(a): A sub-graph of BBN formed by Algorithm 2 for GloVe Pennington et al. (2014) trained on 2017-January dump of Wikipedia; we discuss the structure of the graph with respect to the word “nurse”. We illustrate four possible scenarios with respect to their effect on GIPE, with θ_s = 0.05: (b) An edge with β ¡ θ_s may not contribute to γ_i or $ηwi$⁠; (c) An outgoing edge may contribute to $ηwi$ only; (d) An incoming edge may contribute to γ_i only; (e) Incoming and outgoing edges may contribute to γ_i and $ηwi$ respectively. Every node pair association can be categorized as one of the aforementioned four cases.