Unlike native speech perception, which is robust, automatic and efficient even in sub-optimal conditions, second-language (L2) speech perception is notoriously problematic even for highly proficient and experienced L2 listeners. L2 sounds that are phonologically contrastive in the L2 but not in the L1 are often miscategorized and misconstrued by L2 listeners, which renders spoken word comprehension difficult. In this talk, I discuss acquisition and processing of consonantal hardness/softness in L2 Russian by L1 English speakers. I review evidence from different levels of linguistic analysis (phonetic, word level, and sentence level) to examine how low-level (bottom-up) information derived from the phonetic-acoustic input is combined with higher-order (top-down) information derived from sentential context (semantic, syntactic, and morphological). Experimental evidence suggests that imperfect encoding of an L2 phonological feature leads to phonologically and lexically ambiguous representations (e.g., the words ‘rock’ and ‘lock’ can be confused if phonological representations for /r/ and /l/ are not sufficiently robust) but that L2 listeners utilize contextual information for meaning disambiguation. Notably, morphological and syntactic constraints appear to be more effective in constraining the choice between two phonolexically ambiguous words than semantic constraints. The implications of the findings are discussed.