Interpretability of trajectory-alignment paths

Determine whether alignment paths between machine and human coherence or diversity trajectories can explain particular textual properties, including repetition and topic shifts.

Background

The temporal facet uses dynamic time warping to compare the ordered evolution of coherence and diversity measurements in machine continuations with human reference trajectories. Although the alignment path identifies where measured trajectories differ, the paper does not establish that these numerical differences correspond to recognizable textual phenomena.

Resolving this problem would connect trajectory-level metric behavior to interpretable textual properties such as repetition or topic changes, thereby clarifying whether temporal alignment provides explanatory value beyond correlation with human quality ratings.

References

An alignment path provides a way to inspect where measured trajectories differ. Its usefulness for explaining particular textual properties, such as repetition or topic shifts, remains an empirical question.

— Reference-Based Analysis of Coherence and Diversity in Open-Ended Text Generation  (2609.28080 - Arias, 23 Sep 2026) in Section 3, subsection “Temporal Alignment”

The interval permits meaningful differences in either direction, leaving their relative performance uncertain.

— Reference-Based Analysis of Coherence and Diversity in Open-Ended Text Generation  (2609.28080 - Arias, 23 Sep 2026) in Section 5, subsection “RQ1: Temporal and Summary Associations”

The overall-quality judgments leave open which textual properties drive the associations.

— Reference-Based Analysis of Coherence and Diversity in Open-Ended Text Generation  (2609.28080 - Arias, 23 Sep 2026) in Section 6, “Discussion and Practical Use”

The educators penalized a property of the sequence as a whole, and a metric that reaches this property by adding a trajectory to a per-activity score has not yet been shown to exist.

— The learner who does not learn: when optimizing a pedagogical metric degrades LLM tutoring  (2610.12125 - Figaredo et al., 8 Oct 2026) in Section 5.1, Discussion