Losing My Composure: Predicting Compositionality Over Time

2026-07-13

AI Summary

This study investigates the diachronic evolution of compositionality in German and English noun compounds, critically examining the long-standing hypothesis that compositionality declines over time. To this end, we construct the first decade-stamped, human-rated dataset of compositionality scores and introduce a novel task: compositionality trend prediction. Leveraging corpus slices spanning one to five decades, we systematically compare static and contextualized semantic vector models under three training regimes—single-decade, incrementally expanding windows, and half-century aggregate training. Our empirical results reveal only a weak downward trend in compositionality, failing to support the strong decline hypothesis. Moreover, models trained on narrow temporal windows align more closely with human judgments, and static representations perform on par with contextualized models in capturing diachronic compositionality patterns.

Abstract

We explore the phenomenon of semantic change of German and English noun compounds, with the objective of investigating and modeling gradual changes of meanings and degrees of compositionality in the past and over time. To do so, we introduce the Compositionality Trend Prediction task, which is evaluated against a novel dataset of in-context compositionality ratings sampled across several decades of diachronic corpora for 23 German and 26 English target compounds, uniquely providing per-decade ratings and corresponding trends over time. These per-decade compositionality ratings allow us to investigate empirically untested hypotheses of generalized trends in compositionality over time, such as the idea that compounds should become less compositional (less transparent) over time. Beyond our empirical observations from the diachronic compositionality annotations, we perform experiments with semantic vector representations of varying complexity, as well as several temporal granularities for training these representations on diachronic data, resulting in about 100 models of each representation type, each covering a different 1--5 decade slice of a diachronic corpus. Contrary to the decisive tendency posited in the literature, we find only a small negative trend in compositionality over time in our target compounds. In our computational experiments, we find that using models trained on narrow time slices of diachronic data (single decades, or incrementally expanding temporal windows) align better with the per-decade compositionality ratings than those trained on an entire half-century window, the latter setting being an analog for the prevalent modeling approach of training representations on an entire half of a corpus' data. Additionally, we find static representations to be competitive with contextual representations in the Compositionality Trend Prediction task.

Problem

• compositionality
• semantic change
• noun compounds
• diachronic corpora
• temporal trends

Innovation

• compositionality trend prediction
• diachronic semantic change
• noun compounds
• temporal granularity
• static vs contextual embeddings