical retrieval