Gov2Vec: Learning Distributed Representations of Institutions and Their Legal Text

摘要

We compare policy differences across institutions by embeddingrepresentations of the entire legal corpus of each institution and thevocabulary shared across all corpora into a continuous vector space. We applyour method, Gov2Vec, to Supreme Court opinions, Presidential actions, andofficial summaries of Congressional bills. The model discerns meaningfuldifferences between government branches. We also learn representations for morefine-grained word sources: individual Presidents and (2-year) Congresses. Thesimilarities between learned representations of Congresses over time andsitting Presidents are negatively correlated with the bill veto rate, and thetemporal ordering of Presidents and Congresses was implicitly learned from onlytext. With the resulting vectors we answer questions such as: how does Obamaand the 113th House differ in addressing climate change and how does this varyfrom environmental or economic perspectives? Our work illustratesvector-arithmetic-based investigations of complex relationships between wordsources based on their texts. We are extending this to create a morecomprehensive legal semantic map.