Researchers investigate semantic change in 23 German and 26 English noun compounds by introducing the Compositionality Trend Prediction task, evaluated against a novel dataset of per-decade diachronic ratings. Contrary to literature positing a decisive tendency for compounds to become less compositional, the study finds only a small negative trend over time.
- The dataset provides unique per-decade compositionality ratings and corresponding trends for 23 German and 26 English target compounds.
- Experiments involved approximately 100 models of varying semantic vector representations trained on different 1-5 decade slices of diachronic corpora.
- Models trained on narrow time slices, such as single decades or incrementally expanding windows, aligned better with ratings than those trained on entire half-century windows.
- Static representations were found to be competitive with contextual representations in the Compositionality Trend Prediction task.
The findings challenge the hypothesis that compounds generally become less transparent over time and suggest that modeling temporal granularity is crucial for accurate compositionality prediction.