Reading Between the Lines: Overcoming Data Sparsity for Accurate Classification of Lexical Relationships