Event processing follows a decoupled model of interaction in space, time, and synchronization. However, another dimension of semantic coupling also exists and poses a challenge to the scalability of event processing systems in highly semantically heterogeneous and dynamic environments such as the Internet of Things (IoT). Current state-of-the-art approaches of content-based and concept-based event systems require a significant agreement between event producers and consumers on event schema or an external conceptual model of event semantics. Thus, they do not address the semantic coupling issue. This article proposes an approach where participants only agree on a distributional statistical model of semantics represented in a corpus of text to derive semantic similarity and relatedness. It also proposes an approximate model for relaxing the semantic coupling dimension via an approximation-enabled rule language and an approximate event matcher. The model is formalized as an ensemble of semantic and top-k matchers along with a probability model for uncertainty management. The model has been empirically validated on large sets of events and subscriptions synthesized from real-world smart city and energy management systems. Experiments show that the proposed model achieves more than 95% F1Score of effectiveness and thousands of events/sec of throughput formedium degrees of approximation while not requiring users to have complete prior knowledge of event semantics. In semantically loosely-coupled environments, one approximate subscription can compensate for hundreds of exact subscriptions to cover all possibilities in environments which require complete prior knowledge of event semantics. Results indicate that approximate semantic event processing could play a promising role in the IoT middleware layer.