Pen input supported deictic referencing: 85% of V1 queries contained at least one visible pen mark, and ease of referencing improved significantly over the baseline¶

Claim · All claims
Evidence · 2 studies · 1 causal r1, 1 design r3 · q3
Currency · no dated study recorded · not yet checked against its sources · not reviewed by a person

Subclaims¶

q3 i? In the V1 dataset, underlining (45%), circles (29%), boxes (9%) and braces (2%) dominated, and 85% of queries contained at least one visible mark. → penmark q3 i? Ease of referencing specific elements was significantly higher for V1 than baseline (6.00 vs. 4.38, p=0.015). → ease-ref

Evidence¶

penmark¶

Jeongmin Rhee, Changhee Lee, Hyunwoo Kim, Kiroong Choe, Bohyoung Kim, Sungahn Ko, and Jinwook Seo. 2026. Penquiry: A Pen-based Interactive In-situ Q&A System Leveraging LLMs. https://arxiv.org/abs/2609.19870 doi:10.48550/arxiv.2609.19870

q3 · i? · design · r3

Analysis of the 204 queries in the Penquiry V1 dataset from Study 1, classified using Marshall's annotation taxonomy; the authors state this "overwhelming reliance on visual deictic cues demonstrates that pen input effectively supports precise referencing behaviors."

"Participants predominantly used four pen-mark types: underlining (45%), circles (29%), boxes (9%), and braces (2%) (Appendix A.3). Notably, 85% of queries contained at least one visible mark."

ease-ref¶

Jeongmin Rhee, Changhee Lee, Hyunwoo Kim, Kiroong Choe, Bohyoung Kim, Sungahn Ko, and Jinwook Seo. 2026. Penquiry: A Pen-based Interactive In-situ Q&A System Leveraging LLMs. https://arxiv.org/abs/2609.19870

q3 · i? · causal · r1

Study 1 questionnaire item (N=16, within-subjects) on ease of referencing specific elements: V1 scored 6.00 versus baseline 4.38, p=0.015. Interview data (e.g., P16) indicated pointing avoided lengthy textual descriptions of graph regions.

"Regarding the “Ease of referencing specific elements, ”PenquiryV1 showed a statistically significant improvement over the baseline (6.00 vs. 4.38, p = 0.015*), allowing participants to easily designate specific targets."

Discussion¶

Related Claims¶

• Unmediated pen input introduced new pen-specific barriers: lack of visual feedback on referential intent and high interaction cost of handwriting complex queries — related
• In Study 1, learners preferred Penquiry V1 for part-scoped, specific-target and short questions, and favored the baseline for general topics and longer interpretive questions — related
• In Study 1, Penquiry V1 reduced physical demand and better maintained learning flow, and received higher TAM ratings, than a keyboard baseline — related
• Learners using standard keyboard-based LLM workflows during pen-based study face a Referential Barrier and an Expressive Barrier — related
• In Study 2, Content Snapping's immediate visual confirmation resolved learners' referential uncertainty about system recognition — related