Pith. sign in

REVIEW 3 major objections 5 minor 57 references

ConversAR: Exploring Embodied LLM-Powered Group Conversations in Augmented Reality for Second Language Learners

T0 review · 3 major / 5 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read ConversAR, an AR app with two gpt-4o-powered embodied agents, gives second-language learners a group conversation practice environment that its 10-user study links to reduced speaking anxiety and increased autonomy.

desk verdict A genuine first demo of multi-agent AR L2 group conversation, but the simulated scene understanding and self-report-only evaluation mean the abstract overclaims; worth a serious referee with revisions. read the letter →

arxiv 2505.24000 v1 pith:VFXL43DW submitted 2025-05-29 cs.HC

Jad Bendarkawi , Ashley Ponce , Sean Mata , Aminah Aliu , Yuhan Liu , Lei Zhang , Amna Liaqat , Varun Nagaraj Rao
show 1 more author
Andrés Monroy-Hernández
This is my paper
classification cs.HC
keywords AugmentedRealityLargeLanguageModelsEmbodiedAgentsSecondAcquisitionGroupConversationSpeakingAnxietyLearnerAutonomyWizardofOz
verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

This paper argues that group conversation practice is valuable for second-language (L2) learners but often blocked by speaking anxiety, and that existing AR tools only support one-on-one chats. To fill the gap, the authors built ConversAR, a Meta Quest 3 app in which two embodied, gpt-4o-powered agents hold a group conversation with the learner in the target language, grounded in a snapshot of the physical environment. In a 10-participant study with intermediate-to-advanced Spanish learners, participants reported reduced speaking anxiety, increased learner autonomy, and greater willingness to take linguistic risks compared with their perceptions of in-person group practice. The paper's contribution is the design and early evaluation of the first AR system to use multiple embodied LLM agents for real-time L2 group conversation practice.

What carries the argument

The load-bearing mechanism is the pairing of two embodied LLM agents with a separate 'Moderator LLM' that manages turn-taking. The two agents (gpt-4o) take turns speaking to each other and the user, with a three-second gap that lets the user interject by holding the controller trigger; after the user speaks, the Moderator selects which agent responds. Speech goes through a whisper-1 speech-to-text pipeline and the audio API's text-to-speech, with live captions under each agent. A zero-shot prompted parameterized prompt injects the target language, level, scene context, and full conversation history, keeping the dialogue grounded. Scene understanding is presented as vision-based object detection, but in this iteration it is simulated: a researcher hardcodes the object list and scene context, and the user sees a "Detecting Environment..." indicator.

What would settle it

Run ConversAR in a room the researchers never visited, with a real automated object-detection pipeline in place of the hardcoded scene context, and compare self-reported and physiological speaking anxiety against a matched in-person group conversation. If the anxiety reduction disappears, or if participants notice that the system's understanding of the room is wrong, the claim that vision-based scene understanding contributes to the observed benefits would not survive.

Watch

Extended reading notes

Core claim

The central claim is that a multi-agent AR conversation environment can give L2 learners the practice benefits of group talk, such as turn-taking, interjecting, listening to different voices, and observing others handle the conversation, without the social pressure of a human group. ConversAR places two humanoid gpt-4o agents in the user's physical space; the agents converse with each other and the user, a separate 'Moderator LLM' decides which agent responds after the user speaks, and live captions plus natural text-to-speech voices support comprehension. Evaluated on 10 university students of intermediate or higher Spanish, the system yielded self-reported speaking anxiety lower than perceptions of in-person methods, higher perceived learner autonomy, and comparable engagement, with participants saying they felt free to make mistakes and to steer the topic. The authors position this as the first system using AR with multiple embodied LLM agents for real-time L2 group conversation practice.

Load-bearing premise

The system's scene understanding is not actually implemented: a researcher pre-hardcodes the objects and context before the user puts on the headset, and the "Detecting Environment" indicator is a simulation, so the anxiety and autonomy results were measured against a system that may not behave like the advertised vision-based version.

Editorial extensions

If this is right

  • If the effect holds, L2 learners can rehearse group conversation dynamics, including turn-taking, interruption, and waiting for a gap, without fear of peer judgment, potentially building fluency before entering real groups.
  • Reducing anxiety and increasing autonomy may encourage learners to take linguistic risks, such as trying new vocabulary or sentence structures, more often than in classroom settings.
  • Anchoring the conversation to the physical environment makes the practice contextual, which prior work links to better retention and motivation.
  • The multi-agent format adds what dyadic practice lacks: passive observation of other speakers and the freedom to pass on a question while other interlocutors continue the discussion.
  • If real object detection replaces the pre-hardcoded scene data, the architecture is designed to generalize across physical contexts such as cafés and offices.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

  • The Wizard of Oz design means the study evaluated a simulated version of the headline feature; the 1–2 second speech-to-text delay and zero recorded social backchannels suggest that an automated vision pipeline may behave differently and that emotional engagement with the agents is still shallow.
  • Because the anxiety comparison relied on self-reports of past in-person experiences rather than a controlled in-person condition, the size of the reported benefit may shrink under direct behavioral or physiological measurement.
  • A natural next test is to vary the number of agents, from dyad to triad and beyond, to map when the group format shifts from supportive to cognitively overwhelming, a concern the paper itself raises for less confident learners.
Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Referee Report

3 major / 5 minor

Summary. The paper presents ConversAR, a Meta Quest 3 AR application that places two embodied, gpt-4o-powered agents in the user's physical environment to enable group conversation practice for second-language (L2) learners. The system is claimed to feature vision-based scene understanding and live captions. The authors report a 10-participant user study with self-report surveys and semi-structured interviews, finding that participants perceived reduced speaking anxiety, increased learner autonomy, and greater willingness to take linguistic risks compared to their remembered in-person L2 practice. The paper also documents system architecture, a pilot study, engagement coding, and a participant table with AP-style conversational depth scores.

Significance. The paper is a useful exploratory contribution to the emerging area of LLM-powered AR language learning. It addresses a genuinely under-explored configuration: multiple embodied LLM agents in AR for group conversation practice, rather than dyadic setups. The authors are transparent about several limitations, including the small sample, self-report measures, and the Wizard of Oz simulation of scene understanding, and they provide detailed appendices (interview guide, survey, engagement metrics, participant table) that support replication and extension. However, the central claim of 'vision-based scene understanding' is contradicted by the disclosed implementation, and the quantitative evidence for the anxiety-reduction comparison is not reported in sufficient detail. These issues affect the paper's central claims, though they are fixable within the scope of an extended abstract.

major comments (3)
  1. [Section 3.1; abstract; Section 1] The abstract and introduction state that ConversAR features 'vision-based scene understanding' and that 'the agents can understand the user's environment through object detection.' Section 3.1 then discloses that object detection is not implemented: a researcher hardcodes objects and scene context before each session, and users see a 'Detecting Environment...' indicator to create an illusion of automated detection. This means the evaluated system is not the system described in the central claim. Because participants always experienced a hand-picked context that matched the physical room, the evaluation cannot separate the effects of automated scene understanding from the effects of a pre-selected, perfectly aligned context. The paper should either revise the abstract and introduction to describe the current prototype as using pre-specified context with a simulated detection step, or implement and evaluate the real-time detection pipeline. As written, the claim is an overclaim.
  2. [Section 5.1.1; Appendix A.3] The claimed reduction in speaking anxiety compared to in-person methods is based on retrospective self-report items from Appendix A.2, and the quantitative results are presented only as Figure 4, without reporting means, standard deviations, effect sizes, or any inferential statistical test. Because the abstract and Section 5.1.1 make a comparative claim ('reduced speaking anxiety'), the paper should report the paired differences (e.g., median or mean difference, range, and the number of participants who reported lower anxiety with ConversAR than with in-person methods). Without these numbers, the magnitude and consistency of the effect cannot be assessed, and the claim is not quantitatively verifiable as reported.
  3. [Section 4; Section 9] With n=10, no control condition, and self-selected participants, the study is a preliminary system demonstration, and the authors appropriately acknowledge these limitations in Section 7. However, the conclusion states that 'the system was associated with reduced speaking anxiety,' which implies an association that the design cannot establish. The framing should be revised consistently to something like 'participants reported lower anxiety with ConversAR than in their remembered in-person experiences,' matching the self-report nature of the data.
minor comments (5)
  1. [Section 2.2.3] The claim of being 'the first system' of its kind should be qualified with 'to the best of our knowledge' or 'among systems surveyed,' since a comprehensive literature scan is not presented.
  2. [Section 3.1] The sentence stating that the Wizard of Oz approach 'mimics the responsiveness and accuracy expected from an automated system' is misleading, because a hand-picked context cannot reproduce realistic detection errors, latency, or variability. Suggest rewording to describe it as a simulation that controls context across participants.
  3. [Section 3.3.3] The phrase 'we determined an optimal gap of 3 seconds' is too strong; since no systematic optimization is reported, suggest 'we set a gap of 3 seconds' or 'we found 3 seconds to work well in pilot testing.'
  4. [Section 4] The engagement metrics were coded by a single researcher with no reported inter-rater reliability; please add a note about whether a second coder was used or list this as a limitation.
  5. [Appendix A.5] The participant table reports mean anxiety (5.1) and mean conversational depth (3.5) but no standard deviations; adding them would help the reader interpret the variability.

Circularity Check

0 steps flagged · score 0.0 of 10

No circularity: reported outcomes are direct self-reports; Wizard-of-Oz scene understanding is a validity limitation rather than a circular reduction.

full rationale

The paper's central claims ('users reported reduced speaking anxiety and increased learner autonomy') are empirical self-report outcomes from a 10-participant survey (Section 5), not quantities derived from fitted inputs. No parameter is fit to the outcome data, no prediction is computed from an equation, and no self-citations appear in the reference list. The Wizard-of-Oz disclosure (Section 3.1: 'one of the researchers hardcodes a list of objects and the scene context onto the device... to create the illusion that the system has detected their immediate surroundings') means the abstract's 'vision-based scene understanding' was not actually tested; this is an overclaim and an external-validity threat, not circularity, because the anxiety and autonomy reports are not defined in terms of, or statistically forced by, the hardcoded context. The supportive agent personas are a deliberate design choice, and the self-report measures could have come out negative; they are not constructed to equal the design goal by definition. The novelty claim is benchmarked against cited external systems, not against the authors' own prior theorems. Section 7's limitations appropriately acknowledge the self-report comparison and small sample. Verdict: no significant circularity; score 0.

Assumptions & free parameters 1 free parameters · 3 assumptions · 0 invented entities

The central claims rely on three main assumptions: the competence of external LLM/STT models, the adequacy of the Wizard of Oz simulation for the claimed scene understanding feature, and the validity of self-report measures. The 3-second inter-agent gap is the only hand-tuned numeric parameter. No new physical or conceptual entities are introduced beyond the software agents themselves.

free parameters (1)
  • Conversational gap between agent responses = 3 seconds
    Chosen via pilot studies as the optimal pause between agent utterances (Section 3.3.3); not derived from theory or independently validated.
assumptions (3)
  • domain assumption OpenAI gpt-4o and whisper-1 perform adequately for real-time Spanish conversation
    The entire system depends on the closed-source models; the paper only reports anecdotal quality observations (Section 3.3.2).
  • domain assumption Wizard of Oz hardcoded scene context is a faithful stand-in for vision-based scene understanding
    Section 3.1 discloses that the object detection is simulated by a researcher, while the abstract claims 'vision-based scene understanding'; validity of the simulation for the study's conclusions is unexamined.
  • domain assumption Self-reported Likert scales and interviews validly measure speaking anxiety and learner autonomy
    The central findings are based on participants' self-assessment after a single 10-minute session, with no behavioral or physiological verification (Sections 4 and 5).

how reviews work

0 comments
Cite this review

Pith. "Pith review of ConversAR: Exploring Embodied LLM-Powered Group Conversations in Augmented Reality for Second Language Learners." pith.science (2026). https://pith.science/paper/VFXL43DW

@misc{pith2026250524000,
  author       = {Pith},
  title        = {Pith review of: ConversAR: Exploring Embodied LLM-Powered Group Conversations in Augmented Reality for Second Language Learners},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/VFXL43DW}},
  note         = {Machine review of arXiv:2505.24000}
}
read the original abstract

Group conversations are valuable for second language (L2) learners as they provide opportunities to practice listening and speaking, exercise complex turn-taking skills, and experience group social dynamics in a target language. However, most existing Augmented Reality (AR)-based conversational learning tools focus on dyadic interactions rather than group dialogues. Although research has shown that AR can help reduce speaking anxiety and create a comfortable space for practicing speaking skills in dyadic scenarios, especially with Large Language Model (LLM)-based conversational agents, the potential for group language practice using these technologies remains largely unexplored. We introduce ConversAR, a gpt-4o powered AR application, that enables L2 learners to practice contextualized group conversations. Our system features two embodied LLM agents with vision-based scene understanding and live captions. In a system evaluation with 10 participants, users reported reduced speaking anxiety and increased learner autonomy compared to perceptions of in-person practice methods with other learners.

Figures

Figures reproduced from arXiv: 2505.24000 by the authors.

Figure 1
Figure 1. Demonstration of the ConversAR system. (A) Title screen where the user selects the language (Spanish) and proficiency level (intermediate) then observes the system detecting the surrounding environment. (B) The user holds the right controller trigger to speak into the system. (C) One of the LLM agents responds to the user with live captions appearing underneath them. ABSTRACT Group conversations are valuable for sec… view at source ↗
Figure 2
Figure 2. System architecture of ConversAR. The architecture highlights core components and data flow within the system.1 AR approach, users can engage in conversations that reflect their surroundings for more relevant L2 practice across diverse contexts. We iteratively developed and refined the system by conducting a pilot study with two adult university undergraduate students enrolled in an intermediate Spanish course. Each… view at source ↗
Figure 3
Figure 3. Example conversation flow between a user and the two agents in English, situated in a café. After the user speaks, [PITH_FULL_IMAGE:figures/full_fig_p004_3.png] view at source ↗
Figures from the paper (2 more)
Figure 4
Figure 4. Figure 4: Graphs comparing user response on the perceived [PITH_FULL_IMAGE:figures/full_fig_p010_4.png]
Figure 5
Figure 5. Figure 5: Engagement Breakdown per Participant. The num [PITH_FULL_IMAGE:figures/full_fig_p010_5.png]

Discussion (0). Sign in to comment.

Reference graph

Works this paper leans on

57 extracted references · 38 canonical work pages

  1. [1]

    Suhni Abbasi and Hameedullah Kazi. 2014. Measuring Effectiveness of Learning Chatbot Systems on Student’s Learning Outcome and Memory Retention. Asian Journal of Applied Science and Engineering 3 (Aug. 2014), 57–66. https://doi.org/ 10.15590/ajase/2014/v3i7/53576

    work page doi:10.15590/ajase/2014/v3i7/53576 2014

  2. [2]

    Alireza Ahmadi and Zahra Montasseri. 2019. Interactional Competence in Paired vs. Group Oral Tests. Teaching English Language 13, 1 (April 2019), 1–26. https: //doi.org/10.22132/tel.2019.86905 Publisher: Teaching English Language and Literature Society of Iran (TELLSI)

    work page arXiv 2019

  3. [3]

    Eshtiak Ahmed and Aino Ahtinen. 2021. Design Implications for a Virtual Language Learning Companion Robot: Considering the Appearance, Interaction and Rewarding Behavior. In Proceedings of the 9th International Conference on Human-Agent Interaction. ACM, Virtual Event Japan, 56–65. https://doi.org/10. 1145/3472307.3484163

    work page arXiv 2021

  4. [4]

    Emmanuel Ayedoun, Yuki Hayashi, and Kazuhisa Seta. 2015. A Conversational Agent to Encourage Willingness to Communicate in the Context of English as a Foreign Language. Procedia Computer Science 60 (Jan. 2015), 1433–1442. https://doi.org/10.1016/j.procs.2015.08.219

    work page doi:10.1016/j.procs.2015.08.219 2015

  5. [5]

    Gene Ball and Jack Breese. 2000. Emotion and Personality in a Conversational Agent. In Embodied Conversational Agents, Justine Cassell, Joseph Sullivan, Scott Prevost, and Elizabeth F. Churchill (Eds.). The MIT Press, Cambridge, MA, USA, 189–219. https://doi.org/10.7551/mitpress/2697.003.0009

    work page doi:10.7551/mitpress/2697.003.0009 2000

  6. [6]

    John Seely Brown, Allan Collins, and Paul Duguid. 1989. Situated Cognition and the Culture of Learning. Educational Researcher 18, 1 (1989), 32–42. https: //doi.org/10.3102/0013189X018001032

    doi:10.3102/0013189x018001032 1989

  7. [7]

    Thorisson

    Justine Cassell and Kristinn R. Thorisson. 1999. The power of a nod and a glance: Envelope vs. emotional feedback in animated conversational agents. Applied Artificial Intelligence 13, 4-5 (May 1999), 519–538. https: //doi.org/10.1080/088395199117360 Publisher: Taylor & Francis _eprint: https://doi.org/10.1080/088395199117360

    work page doi:10.1080/088395199117360 1999

  8. [8]

    Rubén Chacón-Beltrán. 2010. The Acquisition of Sociolinguistic Competence in a Study Abroad Context by Vera Regan, Martin Howard and Isabelle Lemée.Journal of Sociolinguistics 14, 3 (2010), 423–424. https://doi.org/10.1111/j.1467-9841. 2010.00447_7.x _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/j.1467- 9841.2010.00447_7.x

    work page Pith review arXiv 2010

Show all 57 references
  1. [9]

    Jingying Chen, Jian Dai, Keke Zhu, and Liujie Xu. 2022. Effects of extended reality on language learning: A meta-analysis. Frontiers in Psychology 13 (Sept. 2022), 13 pages. https://doi.org/10.3389/fpsyg.2022.1016519 Publisher: Frontiers

    2022

  2. [10]

    Cheng, Meng Guo, Melissa Ran, Arpit Ranasaria, Arjun Sharma, Anthony Xie, Khuyen N

    Alan Y. Cheng, Meng Guo, Melissa Ran, Arpit Ranasaria, Arjun Sharma, Anthony Xie, Khuyen N. Le, Bala Vinaithirthan, Shihe (Tracy) Luan, David Thomas Henry Wright, Andrea Cuadra, Roy Pea, and James A. Landay. 2024. Scientific and Fantastical: Creating Immersive, Culturally Rele...

    2024

  3. [11]

    College Board. 2018. AP Spanish Language and Culture Scoring Guidelines. College Board. https://secure-media.collegeboard.org/ap/pdf/ap18-sg-spanish-language. pdf

    2018

  4. [12]

    Yan Cong. 2024. AI Language Models: An Opportunity to Enhance Lan- guage Learning. Informatics 11, 3 (Sept. 2024), 49. https://doi.org/10.3390/ informatics11030049 Number: 3 Publisher: Multidisciplinary Digital Publishing Institute

    2024

  5. [13]

    Yan Cong. 2025. Demystifying large language models in second language development research. Computer Speech & Language 89 (Jan. 2025), 101700. https://doi.org/10.1016/j.csl.2024.101700

    2025

  6. [14]

    Robert O Davis, Taejung Park, and Joseph Vincent. 2023. A Meta-Analytic Review on Embodied Pedagogical Agent Design and Testing Formats. Journal of Educational Computing Research 61, 1 (March 2023), 30–67. https://doi.org/10. 1177/07356331221100556 Publisher: SAGE Publications Inc

    2023

  7. [15]

    Divekar*, Jaimie Drozdal*, Samuel Chabot*, Yalun Zhou, Hui Su, Yue Chen, Houming Zhu, James A

    Rahul R. Divekar*, Jaimie Drozdal*, Samuel Chabot*, Yalun Zhou, Hui Su, Yue Chen, Houming Zhu, James A. Hendler, and Jonas Braasch. 2022. Foreign lan- guage acquisition via artificial intelligence and extended reality: design and evaluation. Computer Assisted Language Learning...

    2022

  8. [16]

    Ermert, Chinthusa Mohanathasan, Sabine J

    Jonathan Ehret, Andrea Bönsch, Patrick Nossol, Cosima A. Ermert, Chinthusa Mohanathasan, Sabine J. Schlittmeier, Janina Fels, and Torsten W. Kuhlen

  9. [17]

    Rod Ellis. 1997. The empirical evaluation of language teaching materials. ELT Journal 51, 1 (Jan. 1997), 36–42. https://doi.org/10.1093/elt/51.1.36

    1997 doi

  10. [18]

    Rod Ellis. 2012. Language Teaching Research and Language Pedagogy . John Wiley & Sons, Hoboken, NJ, USA. Google-Books-ID: XZ4dqBFDiwEC

    2012

  11. [19]

    Talking Circle

    Gisela Ernst. 1994. "Talking Circle": Conversation and Negotiation in the ESL Classroom. TESOL Quarterly 28, 2 (1994), 293–322. https://doi.org/10.2307/ 3587435 Publisher: [Wiley, Teachers of English to Speakers of Other Languages, Inc. (TESOL)]

    1994

  12. [20]

    Murunga Felicity. 2018. Speaking Anxiety and its Effects on Participation in Group Discussion in L2 Classrooms. International Journal of English Language Teaching 6, 3 (2018), 20–33

    2018

  13. [21]

    Zhengdong Gan. 2010. Interaction in group oral assessment: A case study of higher- and lower-scoring students. Language Testing 27, 4 (Oct. 2010), 585–602. https://doi.org/10.1177/0265532210364049 Publisher: SAGE Publications Ltd

    2010 doi

  14. [22]

    Shanna Li Ching Hollingworth and Wesley Willett. 2023. FluencyAR: Augmented Reality Language Immersion. In Adjunct Proceedings of the 36th Annual ACM Symposium on User Interface Software and Technology . ACM, San Francisco CA USA, 1–3. https://doi.org/10.1145/3586182.3616670

    2023

  15. [23]

    Horwitz, Michael B

    Elaine K. Horwitz, Michael B. Horwitz, and Joann Cope. 1986. Foreign Language Classroom Anxiety. The Modern Language Journal 70, 2 (1986), 125–132. https: //doi.org/10.2307/327317 Publisher: [National Federation of Modern Language Teachers Associations, Wiley]

    1986 doi

  16. [24]

    Nan Huang. 2022. Revisiting L2 pragmatic competence through implicit vs. explicit instructional framework. Frontiers in Psychology 13 (Aug. 2022), 8 pages. https://doi.org/10.3389/fpsyg.2022.987729 Publisher: Frontiers

    2022

  17. [25]

    Dhruv Jain, Bonnie Chinh, Leah Findlater, Raja Kushalnagar, and Jon Froehlich

  18. [26]

    Curenton

    Cecilia Jarquin Tapia, Sarah Surrain, and Stephanie M. Curenton. 2022. The Importance of Dyadic Classroom Conversations for Dual Language Learners. The Reading Teacher 75, 6 (2022), 777–781. https://doi.org/10.1002/trtr.2089 _eprint: https://onlinelibrary.wiley.com/doi/pdf/10....

    2022 doi

  19. [27]

    Lewis Johnson and James C

    W. Lewis Johnson and James C. Lester. 2018. Pedagogical Agents: Back to the Future. AI Magazine 39, 2 (July 2018), 33–44. https://doi.org/10.1609/aimag. v39i2.2793 Number: 2

    2018 doi

  20. [28]

    Minsol Kim, Aliea L Nallbani, and Abby Rayne Stovall. 2024. Exploring LLM- based Chatbot for Language Learning and Cultivation of Growth Mindset. In Extended Abstracts of the 2024 CHI Conference on Human Factors in Computing Systems (CHI EA ’24). Association for Computing Mach...

    2024

  21. [29]

    Craig Lambert, Jenefer Philp, and Sachiko Nakamura. 2017. Learner-generated content and engagement in second language task performance. Lan- guage Teaching Research 21, 6 (Nov. 2017), 665–680. https://doi.org/10.1177/ 1362168816683559 Publisher: SAGE Publications

    2017

  22. [30]

    Hyungmin Lee, Chen-Chun Hsia, Aleksandr Tsoy, Sungmin Choi, Hanchao Hou, and Shiguang Ni. 2023. VisionARy: Exploratory research on Contextual Language Learning using AR glasses with ChatGPT. In Proceedings of the 15th Biannual Conference of the Italian SIGCHI Chapter . ACM, To...

    2023

  23. [31]

    Sangmin-Michelle Lee, Xiaoran Wang, Ilee Park, and Riski Lestiono. 2024. It feels so real! Situated authentic language learning in immersive virtual reality. Education and Information Technologies 29 (June 2024), 24023–24045. https: //doi.org/10.1007/s10639-024-12807-8

    2024 doi

  24. [32]

    Natalie Lefkowitz. 2009. Second Language Learning and Language Teaching (4th ed.). Vivian Cook. London: Arnold, 2008. Pp. x + 306. Studies in Sec- ond Language Acquisition 31, 4 (Dec. 2009), 660–661. https://doi.org/10.1017/ S0272263109990180

    2009

  25. [33]

    Kuo-Yu Liu, Yuanshan Chen, Ming-Fang Lin, Li-Jung Daphne Huang, and Cheah Ping Xiang. 2024. Developing a VR-based contextualized language learning system to Enhance Junior High School Students’ Pragmatic Competence. In Proceedings of the 5th ACM International Conference on Mul...

    2024

  26. [34]

    Song Liu, Shengbing Gao, and Xiaoyan Ji. 2023. Beyond borders: exploring the impact of augmented reality on intercultural competence and L2 learning motivation in EFL learners. Frontiers in Psychology 14 (Oct. 2023), 15 pages. https://doi.org/10.3389/fpsyg.2023.1234905 Publish...

    2023

  27. [35]

    MacIntyre and R

    Peter D. MacIntyre and R. C. Gardner. 1994. The Subtle Effects of Language Anxiety on Cognitive Processing in the Second Language. Language Learning 44, 2 (1994), 283–305. https://doi.org/10.1111/j.1467-1770.1994.tb01103.x _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.11...

    1994

  28. [36]

    Kathy A. Mills. 2022. Potentials and Challenges of Extended Reality Technologies for Language Learning. Anglistik 33, 1 (2022), 147–163. https://doi.org/10.33675/ ANGL/2022/1/13 Publisher: Universitätsverlag WINTER GmbH Heidelberg

    2022

  29. [37]

    Ezihaslinda Ngah and Siti Hamin Stapa. 2019. Exploring the Relationship between Interactional Competence and Group Oral Communication. International Journal of Language Education and Applied Linguistics 9, 2 (Dec. 2019), 55–65. https: //doi.org/10.15282/ijleal.v9.300

    2019 doi

  30. [38]

    Norris and Lourdes Ortega

    John M. Norris and Lourdes Ortega. 2000. Effectiveness of L2 Instruc- tion: A Research Synthesis and Quantitative Meta-analysis. Language Learn- ing 50, 3 (2000), 417–528. https://doi.org/10.1111/0023-8333.00136 _eprint: https://onlinelibrary.wiley.com/doi/pdf/10.1111/0023-833...

    2000

  31. [39]

    Hinojos, Aaron E

    David Novick, Laura J. Hinojos, Aaron E. Rodriguez, Adriana Camacho, and Mahdokht Afravi. 2018. Conversational Interaction with Multiple Agents Initiated via Proxemics and Gaze. In Proceedings of the 6th International Conference on Human-Agent Interaction. ACM, Southampton Uni...

    2018

  32. [40]

    Mengxu Pan, Alexandra Kitson, Hongyu Wan, and Mirjana Prpa. 2024. ELLMA-T: an Embodied LLM-agent for Supporting English Language Learning in Social VR. http://arxiv.org/abs/2410.02406 arXiv:2410.02406

    2024 arXiv

  33. [41]

    Wim Pouw and Judith Holler. 2022. Timing in conversation is dynamically adjusted turn by turn in dyadic telephone conversations. Cognition 222 (May 2022), 105015. https://doi.org/10.1016/j.cognition.2022.105015

    2022

  34. [42]

    Pranab Sahoo, Ayush Kumar Singh, Sriparna Saha, Vinija Jain, Samrat Mondal, and Aman Chadha. 2024. A Systematic Survey of Prompt Engineering in Large Language Models: Techniques and Applications. https://doi.org/10.48550/arXiv. 2402.07927 arXiv:2402.07927 [cs]

    2024 doi arXiv

  35. [43]

    Plecher, Christian Eichhorn, and Gudrun Klinker

    Isabel Schorr, David A. Plecher, Christian Eichhorn, and Gudrun Klinker. 2024. Foreign language learning using augmented reality environments: a systematic review. Frontiers in Virtual Reality 5 (Feb. 2024), 15 pages. https://doi.org/10. 3389/frvir.2024.1288824 Publisher: Frontiers

    2024

  36. [44]

    Sinan Sonlu, Bennie Bendiksen, Funda Durupinar, and Uğur Güdükbay. 2024. The Effects of Embodiment and Personality Expression on Learning in LLM-based Educational Agents. https://arxiv.org/abs/2407.10993v1

    2024 arXiv

  37. [45]

    Aimee Sousa Calepso, Natalie Hube, Noah Berenguel Senn, Vincent Brandt, and Michael Sedlmair. 2022. cARdLearner: Using Expressive Virtual Agents when Learning Vocabulary in Augmented Reality. InExtended Abstracts of the 2022 CHI Conference on Human Factors in Computing Systems...

    2022

  38. [46]

    Robert Stroud. 2017. Second Language Group Discussion Participation: A Closer Examination of ’Barriers’ and ’Boosts’. Proceedings of the International Conference on Education and Learning (ICEL) 2017 1 (01 2017), 40–57

    2017

  39. [47]

    John Sweller, Stéphanie Roussel, and André Tricot. 2022. Cognitive Load Theory and Instructional Design for Language Learning. In The Cambridge Handbook of Working Memory and Language , John W. Schwieter and Zhisheng (Edward) Wen (Eds.). Cambridge University Press, Cambridge, ...

    2022 doi

  40. [48]

    Vander Tavares. 2016. The Role of Peer Interaction and Second Language Learning for ESL Students in Academic Contexts: An Extended Literature Review . Master’s thesis. York University, Toronto, Canada. http://hdl.handle.net/10315/34200

    2016

  41. [49]

    Lili Tian and Yan Jiang. 2021. L2 Proficiency Pairing, Task Type and L1 Use: A Mixed-Methods Study on Optimal Pairing in Dyadic Task-Based Peer Interaction. Frontiers in Psychology 12 (June 2021), 12 pages. https://doi.org/10.3389/fpsyg. 2021.699774 Publisher: Frontiers

    2021

  42. [50]

    Dolly Jesusita Young. 1991. Creating a Low-Anxiety Classroom Environment: What Does Language Anxiety Research Suggest? The Modern Language Jour- nal 75, 4 (1991), 426–439. https://doi.org/10.2307/329492 Publisher: [National Federation of Modern Language Teachers Associations, Wiley]

    1991 doi

  43. [51]

    Reza Zabihi and Behnoush Ghahramanzadeh. 2022. Proficiency pairing and engagement with language in L2 peer interaction. System 105 (April 2022), 102725. https://doi.org/10.1016/j.system.2022.102725

    2022

  44. [52]

    Zhihui Zhang and Xiaomeng Huang. 2024. The impact of chatbots based on large language models on second language vocabulary acquisition. Heliyon 10, 3 (2024), e25370. https://doi.org/10.1016/j.heliyon.2024.e25370

    2024 doi

  45. [53]

    Yijun Zhao, Jiangyu Pan, Yan Dong, Tianshu Dong, Guanyun Wang, Fangtian Ying, Qihang Shen, and Jiacheng Cao. 2024. Language Urban Odyssey: A Serious Game for Enhancing Second Language Acquisition through Large Language Mod- els. In Extended Abstracts of the CHI Conference on H...

    2024

  46. [54]

    Yuying Zhi and Lihuan Wu. 2023. Extended reality in language learning: A cog- nitive affective model of immersive learning perspective. Frontiers in Psychology 14 (Feb. 2023), 5 pages. https://doi.org/10.3389/fpsyg.2023.1109025 Publisher: Frontiers

    2023

  47. [55]

    in-person methods

    Katja Zibrek and Rachel McDonnell. 2014. Does render style affect perception of personality in virtual humans?. In Proceedings of the ACM Symposium on Applied Perception. ACM, Vancouver British Columbia Canada, 111–115. https: //doi.org/10.1145/2628257.2628270 ConversAR: Explo...

    2014

  48. [2018]

    In Proceedings of the 2018 ACM Conference Companion Publication on Designing Interactive Systems

    Exploring Augmented Reality Approaches to Real-Time Captioning: A Preliminary Autoethnographic Study. In Proceedings of the 2018 ACM Conference Companion Publication on Designing Interactive Systems. ACM, Hong Kong China, 7–11. https://doi.org/10.1145/3197391.3205404

    2018

  49. [2023]

    In Proceedings of the 23rd ACM International Con- ference on Intelligent Virtual Agents

    Who’s next?: Integrating Non-Verbal Turn-Taking Cues for Embod- ied Conversational Agents. In Proceedings of the 23rd ACM International Con- ference on Intelligent Virtual Agents . ACM, Würzburg Germany, 1–8. https: //doi.org/10.1145/3570945.3607312

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.