Easy to Hear, Easy to Select?: Investigating Speech Rate as a Potential Deceptive Pattern in Non-Native Users’ Decision-Making

Easy to Hear, Easy to Select?: Investigating Speech Rate as a Potential Deceptive Pattern in Non-Native Users’ Decision-Making Yuichiro Kinoshita, Satoshi Nakamura (Meiji University)

Inspiration Chose what I could understand Just said “Yes” without fully understanding Can these designs steer non-native speakers’ decision-making? 2

Deceptive Patterns (Dark Patterns) Deceptive Patterns (DPs): Manipulative user interfaces that benefit service providers Multivitamin €30 Subscribe & Save One-time purchase Subscription emphasis Create your account Email Password I would like to receive updates via email Default opt-in setting DPs were used in 95% of the top 240 mobile apps [1] DPs can cause users financial loss and psychological burden [1] Di Geronimo et al.: UI Dark Patterns and Where to Find Them: A Study on Mobile Applications and User Perception. CHI’20 3

Deceptive Patterns in VUIs Most prior studies have focused on DPs in GUIs DPs also exist in voice user interfaces (VUIs) [2] VUIs may become more widely used in commercial settings - Voice AI is already used in drive-through ordering The use of DPs may also increase in VUIs [2] Owens et al.: Exploring Deceptive Design Patterns in Voice Interfaces. EuroUSEC’22 4

Related Work on DPs in VUIs Manipulating speech rate and pitch can steer users’ decision-making [3] Specific vocal characteristics can increase users’ trust in a system [4] Assuming interaction in users’ native language [3] Dubiel et al.: Impact of Voice Fidelity on Decision Making: A Potential Dark Pattern?. IUI’24 [4] Zhang et al.: The Manipulative Power of Voice Characteristics: Investigating Deceptive Patterns in Mandarin Chinese Female Synthetic Speech. Proc. ACM Interactive, Mobile, Wearable and Ubiquitous Technologies. 2025 5

Interacting in a Non-native Language VUIs do not adequately support users with diverse backgrounds [5] Non-native speakers often find it difficult to use VUIs [6] Non-native speakers may be more vulnerable to DPs in VUIs Focused on manipulating speech rate Investigated its influence on non-native speakers’ decision-making [5] Wenzel et al.: Designing for Harm Reduction: Communication Repair for Multicultural Users’ Voice Interactions. CHI’24 [6] Pyae et al.: Investigating the Role of User’s English Language Proficiency in Using a Voice User Interface: A case of Google Home Smart Speaker. CHI EA’19 6

Research Questions RQ1: Can non-native speakers’ selections be biased by deliberately manipulating speech rate? RQ2: Does the effect on non-native speakers’ selections change when the speech rate difference between options varies? 7

Experimental Scenario and Options Overseas travel scenario - Included a variety of choice situations Allowed participants to imagine using a non-native language Presented 15 questions, each with two options - e.g., “You are going to visit a garden. Which do you choose?” - Garden with roses and trees / Garden with lavender and trees Kept the number of words consistent across the two options Changed only part of each option to encourage participants to listen to the end 8

Experimental Setup Manipulated speech rate in five of the 15 questions - One option was faster, while the other was slower - Randomized which option was presented faster and which option was presented first - Audio was generated using Google Cloud Text-To-Speech Used four different speech rates and set six conditions covering all combinations - 113, 139, 165, and 191 WPM Between-participants experiment 9

Experimental System (191 & 113 WPM) 10

Results [1/3] Participants - Japanese people who were non-native speakers of English - 353 (175 males, 176 females, 1 transgender, 1 undisclosed) in total - 50–60 participants were randomly allocated to each of the six conditions Average number of audio playbacks - About 1.2 across all conditions - No significant differences across conditions 11

Results [2/3] Condition (WPM) Speed Diff. (WPM) 113 – 139 26 55.2% 44.8% .745 139 – 165 26 53.6% 46.4% .936 165 – 191 26 54.7% 45.3% .745 113 – 165 52 52.3% 47.7% .936 139 – 191 52 53.5% 46.5% .936 113 – 191 78 57.9% 42.1% .083 Slower selection Faster selection Adj. p-value

Results [3/3] Applied a binomial generalized linear mixed model (GLMM) - Fixed effect: Speech rate difference - Random effect: Participant No significant effect of speech rate difference Factors Estimate Std. Error Z p-value Intercept 0.105 0.133 0.795 .426 Speed Diff. 0.002 0.003 0.630 .529 13

Discussion [1/2] RQ1: Can non-native speakers’ selections be biased by deliberately manipulating speech rate? No evidence of selection bias from manipulating speech rate Consistent with prior work examining the effects of speech rate on permission requests in users’ native languages [7] - Neither the faster nor the slower option was selected more frequently Speech rate manipulation alone has little effect on decision-making in both native and non-native contexts [7] Leschanowsky et al.: Exploring the Impact of Modality and Speech Rate Manipulation in Voice Permission Requests— Limits of Applicability and Potential for Influencing Decision-Making. International Journal of Human-Computer Studies. 2025 14

Discussion [2/2] RQ2: Does the effect on non-native speakers’ selections change when the speech rate difference between options varies? Slight differences in selection rates, but no significant differences across conditions Humans tend to minimize cognitive effort and prefer information that is easier to process [8][9] Further research is needed to reach a conclusion [8] Oreg et al.: Prone to Bias: Development of a Bias Taxonomy from an Individual Difference Perspective. Review of General Psychology. 2009 [9] Reber et al.: Effects of Perceptual Fluency on Affective Judgments. Psychological Science. 1998 15

Limitations Participants were Japanese non-native speakers of English - Results may differ across cultures Used an overseas travel scenario - Findings may not generalize to other contexts Potentially limited sample size - Selection rate difference was 15 percentage points (113–191 WPM condition) - We determined the sample size assuming a larger difference in selection rates 16

Conclusion & Future Work No clear evidence that manipulating speech rate functions as a deceptive pattern for non-native speakers Investigate other scenarios where deceptive patterns could be applied, such as shopping Examine fully voice-based interactions, where options are presented by voice and participants respond by voice 17