AETAS: Analysis of Evolving Temporal Affect and Semantics for Legal History

TL;DR

Addressing semantic drift in historical legal discourse and the need for interpretable, reproducible tooling, the paper introduces AETAS, an auditable pipeline that bins data by decade, trains SGNS embeddings, aligns spaces with orthogonal Procrustes, and quantifies drift via $d = 1 - \cos(v_t, v_{t'})$, while grounding results on a mercy-retribution axis. The authors contribute a fully scripted workflow, narrative-ready visualizations, and a suite of robustness checks to support interpretation beyond black-box metrics. Empirical results show substantive drift in justice, transportation, charity, insanity, and poverty across 1720–1913, with notable but sometimes modest net effects after accounting for within-bin instability. The framework serves as a reusable template for diachronic semantic analysis in other historical corpora and for linking lexical change to socio-legal history.

Abstract

Digital-humanities work on semantic shift often alternates between handcrafted close readings and opaque embedding machinery. We present a reproducible expert-system style pipeline that quantifies and visualises lexical drift in the Old Bailey Corpus (1720--1913), coupling interpretable trajectories with legally meaningful axes. We bin proceedings by decade with dynamic merging for low-resource slices, train skip-gram embeddings, align spaces through orthogonal Procrustes, and measure both geometric displacement and neighborhood turnover. Three visual analytics outputs, which are drift magnitudes, semantic trajectories, and movement along a mercy-versus-retribution axis, expose how justice, crime, poverty, and insanity evolve with penal reforms, transportation debates, and Victorian moral politics. The pipeline is implemented as auditable scripts so results can be reproduced in other historical corpora.

Figures (7)

• Figure 1: Semantic drift magnitudes by span and domain. Points show cosine drift; stems indicate displacement relative to each time window.
• Figure 2: Aligned semantic trajectories for focal terms. Arrows indicate chronological order toward 1900s.
• Figure 3: Shifts along a mercy-retribution axis. Positive scores indicate proximity to mercy-related terms; negatives lean toward punitive semantics.
• Figure 4: Neighbor stability when increasing $k$ from 5 to 20 (Jaccard of start/end neighborhoods). Larger $k$ raises overlap but many terms still shift into new contextual neighborhoods.
• Figure 5: Mercy-retribution projections under leave-one-out seed variants. Bands span min/max per year; lines show mean. Justice remains negative in 1900s across seed choices.