Design and development of domain-specific modeling languages are crucial activities in model-driven engineering. At the core of these languages we find metamodels, i.e. descriptions of concepts and rules to combine those concepts in order to build valid models. Both in research and practice, metamodels are created and updated more or less frequently to meet certain business requirements. Although there exist several workbenches for metamodeling, some textual (e.g., JetBrains MPS) and some graphical (e.g., Eclipse Modeling Framework - EMF), it still remains a sensitive and complex task, where several stakeholders with different skill-sets need to be able to properly exchange ideas and reach agreements.To maximize the throughput of metamodeling activities, in this paper we propose a Blended Metamodeling Framework (BMF) that enables the development of metamodels through both graphical and textual (natural language) notations interchangeably, by utilizing the concepts of Natural Language Processing and model-driven engineering. The feasibility of the framework is demonstrated via the Portable test and Stimulus Standard (PSS) use case, where a DSML is developed by seamlessly blending the use of textual (natural language) and graphical (EMF) notations. Moreover, for demonstration purposes we also generate a domain-specific language structure reflecting the metamodel in JetBrains MPS.