Create small controlled presentation mini-sets as a near-term community benchmark¶

Strategy · All strategies
Evidence · 1 claim (1 for) · 1 study (1 review), q2 · 0 of 1 report an effect size · 1 claim rests on one study

Description¶

To address the corpus gap, the survey recommends "creating small, controlledpresentation mini-sets, consisting of 5–10 minute talks per speaker with slide markers, emphasis annotations, and diverse accent coverage, as a near-term community benchmark." Section 7 elaborates the target composition: 20–30 speakers from 5+ L1 backgrounds delivering technical talks with timestamped slide boundaries, phone-level transcriptions, and expert prosody ratings, enabling head-to-head system comparison and shared evaluation campaigns.

Design Implications¶

Context¶

Requirements¶

• Slide markers, emphasis annotations, phone-level transcriptions, expert prosody ratings, and diverse L1 coverage across speakers.

Constraints¶

• The survey notes the datasets it curates and recommends are limited in scale; large-scale presentation corpora with diverse accents and domains would strengthen evaluation.

Target Learners¶

• L2 English presenters across diverse L1 backgrounds

Target Learning Goals¶

• Enabling evaluation of coaching systems on all five presentation-skill dimensions

Claims¶

• No existing corpus simultaneously provides L2 speech, long-form presentation structure, accent labels, and prosodic annotations [+W]

Related Strategies¶

• Calibrate coaching thresholds per L1 cohort and avoid penalizing identity-marking accent features
• TTS-based coaching pipeline: anchor and target exemplars, chunked recording, alignment, and focused drills
• Build new frontier-model benchmarks from new hard tasks with expert verification and adequate curation resources

Examples¶

-

Key Sources¶

• Wen Liang, Li Siyan, Zackary Rackauckas, and Julia Hirschberg. (2026). A Survey of Automated Presentation Coaching: Systems, Methods, and Open Challenges. https://arxiv.org/abs/2606.27380