PaperlayerView exact revision on arXiv

Source-linked AI summary

Replicating a Disjoint-Set Union Experiment over Various Notions of Micro Units to assess Translation Effort

Michael Carl

arXiv:2609.07748v1cs.CL

TL;DR

The paper examines how pause-based segmentation affects the distribution of editing strategies across micro- and macro-units. It replicates the analysis using five segmentation methods and finds that TT-based segmentation differs from similar time-based methods, motivating its conclusion that product-based units offer a more linguistically grounded framework.

  • Problem

    The effect of pause-threshold segmentation on editing-strategy distributions is unclear, although the choice of threshold changes how keystrokes are grouped into micro- and macro-units.

    Source: A replication experiment · p. 2

  • Method

    The replication applies five segmentation methods to automatically mapped target-word IDs in CRITT TPR-DB sessions and extracts DSUs automatically.

    Source: A replication experiment · p. 2

  • Results

    TT-based segmentation assigns roughly 85% of strategies to P0 and less than 10% to P1, unlike the approximately similar distributions from time-based methods.

    Source: Evaluation · pp. 3–4

  • Takeaways & Limitations

    The authors conclude that product-based MUs anchor revision strategies to stable textual units in a linguistically grounded, analytically transparent framework.

    Source: Conclusion · p. 6

    Temporal segmentation may obscure linguistically meaningful revision behavior, reducing precision when the exact scope of a revision target matters.

    Source: Conclusion · p. 5

Abstract

from arXiv · show

The paper describes a replication experiment to assess "the distribution of editing procedures across micro and macro units as an indicator of the strain of text production". We investigate various pause thresholds to isolate Micro and Macro units and conclude that process-based units might be more suitable entity.

A replication experiment

The replication compares five ways of segmenting micro-units (MUs) to examine how segmentation affects editing strategies and derived macro-units (MXs). It uses automatically mapped target-word IDs and automated DSU extraction on 478 translation sessions.

A replication experiment

Evaluation

The four time-based methods produce similar editing-strategy distributions, while TT-based segmentation differs substantially. Results also vary by language pair, and TT-based MXs contain fewer MUs than time-based MXs.

Evaluation

  • Time-based methods yield approximately similar editing-strategy distributions, unlike TT-based segmentation, which concentrates strategies differently.Source: Evaluation · pp. 3–4
  • Figure 2 shows around 6 MUs per MX for EN-AR and ZH-EN, 3 to 4 for EN-HI and EN-DE, and fewer for other language pairs.Source: Figure 2 · p. 4, Evaluation · p. 4
  • 82%-87% of strategies are P0 or P1 combined; TT-based segmentation assigns roughly 85% to P0 and less than 10% to P1.Source: Evaluation · p. 4
  • Table 3 reports 3 TTids and 3.77 MUs per MUD-based MX, compared with exactly 1 TTid and 1.17 MUs per TT-based MX.Source: Table 3 · p. 4, Evaluation · p. 5

Conclusion

Temporal segmentation methods resemble one another but differ systematically from text-based TT, and pause-based units may misalign with linguistically meaningful revision questions. TT word-based units offer a more transparent framework, while CRITT features make DSUs unnecessary for investigating revision dynamics.

Conclusion

  • Temporal methods show similar MX sizes and P0/P1 distributions, yet differ systematically from text-based TT segmentation.Source: Conclusion · p. 5
  • Temporal segmentation risks obscuring meaningful revision behavior by dividing production at pauses rather than textual boundaries.Source: Conclusion · p. 5
  • Larger MXs under static K1000 and translator-specific MUD reduce analytical precision when the exact scope of a revision target matters.Source: Conclusion · p. 5
  • For cohesive chains, temporal MUs may be misaligned; pauses within a word may inflate P1 counts without clear interpretive value.Source: Conclusion · p. 5
  • TT segmentation maps each MX to one target-text word, anchoring revision procedures to stable textual units for more transparent analysis.Source: Conclusion · p. 6
  • CRITT TPR-DB features support methodologically consistent, theoretically aligned study of revision dynamics without computing DSUs.Source: Conclusion · p. 6
Loading 2609.07748v1…