The Hyperspace Analogue to Language (HAL) model encodes contextual experience in a representational form that cuts across traditional semantic, grammatical, syntactic boundaries. Attempts to model subsymbolic representations have been limited for three reasons: contrived representations, narrow focus on one aspect of context (semantics, grammar, etc.), not truly subsymbolic. HAL's vector representations avoid these limitations by capitalizing on global lexical co-occurrence and captures contexts in which words occur and, as such, have explanatory power that captures a broad range of cognitive phenomena. A new set of experiments will be presented that compares the global co-occurrence learning procedure with a recurrent connectionist network. A set of criticisms have been leveled against both classes of models. These include that the models are frequency counters and do not capture real semantic or grammatical structure; that they are a throwback to old-style associationist models and are outdated; and are impressive tools, but hardly representational theories of meaning. These criticisms will be countered with empirical results that show that context is a valid and fundamental carrier of information pertaining to word meaning - both at the semantic and grammatical level.