LIS users recognize Pepper's isolated signs far better than its full signed sentences in an exploratory evaluation¶

Claim · All claims
Evidence · 1 study (2 entries) · design r2 · q2
Currency · newest study 2025 · not yet checked against its sources · not reviewed by a person · recheck by 2028: tool and model accuracy changes quickly

Subclaims¶

q2 i? Performance on open-ended signed-sentence interpretation was notably lower than on single-choice isolated-sign recognition among 12 LIS-proficient participants. → Giulia Boaa 2025 q2 i? For the sentence 'Ho mangiato una mela', only one participant correctly interpreted the full sentence, with most confusing the sign for apple with the sign for water. → Giulia Boaa 2025 (2)

Evidence¶

Giulia Boaa 2025¶

Giulia Boaa, Marco Boaa, Cristina Gena, Alessandro Mazzei, Massimo Donini, and Alberto Lillo. (2025). Using the Pepper Robot to Support Sign Language Communication. https://github.com/CodeCruncher63/ISCR2025 doi:10.1007/978-981-95-2382-5_8

q2 · i? · design · r2

Exploratory online questionnaire study with 12 participants (4 LIS interpreters, 8 Deaf LIS users) comparing 15 single-choice sign videos against 2 open-ended signed sentences. The article reports the open-ended section was "notably lower than in the single-choice sec0on".

"Performance in the open-ended sec0on was notably lower than in the single-choice sec0on, sugges0ng greater difficulty in interpre0ng full signed sentences:"

Giulia Boaa 2025 (2)¶

Giulia Boaa, Marco Boaa, Cristina Gena, Alessandro Mazzei, Massimo Donini, and Alberto Lillo. (2025). Using the Pepper Robot to Support Sign Language Communication. https://github.com/CodeCruncher63/ISCR2025

q2 · i? · design · r2

Open-ended sentence task in the same 12-participant questionnaire: for the sentence mela–mangiare–fatto, "Only one par0cipant correctly interpreted the full sentence" while the other 11 recognized at least 2 of 3 signs.

"Most par0cipants interpreted the sequence as “Ho bevuto e mangiato” (I drank and ate), likely due to the confusion between the signs for “apple” and “water”. Only one par0cipant correctly interpreted the full sentence."

Discussion¶

Related Claims¶

• A MATLAB inverse-kinematics pipeline automated generation of 38 of 52 LIS signs on Pepper and cut authoring time from 45–60 to under 10 minutes per sign — related
• Recognition of individual Pepper-produced LIS signs varies widely, with several signs significantly above chance and others not significant — related
• Thematic coding of sentence misinterpretations yields three categories: lexical confusion, structure misinterpretation, and partial recognition with wrong sequencing — related