Square One Bias in NLP: Towards a Multi-Dimensional Exploration of the Research Manifold