{
  "schema": "vedokrok.public-item.v1",
  "release_id": "VEDOKROK-SEL-5d86e1b23401e578",
  "url": "/knowledge/do-not-score-every-overlap-as-an-interruption",
  "id": "MHC-D-RESEARCH-9337",
  "version": "0.1.0",
  "title": "Do not score every overlap as an interruption",
  "summary": "Two voices at once can be a collision. It can also be the listener saying, 'I am with you—keep going.'",
  "kind": "anti_pattern",
  "body": "Natural conversation contains cooperative overlap. Short backchannels such as 'mhm', 'yeah' or 'okay' often occur while the current speaker is still talking and can signal attention without taking the floor. Reviews of turn-taking therefore distinguish overlap from interruption, and naturalistic corpora show that backchannels are extremely common. A useful analyzer should first ask what the second voice did: support the current turn, complete it collaboratively, compete for the floor, or actually derail it. Do not let voice-activity overlap become a moral verdict on conversational behavior.\n\nA brief low-content listener response occurs under the speaker and the speaker continues smoothly.\nTreat it as possible supportive backchannel/cooperative overlap, not an interruption by default.\n\nThe second speaker launches propositional content, raises competition for the floor or causes the first speaker to stop/restart.\nInspect it as possible competitive overlap or interruption using context, not overlap duration alone.\n\nBackchannels are frequent but follow-up questions, validation, comprehension or task outcomes are poor.\nDo not call the conversation high-quality from backchannel frequency alone.",
  "limits": [
    "Classifying conversational intent automatically remains difficult. Backchannel forms, overlap norms and timing vary across languages, cultures, relationships, neurotypes and conversational settings. Do not use overlap or backchannel rate to rank politeness, personality, social skill or neurodevelopmental status. When the distinction cannot be made reliably, report uncertainty rather than forcing a binary label."
  ],
  "topics": [
    "communication"
  ],
  "intents": [],
  "source_ids": [
    "spoken-turn-taking-review-2020",
    "spoken-backchannel-fluenceme-2025",
    "spoken-candor-conversation-corpus-2023",
    "spoken-listening-connection-2025-detail"
  ],
  "evidence": [
    {
      "claim": "Current turn-taking research distinguishes cooperative overlap such as backchannels from competitive overlap and interruption; simultaneous speech alone is not sufficient to label an interruption.",
      "source_id": "spoken-turn-taking-review-2020",
      "role": "supports",
      "note": "Claim limit: The categories still require contextual interpretation, and acoustic features alone do not perfectly identify speaker intent.\nSource reading history: open_review_passages; 2026-09-30; Review section on overlaps, backchannels and interruptions. The review distinguishes cooperative overlap, competitive overlap and interruption, and explicitly warns that overlap is not equivalent to failed turn-taking or interruption.\nAgent release review: Agent checked the exact card fields and its claim, source, locator, example, check and stated boundary. This bounded projection preserves the source pack’s screened evidence scope and does not claim independent scientific or human editorial review.",
      "locator": "Section on overlaps, backchannels and interruptions"
    },
    {
      "claim": "In a 2025 face-to-face conversation study, 66% of observed listener backchannels occurred inside the current speaker's turn and therefore commonly overlapped the speaker without functioning as an attempt to take the floor.",
      "source_id": "spoken-backchannel-fluenceme-2025",
      "role": "supports",
      "note": "Claim limit: The exact proportion comes from one German conversational dataset and should not be treated as a universal benchmark.\nSource reading history: open_full_text; 2026-09-30; German face-to-face conversations; distribution of backchannels and fluencemes within/around turns, acoustic form and head movement. In this dataset most backchannels occurred turn-internally and therefore overlapped the current speaker. The study is small and language-specific, so the exact distribution should not be generalized as a universal rate.\nAgent release review: Agent checked the exact card fields and its claim, source, locator, example, check and stated boundary. This bounded projection preserves the source pack’s screened evidence scope and does not claim independent scientific or human editorial review.",
      "locator": "Distribution of backchannels by turn location"
    },
    {
      "claim": "In the large naturalistic CANDOR conversation corpus, listeners used backchannels in 33.7% of all speaker turns and 65.5% of turns at least five words long, illustrating that minimal listener feedback is a common feature of ordinary conversation.",
      "source_id": "spoken-candor-conversation-corpus-2023",
      "role": "contextualizes",
      "note": "Claim limit: Prevalence does not establish an optimal rate, and conversation quality cannot be inferred from backchannel count alone.\nSource reading history: open_full_text; 2026-09-30; Naturalistic video-conversation corpus; listener backchannel frequency and distribution. Backchannels were common across natural conversation, including in 65.5% of speaker turns of at least five words. Frequency demonstrates normality, not that more backchannels are always better.\nAgent release review: Agent checked the exact card fields and its claim, source, locator, example, check and stated boundary. This bounded projection preserves the source pack’s screened evidence scope and does not claim independent scientific or human editorial review.",
      "locator": "Backchannel frequency analysis"
    },
    {
      "claim": "A 2025 listening study operationalized richer verbal listening through relevant follow-up questions and verbal validation while excluding simple short backchannels such as 'yeah' or 'mhmm' from those verbal quality indicators.",
      "source_id": "spoken-listening-connection-2025-detail",
      "role": "limits",
      "note": "Claim limit: Backchannels can still be useful listener feedback; the distinction shows only that their presence alone is not sufficient evidence of high-quality listening.\nSource reading history: open_full_text; 2026-09-30; Behavioral coding of follow-up questions, verbal validation and global high-quality listening; simple short backchannels excluded from the verbal listening indicators. The study linked richer listening behaviors to connection measures, but deliberately did not count simple 'yeah/mhmm' backchannels as sufficient verbal indicators of high-quality listening. This limits any score that equates backchannel frequency with listening quality.\nAgent release review: Agent checked the exact card fields and its claim, source, locator, example, check and stated boundary. This bounded projection preserves the source pack’s screened evidence scope and does not claim independent scientific or human editorial review.",
      "locator": "Behavioral indicators of high-quality listening"
    }
  ],
  "use_when": [
    "A conversation recording or speech-analysis system flags simultaneous speech, backchannels or short listener responses as interruptions and treats a higher overlap count as worse communication."
  ],
  "avoid_when": [
    "Classifying conversational intent automatically remains difficult. Backchannel forms, overlap norms and timing vary across languages, cultures, relationships, neurotypes and conversational settings. Do not use overlap or backchannel rate to rank politeness, personality, social skill or neurodevelopmental status. When the distinction cannot be made reliably, report uncertainty rather than forcing a binary label."
  ],
  "recognition": "Natural conversation contains cooperative overlap. Short backchannels such as 'mhm', 'yeah' or 'okay' often occur while the current speaker is still talking and can signal attention without taking the floor. Reviews of turn-taking therefore distinguish overlap from interruption, and naturalistic corpora show that backchannels are extremely common. A useful analyzer should first ask what the second voice did: support the current turn, complete it collaboratively, compete for the floor, or actually derail it. Do not let voice-activity overlap become a moral verdict on conversational behavior.\n\nA brief low-content listener response occurs under the speaker and the speaker continues smoothly.\nTreat it as possible supportive backchannel/cooperative overlap, not an interruption by default.\n\nThe second speaker launches propositional content, raises competition for the floor or causes the first speaker to stop/restart.\nInspect it as possible competitive overlap or interruption using context, not overlap duration alone.\n\nBackchannels are frequent but follow-up questions, validation, comprehension or task outcomes are poor.\nDo not call the conversation high-quality from backchannel frequency alone.",
  "example": "During a technical explanation, the listener says 'yeah' softly halfway through a clause and the speaker continues without changing course. A VAD-only metric sees overlap; a conversation-aware metric should not automatically count that moment as rude interruption.",
  "check": "Overlap labels distinguish at least supportive listener feedback from attempts to take the floor, and any listening-quality score uses outcomes or richer listening behavior beyond raw backchannel frequency.",
  "published_at": "2026-11-13T05:54:13Z",
  "review": {
    "kind": "agent",
    "label": "Agent editorial and scoped source review",
    "reviewed_at": "2026-11-13T05:54:13Z"
  },
  "knowledge_origin": {
    "release_id": "MHC-RPUB-20261113-0d6697fd",
    "corpus_commit": "0d6697fd15f7fb670cfa465c54a411e91752cae4",
    "manifest_sha256": "038080ffd768d7c1ae21ae65ea2a3b8e36664c28df759e06d19f63f6e059bf29",
    "projection_sha256": "489520d20496b8e53d4bcf37ddaae13d1b66f4405e33ae1666ce681073b24c78",
    "published_at": "2026-11-13T05:54:13Z"
  },
  "editorial_overlay": null,
  "editorial": null,
  "notices": [],
  "sources": [
    {
      "id": "spoken-turn-taking-review-2020",
      "title": "Turn-taking in Conversational Systems and Human-Robot Interaction: A Review",
      "url": "https://doi.org/10.1016/j.csl.2020.101178"
    },
    {
      "id": "spoken-backchannel-fluenceme-2025",
      "title": "The speaker's “okay” vs. the listener's “okay”: exploring lexical, phonetic, and multimodal variation of backchannels and fluencemes in conversation",
      "url": "https://doi.org/10.3389/fcomm.2025.1655049"
    },
    {
      "id": "spoken-candor-conversation-corpus-2023",
      "title": "The CANDOR corpus: Insights from a large multimodal dataset of naturalistic conversation",
      "url": "https://doi.org/10.1126/sciadv.adf3197"
    },
    {
      "id": "spoken-listening-connection-2025-detail",
      "title": "High-quality listening behaviors linked to social connection between strangers",
      "url": "https://doi.org/10.1038/s44271-025-00342-2"
    }
  ],
  "relations": [
    {
      "from": "MHC-D-RESEARCH-9337",
      "to": "MHC-D-RESEARCH-9330",
      "type": "useful_with",
      "url": "/knowledge/pause-where-the-listener-can-join-the-thought"
    },
    {
      "from": "MHC-D-RESEARCH-9337",
      "to": "MHC-D-RESEARCH-1245",
      "type": "useful_with",
      "url": "/knowledge/treat-delivery-metrics-as-diagnostics-not-commandments"
    },
    {
      "from": "MHC-D-RESEARCH-9337",
      "to": "MHC-D-BRALI-0029",
      "type": "useful_with",
      "url": "/knowledge/check-what-you-heard-before-answering"
    }
  ],
  "collections": [
    {
      "id": "RC-MHC-UNION-SPOKEN-INTERACTION-TIMING-AND-GESTURE-2026",
      "title": "Make live speech easier to follow, answer and join",
      "url": "/collections/make-live-speech-easier-to-follow-answer-and-join"
    }
  ],
  "topic_hubs": [
    {
      "id": "communication",
      "title": "Communication and feedback",
      "url": "/topics/communication-and-feedback"
    }
  ],
  "solution_packs": []
}
