Designing Transparent AI-Mediated Language Support for Intergenerational Family Communication Conference: Designing Interactive Systems Conference; June 13–17, 2026; Singapore, Singapore Designing Interactive Systems Conference (DIS Companion ’26), June 13–17, 2026, Singapore, Singapore DOI: 10.1145/3802974.3809413 ISBN: 979-8-4007-2632-3/2026/06 CCS: Human-centered computing User studies CCS: Human-centered computing Empirical studies in HCI CCS: Computing methodologies Natural language generation

Sora Kang, Youjin Hwang, Joonhwan Lee · 2026

Read Designing Transparent AI-Mediated Language Support for Intergenerational Family Communication Conference: Designing Interactive Systems Conference; June 13–17, 2026; Singapore, Singapore Designing Interactive Systems Conference (DIS Companion ’26), June 13–17, 2026, Singapore, Singapore DOI: 10.1145/3802974.3809413 ISBN: 979-8-4007-2632-3/2026/06 CCS: Human-centered computing User studies CCS: Human-centered computing Empirical studies in HCI CCS: Computing methodologies Natural language generation by Sora Kang, Youjin Hwang, Joonhwan Lee (2026) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2604-19276.txt.

Introduction

Kang, Hwang, and Lee build GenSync, a chat tool that helps family members across generations understand each other by showing translation openly. In a controlled study with 16 family pairs, they compare no translation, hidden translation, and open translation showing both the original and the interpreted text. Open translation supports conversation quality, closeness, and ease of use. Hidden translation, by contrast, often breaks the flow of talk.

Catalog record

CREATORSora Kang, Youjin Hwang, Joonhwan Lee
YEAR2026
COLLECTIONPAPERS
FORMATTXT
FILE SIZE19590
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2604.19276

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Designing Transparent AI-Mediated Language Support for Intergenerational Family Communication Conference: Designing Interactive Systems Conference; June 13–17, 2026; Singapore, Singapore Designing Interactive Systems Conference (DIS Companion ’26), June 13–17, 2026, Singapore, Singapore DOI: 10.1145/3802974.3809413 ISBN: 979-8-4007-2632-3/2026/06 CCS: Human-centered computing User studies CCS: Human-centered computing Empirical studies in HCI CCS: Computing methodologies Natural language generation Sora Kang, Youjin Hwang, Joonhwan Lee Abstract . Intergenerational linguistic differences pose challenges to effective and intimate family communication. This paper presents GenSync, a chat-based interface that supports intergenerational understanding through different forms of translation visibility. We conducted a controlled within-subjects study with 16 family dyads (32 participants), comparing three conditions: no translation, black-box translation, and transparent translation that displays both original and interpreted messages. The results show that translation visibility plays a critical role in shaping conversational experiences. Transparent translation supported conversational quality, intimacy, and usability, while black-box translation often disrupted conversational flow. These findings position intergenerational language support as a form of interpretive mediation and contribute design implications for AI-mediated communication in socially sensitive contexts. Figure 1 . GenSync interface. We evaluated three interface versions: (A) Control, with no translation support; (B) Black-box GenSync, showing only the interpreted message; and (C) Transparent GenSync, showing both the original and interpreted messages. GenSync interface. We evaluated three interface versions: (A) Control, with no translation support; (B) Black-box GenSync, showing only the interpreted message; and (C) Transparent GenSync, showing both the original and interpreted messages. † † cc-license: by 1. Introduction Communication is foundational to human relationships: it enables understanding, empathy, and connection ( 1 ; 17 ) and supports family well-being and mutual respect ( 18 ) . Beyond exchanging information, family communication helps build emotional bonds ( 15 ) and trust ( 13 ) . However, contemporary family interactions increasingly face intergenerational linguistic divides, where generation-specific slang, idioms, and online vernacular can make messages hard to interpret across age groups ( 16 ; 6 ; 3 ; 8 ) . Such gaps may lead to misinterpretations and feelings of distance within families ( 8 ; 10 ) . For example, parents born in the 1970s (and even more so grandparents) may struggle with messages like: “이번 기말고사 완전 억까 에바야 ” (ibeon gimalgosa wan-jeon eok-kka eba; “this final exam is completely unfair and crossing the line”). Recent advances in large language models have renewed interest in using AI to support human communication. Such interventions fall under the domain of AI-Mediated Communication, where an AI’s autonomy and role orientation fundamentally shape interpersonal dynamics ( 9 ) . In socially sensitive contexts like family dialogue, it is crucial to understand how the magnitude of AI intervention—specifically, the visibility of its translations—affects users’ sense of agency. However, it remains an open question how these systems should optimally intervene in intergenerational conversations, and how varying forms of intervention shape relational experiences. While people often adjust their vocabulary depending on their audience, intergenerational family interactions introduce additional social and relational constraints that complicate straightforward language adaptation. To explore this design space, we developed GenSync, a chat-based interface that supports intergenerational understanding by mediating generation-specific expressions. To evaluate GenSync, we conducted a controlled within-subjects study with 16 family dyads (32 participants). Each dyad experienced three interface conditions—(A) Control (no translation), (B) Black-box GenSync (interpreted message only), and (C) Transparent GenSync (original and interpreted messages)—in randomized order. During each condition, dyads engaged in text-based conversations across four everyday contexts (casual, emotional, information exchange, and entertainment-related). Our findings indicate that while translation can support intergenerational understanding, the visibility of translation plays a critical role in shaping how participants verify, interpret, and respond to translated messages in family conversations. Through the design of GenSync and a controlled empirical comparison of translation visibility, this work contributes insights into how AI-mediated language interfaces can be designed for socially sensitive communication contexts. 2. Method 2.1. GenSync GenSync is a GPT-4–based chat system designed to support intergenerational communication by rephrasing generation-specific expressions into more interpretable forms for another generation. The system relies on representative intergenerational expressions embedded in the system prompt as few-shot guidance ( 5 ) , enabling contextually appropriate interpretations without task-specific model training. 2.1.1. Prompt Design Each system prompt comprised three components: (1) a task instruction specifying the direction of intergenerational interpretation (e.g., slang to standard language or vice versa), (2) representative slang expressions provided as examples to guide model behavior, and (3) an output specification constraining responses to be concise and suitable for chat-based interaction while remaining sensitive to relational context (e.g., parent–child communication). 2.1.2. Slang Dictionary We constructed a slang dictionary focused on Kupsikche (K-slang), a form of generational slang commonly used by Generation Z in South Korea in informal, text-based communication. Publicly available expressions were collected from Naver and curated by two authors to remove obsolete, overly niche, or poorly contextualized items. Each entry included the slang expression, its interpretation in standard Korean, and a brief usage example. The resulting dictionary was incorporated directly into the system prompts as few-shot examples. 2.1.3. Implementation GenSync was implemented as a real-time, chat-based interface following interaction patterns familiar from everyday messaging platforms. For the experiment, we deployed three interface versions differing in interpretive transparency (Figure 1): Version A (Control) provided no interpretation support, Version B (Black-box) displayed only the system-generated interpretation, and Version C (Transparent) presented both the original message and its interpretation. 2.2. User Study We recruited 16 Korean family dyads (32 participants), each consisting of one younger family member and one older family member. Younger participants were adolescents aged 11–19 who reported frequent use of generation-specific slang in daily communication, while older participants ranged from 22 to 52 years. Dyads included 13 parent–child pairs and 3 sibling pairs. Age gaps between dyad members ranged from 10 to 36 years, and all participants used messaging applications regularly and communicated with their family member at least once per week. The study was conducted remotely via Zoom, with dyad members joining from separate rooms to simulate everyday text-based communication. After completing a pre-survey, participants were introduced to the GenSync interface and given time to familiarize themselves with its functionality. They were asked to communicate as naturally as they would with their family member. Each dyad experienced all three interface conditions in a randomized order and engaged in conversations across four themes: casual, emotional, informational, and entertainment-related. All interactions were conducted in Korean and subsequently translated into English for analysis by bilingual researchers. 2.3. Measures and Analysis Participants completed surveys during and after the experiment to assess conversation quality, family and intergenerational intimacy, and usability. To assess the system’s impact while minimizing participant fatigue—given our repeated-measures design across four conversation contexts—we selected and adapted single-item measures for each core construct from validated scales. Specifically, perceived usefulness was measured using an item adapted from the Technology Acceptance Model (TAM) ( 7 ) . Conversation quality and relational intimacy were assessed using items derived from established computer-mediated communication (CMC) and family interaction metrics ( 18 ; 15 ) . Overall continuance intention was measured based on the Expectation-Confirmation Model ( 4 ) . All items were rated on a 5-point Likert scale (1 = Strongly Disagree, 5 = Strongly Agree). Quantitative differences across interface conditions were examined using a one-factor

Read the complete text