No abstract

How much does money drive legislative outcomes in the United States? In this article, we use aggregated campaign finance data as well as a Transformer based text embedding model to predict roll call votes for legislation in the US Congress with more than 90% accuracy. In a series of model comparisons in which the input feature sets are varied, we investigate the extent to which campaign finance is predictive of voting behavior in comparison with variables like partisan affiliation. We find that the financial interests backing a legislator’s campaigns are independently predictive in both chambers of Congress, but also uncover a sizable asymmetry between the Senate and the House of Representatives. These findings are cross-referenced with a Representational Similarity Analysis (RSA) linking legislators’ financial and voting records, in which we show that “legislators who vote together get paid together”, again discovering an asymmetry between the House and the Senate in the additional predictive power of campaign finance once party is accounted for. We suggest an explanation of these facts in terms of Thomas Ferguson’s Investment Theory of Party Competition: due to a number of structural differences between the House and Senate, but chiefly the lower amortized cost of obtaining individuated influence with Senators, political investors prefer operating on the House using the party as a proxy.

Neural models of Knowledge Base data typically employ compositional representations of graph objects: entity and relation embeddings are systematically combined to evaluate the truth of a candidate knowledge base entry. Using a model inspired by Harmonic Grammar, we propose to tokenize triplet embeddings by subjecting them to a process of optimization with respect to learned well-formedness conditions on knowledge base entries. The resulting model, known as Gradient Graphs, leads to sizable improvements when implemented as a companion to compositional models. The "supracompositional" triplet token embeddings it produces have interpretable properties that prove helpful in performing inference on the resulting representations.

We present Harmonic Memory Networks (HMem), a neural architecture for knowledge base completion that models entities as weighted sums of pairwise bindings between an entity's neighbors and corresponding relations. Since entities are modeled as aggregated neighborhoods, representations of unseen entities can be generated on the fly. We demonstrate this with two new datasets: WNGen and FBGen. Experiments show that the model is SOTA on benchmarks, and flexible enough to evolve without retraining as the knowledge graph grows.1 Code and datasets are available at github.com/ MatthiasRLalisse/HMemNetworks.

A framework and method are proposed for the study of constituent composition in fMRI. The method produces estimates of neural patterns encoding complex linguistic structures, under the assumption that the contributions of individual constituents are additive. Like usual techniques for modeling compositional structure in fMRI, the proposed method employs pattern superposition to synthesize complex structures from their parts. Unlike these techniques, superpositions are sensitive to the structural positions of constituents, making them irreducible to structure-indiscriminate ("bag-of-words") models of composition. Using data from a study by Frankland and Greene (2015), it is shown that comparison of neural predictive models with differing specifications can illuminate aspects of neural representational contents that are not apparent when composition is not modelled. The results indicate that the neural instantiations of the binding of fillers to thematic roles in a sentence are non-orthogonal, and therefore spatially overlapping.