A Hierarchical Playscript Representation of Distributed Words for Effective Semantic Clustering and Search

Semantic word embeddings have shown to cluster in space based on linguistic similarities that are quantifiably captured using simple vector algebra. Recently, methods for learning distributed word vectors have progressively empowered neural language models to compute compositional vector representations for phrases of variable length. However, they remain limited in expressing more generic relatedness between instances of a larger and non-uniform sized body-of-text. A recent study proposed a formulation that combines a word vector set of variable cardinality to represent a verse, with an iterative distance metric to evaluate similarity in pairs of non-conforming verse matrices. In this work, we expand on this sentence abstraction and apply it to a dialogue text passage that is prescribed in a playscript and uttered by an actor. In contrast to baselines characterized by a bag of features, our model preserves word order and is more sustainable in performing semantic matching at any of a dialogue, act, and play levels. To validate our framework for training word vectors, we analyzed the clustering of the complete play set of Shakespeare by exploring multidimensional scaling for visualization, and experimented with playscript searches of both contiguous and out-of-order parts of dialogues. We report robust results that support our intuition for measuring play-to-play and dialogue-to-play similarity.