<?xml version="1.0" encoding="UTF-8"?>
<modsCollection xmlns="http://www.loc.gov/mods/v3">
<mods ID="kuan-etal-2026-model">
    <titleInfo>
        <title>Model Interruption Bench: Evaluating Real-Time Error Correction in Full-Duplex Voice Agents</title>
    </titleInfo>
    <name type="personal">
        <namePart type="given">Shih-Yun</namePart>
        <namePart type="given">Shan</namePart>
        <namePart type="family">Kuan</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Cheng-Han</namePart>
        <namePart type="family">Chiang</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Guan-Ting</namePart>
        <namePart type="family">Lin</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <name type="personal">
        <namePart type="given">Hung-yi</namePart>
        <namePart type="family">Lee</namePart>
        <role>
            <roleTerm authority="marcrelator" type="text">author</roleTerm>
        </role>
    </name>
    <originInfo>
        <dateIssued>2026-10</dateIssued>
    </originInfo>
    <typeOfResource>text</typeOfResource>
    <relatedItem type="host">
        <titleInfo>
            <title>Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing</title>
        </titleInfo>
        <name type="personal">
            <namePart type="given">Hung-Yi</namePart>
            <namePart type="family">Lee</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Sunipa</namePart>
            <namePart type="family">Dev</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Jessy</namePart>
            <namePart type="family">Li</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <name type="personal">
            <namePart type="given">Desmond</namePart>
            <namePart type="family">Elliott</namePart>
            <role>
                <roleTerm authority="marcrelator" type="text">editor</roleTerm>
            </role>
        </name>
        <originInfo>
            <publisher>Association for Computational Linguistics</publisher>
            <place>
                <placeTerm type="text">Budapest, Hungary</placeTerm>
            </place>
        </originInfo>
        <genre authority="marcgt">conference publication</genre>
        <identifier type="isbn">979-8-89176-459-0</identifier>
    </relatedItem>
    <abstract>Full-duplex voice agents can now listen and speak at the same time, yet no benchmark asks whether they can further exploit this capability—proactively interrupting a user mid-utterance to correct a content error. We introduce the Model Interruption Bench (MIB), the first benchmark for real-time, content-aware error correction in voice agents. MIB frames the task as proactive correction of spoken chain-of-thought math: ~600 step-annotated solutions derived from GSM8K, played to the model under a streaming protocol that scores whether it interrupts in the right window with the right correction. To localize failure, we pair the full-duplex streaming evaluation with turn-based ablations that isolate error detection, and provide an automated LLM-judge pipeline with backchannel filtering and a fine-grained FP/FN sub-type taxonomy to analyze failure cases. Evaluating six voice agents, Freeze-Omni, Gemini 2.5/3.1 Flash Live, GPT-4o Realtime, Moshi, and PersonaPlex, we uncover a striking detection–interruption gap: the three most detection-capable models recognize 59–67% of errors in the turn-based condition but reach below 10% mid-speech TP% when forced to interrupt in real time. The bottleneck is not understanding errors but acting on them in time. Three behavioral archetypes emerge from the sub-type breakdown—passive, post-hoc responder, and aggressive—exposing real-time error interruption as a concrete open challenge for full-duplex voice agents.</abstract>
    <identifier type="citekey">kuan-etal-2026-model</identifier>
    <location>
        <url>https://aclanthology.org/2026.emnlp-main.2349/</url>
    </location>
    <part>
        <date>2026-10</date>
        <extent unit="page">
            <start>53341</start>
            <end>53356</end>
        </extent>
    </part>
</mods>
</modsCollection>

%0 Conference Proceedings
%T Model Interruption Bench: Evaluating Real-Time Error Correction in Full-Duplex Voice Agents
%A Kuan, Shih-Yun Shan
%A Chiang, Cheng-Han
%A Lin, Guan-Ting
%A Lee, Hung-yi
%Y Lee, Hung-Yi
%Y Dev, Sunipa
%Y Li, Jessy
%Y Elliott, Desmond
%S Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing
%D 2026
%8 October
%I Association for Computational Linguistics
%C Budapest, Hungary
%@ 979-8-89176-459-0
%F kuan-etal-2026-model
%X Full-duplex voice agents can now listen and speak at the same time, yet no benchmark asks whether they can further exploit this capability—proactively interrupting a user mid-utterance to correct a content error. We introduce the Model Interruption Bench (MIB), the first benchmark for real-time, content-aware error correction in voice agents. MIB frames the task as proactive correction of spoken chain-of-thought math: ~600 step-annotated solutions derived from GSM8K, played to the model under a streaming protocol that scores whether it interrupts in the right window with the right correction. To localize failure, we pair the full-duplex streaming evaluation with turn-based ablations that isolate error detection, and provide an automated LLM-judge pipeline with backchannel filtering and a fine-grained FP/FN sub-type taxonomy to analyze failure cases. Evaluating six voice agents, Freeze-Omni, Gemini 2.5/3.1 Flash Live, GPT-4o Realtime, Moshi, and PersonaPlex, we uncover a striking detection–interruption gap: the three most detection-capable models recognize 59–67% of errors in the turn-based condition but reach below 10% mid-speech TP% when forced to interrupt in real time. The bottleneck is not understanding errors but acting on them in time. Three behavioral archetypes emerge from the sub-type breakdown—passive, post-hoc responder, and aggressive—exposing real-time error interruption as a concrete open challenge for full-duplex voice agents.
%U https://aclanthology.org/2026.emnlp-main.2349/
%P 53341-53356

Markdown (Informal)

[Model Interruption Bench: Evaluating Real-Time Error Correction in Full-Duplex Voice Agents](https://aclanthology.org/2026.emnlp-main.2349/) (Kuan et al., EMNLP 2026)

• Model Interruption Bench: Evaluating Real-Time Error Correction in Full-Duplex Voice Agents (Kuan et al., EMNLP 2026)

ACL

• Shih-Yun Shan Kuan, Cheng-Han Chiang, Guan-Ting Lin, and Hung-yi Lee. 2026. Model Interruption Bench: Evaluating Real-Time Error Correction in Full-Duplex Voice Agents. In Proceedings of the 2026 Conference on Empirical Methods in Natural Language Processing, pages 53341–53356, Budapest, Hungary. Association for Computational Linguistics.