CoChat
Sign Up
Cited by 2Ethics and Social Impacts of AIDisability Rights and RepresentationDigital Accessibility for Disabilities

Cripping AI: Reimagining AI Through Lived Disability Experiences

Open access
Tang, Xinru, Lin, Ting-an, Li, Jingjin, Wu, Shaomei·2026·Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency
DOI 10.1145/3805689.3806744
Open this paper in CoChat — free account

Abstract

Drawing on crip theory, this paper proposes cripping AI as a guiding framework to center lived disability experiences in AI research and development. Moving beyond calls to make AI “accessible” to people with disabilities, cripping AI seeks to: (1) reveal and dismantle ableist assumptions embedded in how AI is imagined, designed, and evaluated; (2) center disabled ways of knowing (i.e., cripistemologies); (3) respect disabled labor in co-creating accessible practices. We demonstrate how to apply our framework with three cases: deafness and sign language AI, blindness and visual assistive AI, and stuttering and speech AI. We end by outlining three directions for future work, including cripping AI with diverse human bodyminds, across the entire AI pipeline and ecosystem, and in collaboration with other justice-oriented AI efforts.

Verified 2026-10-25 · 46 references not checked yet

AI snapshot

Written from the paper’s full text.

TL;DR

This paper proposes "cripping AI" as a guiding framework that moves beyond making AI merely accessible, instead centering lived disability experiences in how AI is imagined, designed, and evaluated. It develops three tenets—exposing disability politics, honoring cripistemologies, and respecting crip labor—and illustrates them through three case studies on sign language AI, visual assistive AI, and stuttering/speech AI.

Key findings

  • The framework rests on three interconnected tenets: (1) *expose disability politics*—interrogate how AI practices construct certain bodyminds as "disabled" rather than treating disabled/non-disabled as fixed; (2) *honor cripistemologies*—center knowledge produced through disabled bodyminds and communities rather than treating able-bodied perception as ground truth; (3) *respect crip labor*—recognize accessibility as ongoing, friction-filled, collective work in which disabled people play a key role.
  • The paper identifies three ableist assumptions recurring in accessibility-oriented AI: disability viewed as a medical deficit, able-bodied people positioned as the authoritative source of knowledge, and disabled people treated as passive objects of protection.
  • Cited evidence includes: WHO figures that around 1.3 billion people worldwide experience significant disability (~16% of the global population); a survey of 10k+ deaf signers in China finding fewer than 10% can fully understand sign language interpretation on TV broadcasting; and a finding that PWS annotators disagreed with labels in more than 25% of clips in a highly cited English stuttered speech dataset.
  • Three cross-case insights: toward lived-experience-preserving pipelines (preserving diversity, ambiguity, and subjectivity instead of a single ground truth), lived experience as expertise, and design with lived experience in the loop.

Methods

  • Conceptual/theoretical paper: it draws on crip theory and critical disability studies to propose a framework, offered explicitly as "a generative framework rather than a prescriptive method."
  • Three case studies drawn from the authors' own prior and ongoing research: deafness and sign language AI, blindness and visual assistive AI, and stuttering and speech AI (the latter conducted at AImpower.org with people who stutter).
  • No new empirical data collection is reported for this paper; the cases synthesize prior work (e.g., sign language content creation, everyday uncertainty in BLV GenAI use, stuttered speech dataset annotation and benchmarking) to illustrate applying the three tenets.

Limitations

  • Stated by the authors: they acknowledge "the risk of abstracting disabled experiences in our theorization and when we describe disabilities as a group"; that ableism remains a pervasive reality and not all disabled individuals want to be called "crip"; and that claiming a crip identity and access to disability justice can itself reflect a form of privilege, so the framework "must remain situated and context-sensitive in practice."
  • Stated scope limitation: "Given the scope limitation, we did not cover some crucial aspects such as model training, architecture, APIs, and governance."
  • Evident from the positionality statement: none of the authors identify as deaf or blind, none are signers, and only the first author has basic knowledge of sign language—so the deafness and blindness cases are not authored from lived experience of those disabilities (the last author has lived experience of stuttering).

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1145/3805689.3806744

Cites

217 of 290 references verified0 citation errors · 27 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

6 of 318 cited sources checked · 2 supported · 0 flagged · accuracy not yet measured

  • Premisecitation
    § Cripping AI · p. 5

    Recent clinical work increasingly advocates for transdiagnostic approaches that shift focus from diagnostic categories to individuals' needs and experiences.

    “Recent clinical work increasingly advocates for transdiagnostic approaches that shift the focus from diagnostic categories to individuals' needs and experiences [7, 9].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [7] [7] Ian A Apperly, Robert Lee, Sanne W van der Kleij, and Rory T Devine. 2024. A transdiagnostic approach to neurodiversity in a representative population sample: The N+ 4 model. JCPP advances 4, 2 (2024), e12219. https://doi.org/10.1002/jcv2.12219 - source

      The Introduction (“Neurodevelopmental conditions as complex phenotypes”; page unavailable) says that features outside diagnostic categories can be more important to people's experiences and needs, and describes arguments for approaches that better capture this complexity. It does not establish that clinical work increasingly advocates such a shift; it describes a research approach and says some authors favor combining new approaches with, rather than replacing, diagnostic categories. (LLM, conf=0.79)

      Experimental — accuracy not yet measured
    • cites [9]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Premisecitation
    § 2.1 Limitations of 'Making AI Accessible' · p. 3

    Many diagnostic frameworks lack consensus or explanatory validity.

    “many diagnositic frameworks lack consensus or explanatory validity [7, 9, 14, 196, 280].”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [7] [7] Ian A Apperly, Robert Lee, Sanne W van der Kleij, and Rory T Devine. 2024. A transdiagnostic approach to neurodiversity in a representative population sample: The N+ 4 model. JCPP advances 4, 2 (2024), e12219. https://doi.org/10.1002/jcv2.12219 - source

      The Introduction (‘Neurodevelopmental conditions as complex phenotypes’; page unavailable) says traditional diagnostic categories have prompted questions about their validity and may not capture important complexity. The Discussion (‘Assumptions, limitations, and further work’; page unavailable) describes contested boundaries between neurodevelopmental and psychiatric conditions. These passages support concerns about some diagnostic categories, but do not establish that many diagnostic frameworks lack explanatory validity as a matter of fact. (LLM, conf=0.82)

      Experimental — accuracy not yet measured
    • cites [9]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [14]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    In mixed-ability teams, members must deal with inaccessible tools and constantly negotiate between differing and sometimes conflicting needs.

    “in mixed-ability teams, where members must deal with inaccessible tools and constantly negotiate between differing and sometimes conflicting needs [5, 97, 101, 150, 248].”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites [5] [5] Rahaf Alharbi, John Tang, and Karl Henderson. 2023. Accessibility Barriers, Conflicts, and Repairs: Understanding the Experience of Professionals with Disabilities in Hybrid Meetings. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI '23) . Association for Computing Machinery, New York, NY, USA, Article 605, 15 pages. doi:10.1145/3544548. 3581541 - source
    • cites [97]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [101]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Findingcitation
    § Cripping AI · p. 13

    PWS-led practices allowed the authors to capture more versatile, naturalistic, and value-aligned stuttered speech data than datasets curated through expert-led paradigms.

    “PWS-led practices therefore allowed us to capture more versatile, naturalistic, and value-aligned stuttered speech data than datasets curated through expert-led paradigms [140, 144].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [140]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [144]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Findingcitation
    § 4.3 Stuttering and Speech AI · p. 12

    The authors' findings suggest that PWS speakers' expectations for annotations vary across communication contexts.

    “Our findings suggest that PWS speakers' expectations for annotations vary across communication contexts.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [246]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims. 60
  • Classified as probably not a substantial claim. 37

  • May combine two claims.

    Premisecitation
    § Cripping AI · p. 5

    These standards have limitations and have been repeatedly challenged by clinical professionals.

    “these standards have their limitations and have been repeatedly challenged by clinical professionals.”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [7] [7] Ian A Apperly, Robert Lee, Sanne W van der Kleij, and Rory T Devine. 2024. A transdiagnostic approach to neurodiversity in a representative population sample: The N+ 4 model. JCPP advances 4, 2 (2024), e12219. https://doi.org/10.1002/jcv2.12219 - source

      The Introduction (‘Neurodevelopmental conditions as complex phenotypes’; page unavailable) says categorical diagnoses have limitations and that some have questioned their validity. The Introduction (‘Transdiagnostic approaches to psychiatry and neurodevelopmental conditions’; page unavailable) also says researchers have questioned psychiatric diagnostic categories. Neither passage establishes that clinical professionals, specifically, have repeatedly challenged these standards. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
    • cites [9]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    In video conferencing, BLV members might prioritize visual privacy, while deaf members might prefer to have visual cues; both need to navigate error-prone captions.

    “in video conferencing, BLV members might prioritize visual privacy, while deaf members might prefer to have visual cues; both need to navigate error-prone captions [5, 150].”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [5] [5] Rahaf Alharbi, John Tang, and Karl Henderson. 2023. Accessibility Barriers, Conflicts, and Repairs: Understanding the Experience of Professionals with Disabilities in Hybrid Meetings. In Proceedings of the 2023 CHI Conference on Human Factors in Computing Systems (Hamburg, Germany) (CHI '23) . Association for Computing Machinery, New York, NY, USA, Article 605, 15 pages. doi:10.1145/3544548. 3581541 - source

      Sections 4.1.1 and 5.2.1 describe deaf and hard-of-hearing professionals needing visual cues to identify speakers and lipread. Section 4.1.2 reports inaccurate captions in hybrid meetings. The passages do not establish that blind or low-vision members prioritize visual privacy or that they, too, need to navigate caption errors. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites [150]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 10

    Both general-purpose and blind-focused AI systems often miss information crucial to BLV users or assume sighted knowledge when generating visual descriptions.

    “Research shows that both general-purpose and blind-focused AI systems often miss information crucial to BLV users (e.g., nutrition labels on products [66]) or assume sighted knowledge when generating visual descriptions [4].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [66]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [4] [4] Rahaf Alharbi, Angela D Cheong, Jaylin Herskovitz, Robin N. Brewer, and Sarita Schoenebeck. 2025. 'Trying to Piece It Together': Exploring Accessible Error Detection in Emerging Privacy Techniques With Blind People. In Proceedings of the 27th International ACM SIGACCESS Conference on Computers and Accessibility (ASSETS '25) . Association for Computing Machinery, New York, NY, USA, Article 18, 20 pages. doi:10.1145/3663547.3746376 - source

  • May combine two claims.

    Findingcitation
    § Cripping AI · p. 13

    While non-PWS crowdworkers often follow rigid rules or focus on observable acoustic cues, PWS annotators can capture the bodily sensations that arise during and around moments of stuttering.

    “While non-PWS crowdworkers often follow rigid rules or focus on observable acoustic cues, PWS annotators can capture the bodily sensations that arise during and around moments of stuttering [246].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [246]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Findingcitation
    § Cripping AI · p. 13

    This PWS-centered approach helped foster an intimate and supportive environment for PWS participants, and turned data collection into a rewarding community activity for personal growth.

    “This PWS-centered approach not only helped foster an intimate and supportive environment for PWS participants, but also turned data collection into a rewarding community activity for personal growth [143].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [143]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 1 Introduction · p. 2

    AI models exhibit both performance degradation and bias in disability-related tasks.

    “Growing evidence shows that AI models exhibit both performance degradation and bias in disability-related tasks [79, 256, 266].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [79]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [256]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [266]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 1 Introduction · p. 2

    There remains a lack of clear guidance on how to make AI truly supportive of disability.

    “there remains a lack of clear guidance on how to make AI truly supportive of disability.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 2.1 Limitations of 'Making AI Accessible' · p. 3

    Able-bodied people are often treated as the default reference for task goals or as the source of ground-truth labels.

    “Able-bodied people are often treated as the default reference for task goals or as the source of ground-truth labels.”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [65]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [81]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 2.2 Crip Theory and Cripping Tech · p. 3

    This line of work has revealed pervasive ableist assumptions embedded in technologies and made efforts to transform design to center disability.

    “This line of work has revealed pervasive ableist assumptions embedded in technologies, and made efforts to transform design to center disability [101, 105, 108].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [101]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [105]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [108]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 2.2 Crip Theory and Cripping Tech · p. 3

    How to truly center lived disability experiences in AI development remains exploratory.

    “how to truly center lived disability experiences in AI development remains exploratory.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Interpretationreasoning
    § 1 Introduction · p. 2

    These technologies might still reinforce the ableist hierarchy and treat PWDs as passive objects of protection.

    “these technologies might still reinforce the ableist hierarchy and treat PWDs as passive objects of protection.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § 2.1 Limitations of 'Making AI Accessible' · p. 3

    These systems often rely on visual descriptions provided by sighted people for training while overlooking BLV people's sense-making strategies.

    “these systems often rely on visual descriptions provided by sighted people for training [80, 81], while overlooking BLV people's sense-making strategies [66, 244].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [81]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [66]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning
    § 2.1 Limitations of 'Making AI Accessible' · p. 3

    Without careful consideration, resulting AI models and systems may not only fall short in benefiting disabled communities but also reinforce ableism.

    “Without careful consideration, the resulting AI models and systems may not only fall short in benefiting disabled communities but also reinforce ableism.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § 3.1 Expose disability politics: interrogating how power relations are encoded and reinforced in AI practices · p. 4

    Crip theory rejects essentialist views of disability, arguing that disabled people are 'disabled' through interactions with their surrounding environments.

    “rejects essentialist views of disability, arguing that disabled people are 'disabled' through interactions with their surrounding environments”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [117]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [122]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [145]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 3.1 Expose disability politics: interrogating how power relations are encoded and reinforced in AI practices · p. 4

    Diagnostic frameworks such as DSM are often deemed the authoritative standard for understanding mental or neurodevelopmental disabilities.

    “diagnostic frameworks such as DSM are often deemed as the authoritative standard for understanding mental or neurodevelopmental disabilities [41, 167, 234].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [41]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [167]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [234]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning
    § 2.1 Limitations of 'Making AI Accessible' · p. 3

    Without critical examination and careful design, these systems may reinforce flawed medical models and perpetuate ableist worldviews.

    “Without critical examination and careful design, these systems may reinforce flawed medical models and perpetuate ableist worldviews.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 3.2 Honor cripistemologies: centering disabled ways of knowing in AI development · p. 6

    Many navigation systems define the optimal path as the widest open space, but objects on roads often serve as important informational cues for BLV people, making the widest open space a worse environment for them.

    “many navigation systems are designed for obstacle avoidance and define the optimal path as the widest open space. However, objects on roads often serve as important informational cues for BLV people [267], making the widest open space a worse environment for them.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [267]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S1 (asserted).

    Premisecitation
    § Cripping AI · p. 5

    Critics argue that LLMs like ChatGPT prioritize ableist norms such as efficiency or fluency in speech and writing.

    “critics argue that LLMs, like ChatGPT, prioritize ableist norms such as efficiency [105] or fluency in speech and writing [165].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [105]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [165]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.2 Honor cripistemologies: centering disabled ways of knowing in AI development · p. 6

    BLV people must navigate incomplete and subjective information from both AI models and humans, often developing their unique sense-making strategies.

    “BLV people must navigate incomplete and subjective information from both AI models and humans in their everyday information seeking, often developing their unique sense-making strategies [244].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [244]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.2 Honor cripistemologies: centering disabled ways of knowing in AI development · p. 6

    Traditional machine learning tasks often assume a static, single ground truth, such as each image has a single accurate description, each face expresses a specific emotion, or each English sentence has a direct parallel in sign language.

    “These traditional machine learning tasks often assume a static, single ground truth, for example, that each image has a single 'accurate' description [80], each face expresses a specific type of emotion [46], or that each English sentence has a direct parallel in sign language [55].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [46]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [55]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § Cripping AI · p. 5

    Scholars in critical disability studies challenge assumptions about an 'ideal' human standard and position disability as an intrinsic part of humanity.

    “scholars in critical disability studies challenge assumptions about an 'ideal' human standard and position disability as an intrinsic part of humanity [117, 122, 162].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [117]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [122]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [162]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § Cripping AI · p. 5

    The neurodiversity movement challenges traditional biological and psychiatric standards of autism, arguing that it should be recognized as a form of neural diversity.

    “the neurodiversity movement, which challenges traditional biological and psychiatric standards of autism, arguing that it should be recognized as a form of neural diversity [230].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [230]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § Cripping AI · p. 5

    These metrics and evaluation tasks may not truly measure 'general' intelligence but instead privilege a limited set of social and cultural norms about intelligence and abilities.

    “these metrics and evaluation tasks may not truly measure 'general' intelligence but instead privilege a limited set of social and cultural norms about what counts as intelligence and abilities”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [119]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [153]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [180]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    Creating accessibility is ongoing, friction-filled work in which disabled people actively participate, often resulting in additional labor and without an ideal solution.

    “Extensive evidence shows that creating accessibility is ongoing, friction-filled work in which disabled people actively participate [15, 19, 26, 202, 244], often resulting in additional labor [27, 228] and without an ideal solution [202, 244, 248].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [15]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [19]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [26]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.2 Honor cripistemologies: centering disabled ways of knowing in AI development · p. 6

    When a BLV person navigates a space, they actively gather environmental cues and shift cognitive frameworks, triangulating between ego- and allocentric spatial frameworks, interpreting tactile feedback, using a white cane, and collecting interaction cues from a sighted guide.

    “when a BLV person navigates a space, they do not rely on a single source of information, but actively gather environmental cues and shift in cognitive frameworks [212], such as triangulating between ego- (e.g., 'to my right') and allocentric (e.g., 'east of the park') spatial frameworks [68], interpreting tactile feedback from the ground [267], using a white cane to detect environmental features [”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [212]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [68]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [267]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § Cripping AI · p. 5

    The pursuit of Artificial General Intelligence is often guided by a narrowed conception of intelligence and evaluated using standardized benchmarks or comparison to human baselines.

    “the pursuit of Artificial General Intelligence (AGI) is often guided by a narrowed conception of intelligence and often evaluated using standardized benchmarks or by comparison to 'human baselines' [30, 194].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [30]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [194]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    Although dominant narratives often frame AI as 'empowering' disabled people, these systems can also introduce new accessibility issues.

    “Although dominant narratives often frame AI as 'empowering' disabled people, these systems can also introduce new accessibility issues [2, 47, 103].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [2] [2] Rudaiba Adnin and Maitraye Das. 2024. "I look at it as the king of knowledge": How Blind People Use and Understand Generative AI Tools. In Proceedings of the 26th International ACM SIGACCESS Conference on Computers and Accessibility (St. John's, NL, Canada) (ASSETS '24) . Association for Computing Machinery, New York, NY, USA, Article 64, 14 pages. doi:10.1145/3663548.3675631 - source
    • cites [47]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [103]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    Sign language systems often rely on annotation formats that reduce signs to simple tokens and do not follow best practices in linguistics.

    “a recent review found that sign language systems often rely on annotation formats that reduce signs to simple tokens and do not follow best practices in linguistics [56].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [56]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    Back-up channels like human assistance are necessary for people to receive continuous care and to build trust in a service when AI systems are limited.

    “back-up channels like human assistance are necessary for people to receive continuous care and to build trust in a service when AI systems are limited [171].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [171]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    Sign language AI typically encompasses techniques for sign language recognition, generation, and bidirectional translation with written or spoken language, often taking translation as the ultimate goal.

    “It typically encompasses techniques for sign language recognition, generation, and bidirectional translation with written or spoken language, often taking translation as the ultimate goal [24].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [24]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    Under the traditional view of accessibility as a one-off technical problem, evaluation of AI systems might focus on output accuracy or task completion, and disabled people are usually limited to providing data or testing the final system.

    “Under this traditional view, the evaluation of AI systems might focuses on output accuracy or task completion [39, 80, 81, 283], and disabled people are usually limited to providing data or testing the final system [50].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [39]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [81]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 4.1 Deafness and Sign Language AI · p. 8

    Sign language AI is a core technology to serve deaf communities.

    “From a cultural perspective, sign language AI is therefore a core technology to serve deaf communities.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    BLV people's understanding of information often evolves in their interactions with humans or tools; they constantly interpret and evaluate AI outputs and the subjective nature of information while dealing with inaccessible websites and artifacts.

    “BLV people's understanding of information often evolves in their interactions with human or tools. They constantly interpret and evaluate, not only AI outputs, but also the subjective nature of information (e.g., interpreting subjective questions like those about fashion, or the source of the information), while constantly dealing with inaccessible websites and artifacts [20, 244].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [20]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [244]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationcitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 8

    Overreliance on personalized methods risks shifting responsibility and labor onto disabled users themselves and diverting attention from more fundamental interventions.

    “overreliance on them risks shifting responsibility and labor onto disabled users themselves [243], and diverting attention from more fundamental interventions.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [243]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 3.3 Respect crip labor: recognizing AI accessibility as an ongoing, friction-filled, collective work · p. 7

    Remote visual assistance platforms connect BLV users with sighted volunteers to answer visual questions, while later becoming a major source of training AI systems for BLV people.

    “These platforms connect BLV users with sighted volunteers to answer visual questions [80, 81], while later becoming a major source of training AI systems for BLV people.”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [81]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    Although deafness has traditionally been viewed as a 'lack' of hearing, a cultural perspective views deaf people who use sign language as a linguistic and cultural minority.

    “Although deafness has traditionally been viewed as a 'lack' of hearing, a cultural perspective views deaf people who use sign language as a linguistic and cultural minority [191].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [191]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisenone given
    § 4.1 Deafness and Sign Language AI · p. 8

    Signing Exact English represents English through letter-by-letter fingerspelling.

    “Signing Exact English, represents English through letter-by-letter fingerspelling”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    Historically, hearing-centered beliefs have assumed spoken language as the only legitimate form of language, dismissed sign languages as a poor substitute, and pressured deaf children to use lip-reading and speech.

    “Historically, hearing-centered beliefs have assumed spoken language as the only legitimate form of language, dismissed sign languages as a poor substitute, and pressured deaf children to use lip-reading and speech [86].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [86]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisereasoning
    § 4.1 Deafness and Sign Language AI · p. 8

    Deaf communities have heated debates over what counts as a 'legitimate' sign language and how to sign 'deaf' enough.

    “deaf communities have heated debates over what counts as a 'legitimate' sign language and how to sign 'deaf' enough.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 9

    As sign language information is often lacking in public spaces, deaf people have cultivated vibrant sign language spaces, e.g., deaf clubs and online communities.

    “As sign language information is often lacking in public spaces [191, 245, 255], deaf people have cultivated vibrant sign language spaces, e.g., deaf clubs [93, 191] and online communities [245, 249].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [191]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [245]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [255]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    If every regional and urban variety of sign language was given a distinct name, Indonesia alone would have over 500 named sign languages.

    “Kusters et al. noted that if every regional and urban variety was given a distinct name, Indonesia alone would have over 500 named sign languages [131].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [131]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    It was not until the 1960s that sign languages were recognized as fully developed languages, when William Stokoe provided formal linguistic evidence demonstrating that ASL has its own distinct structures and vocabularies.

    “It was not until the 1960s that sign languages were recognized as fully developed languages, when William Stokoe provided formal linguistic evidence demonstrating that ASL has its own distinct structures and vocabularies [240].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [240]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    Frequent language contact between sign language and written/spoken language systems led to varieties such as Pidgin Signed English.

    “the frequent language contact between sign language and written/spoken language systems then led to varieties such as Pidgin Signed English [89].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [89]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    Sign languages often lack standardized forms and are shaped by complex histories and language ideologies.

    “As minority languages, sign languages often lack standardized forms and are shaped by complex histories and language ideologies [131, 179].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [131]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [179]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing). May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 8

    Manually coded languages are often criticized as ableist and might even impact deaf children's language development.

    “they are often criticized as ableist [179] and might even impact deaf children's language development [220].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [179]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [220]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 9

    A survey of 10k+ deaf signers in China found that fewer than 10% can fully understand sign language interpretation on TV broadcasting, as hearing interpreters usually follow Chinese grammar when signing.

    “A survey of 10k+ deaf signers in China found that fewer than 10% can fully understand sign language interpretation on TV broadcasting, as hearing interpreters usually follow Chinese grammar when signing”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [146]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 11

    Clark's interpreter was not professional and had strong personal opinions, but his descriptions were so direct and informative that Clark could interpret for himself what was happening in the room through the interpreter's perspective.

    “As he recalls, the interpreter was not professional and had strong personal opinions, but his descriptions were so direct and informative that Clark could interpret for himself what was happening in the room through the interpreter's perspective [113].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [113]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 9

    There is rich evidence that deaf people co-construct and negotiate meaning with interpreters in communication, as well as with hearing or other deaf participants.

    “There is rich evidence showing how deaf people co-construct and negotiate meaning with interpreters in communication, as well as with hearing or other deaf participants [51, 113, 164, 261].”
    Quote in the paperSubstantial claim0 of 4 sources checked
    • cites [51]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [113]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [164]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [261]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 10

    Tang et al. describe how BLV people must navigate the inherent subjectivity in visual information from both AI systems and humans, a phenomenon they term 'everyday uncertainty'.

    “Tang et al. describe how BLV people must navigate the inherent subjectivity in visual information from both AI systems and humans, a phenomenon they term 'everyday uncertainty' [244].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [244]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationcitation
    § 4.2 Blindness and Visual Assistive AI · p. 10

    The pervasive subjectivity in visual perception questions the dominant paradigm that relies on sighted crowdworkers as the ground truth, especially considering that most sighted crowdworkers do not know what information BLV people need.

    “This pervasive subjectivity in visual perception questions the dominant paradigm that relies on sighted crowdworkers as the ground truth [80], especially considering that most sighted crowdworkers do not know what information BLV people need [229].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [229]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • Rests on an assumption: Visual assistive systems typically use images provided by BLV individuals or sourced from the Internet for training, while relying on sighted crowdworkers to generate corresponding visual descriptions.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § 4.1 Deafness and Sign Language AI · p. 10

    De Meulder noted many deaf people have negotiated with low-quality interpretation services for their whole life, and may see avatars signing as another signing style they will have to put up with and learn to understand.

    “As deaf scholar De Meulder noted, many deaf people have negotiated with low-quality interpretation services for their whole life, and they may see 'avatars' signing as another signing style they'll have to put up with and learn to 'understand'' [50].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [50]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 11

    Emerging AI tools offer potential support but also introduce new forms of labor, such as managing erroneous outputs and prompt engineering.

    “From the lens of labor, while emerging AI tools offer potential support, they also introduce new forms of labor, such as managing erroneous outputs and prompt engineering [244].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [244]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given
    § 4.2 Blindness and Visual Assistive AI · p. 10

    Comparatively little attention has been given to the underlying assumptions in visual assistive AI efforts.

    “However, comparatively little attention has been given to the underlying assumptions in these efforts.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 10

    Garg et al. conducted a survey to identify BLV users' information needs regarding products and re-annotated visual accessibility datasets to reflect these needs.

    “As an example of recent efforts to incorporate BLV perspectives in model development, Garg et al. conducted a survey to identify BLV users' information needs regarding products and re-annotated visual accessibility datasets to reflect these needs [66].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [66]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 10

    The definition and operationalization of blindness differ across legal, medical, and functional frameworks.

    “Similar to deafness, the definition and operationalization of blindness differ across legal, medical (e.g., visual acuity), and functional frameworks (e.g., what assistive technologies a person uses) [68].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [68]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Interpretationreasoning
    § 4.1 Deafness and Sign Language AI · p. 9

    Translanguaging practices broaden the conventional scope of a named language to the whole semiotic space.

    “These translanguaging practices not only challenge the idea of building translation machines for sign language, but also broaden the conventional scope of a named language to the whole semiotic space.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 11

    Clark critiqued the 'accurate' alt-texts he typically encountered as 'a replica, divorced entirely from the original'.

    “By contrast, he critiqued the 'accurate' alt-texts he typically encountered as 'a replica, divorced entirely from the original' [113].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [113]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning
    § 4.2 Blindness and Visual Assistive AI · p. 10

    These embodied forms of knowledge are often lacking in existing AI models and limit their abilities to represent blindness-related items and meet BLV people's needs.

    “However, these embodied forms of knowledge are often lacking in existing AI models and limit their abilities to represent blindness-related items and meet BLV people's needs.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 10

    Visual assistive systems share the goal to provide visual information to BLV people, for example, describing scenes, generating image captions, answering visual questions about images, or identifying objects.

    “While varying in formats and use contexts, these systems share the goal to provide visual information to BLV people, for example, describing scenes [65], generating image captions [81], answering visual questions about images [80], or identifying objects [116].”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [65]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [81]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 10

    For much of history, blindness has been defined in terms of a lack of sight, while vision has been positioned as the authoritative source of knowledge.

    “For much of history, blindness has been defined in terms of a lack of sight, while vision has been positioned as the authoritative source of knowledge [216].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [216]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 11

    John Lee Clark illustrated this paradox with a story of working with a 'biased' yet 'best' interpreter.

    “John Lee Clark, a Deafblind activist, illustrated this paradox with a story of working with a 'biased' yet 'best' interpreter [113].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [113]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 11

    The assumption of a single objective visual description may work for some tasks but often fails in real-world situations where perceived utility of information varies according to purpose of information seeking, temporal constraints, and criticality.

    “Although this assumption may work for some tasks (e.g., asking for a product brand when the product photo is clear and complete), it often fails in real-world situations that BLV users encounter, where the perceived utility of information varies according to the purpose of information seeking (curiosity vs. focused search), temporal constraints, and criticality (for healthcare vs. fun), etc. [244]”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [244]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.3 Stuttering and Speech AI · p. 12

    Recent research and advocacy in stuttering have increasingly affirmed stuttering as a valid mode of communication, defined not by speech disfluencies but by the speaker's subjective experiences, such as spontaneity and embodied sensations.

    “Recent research and advocacy in stuttering, however, have increasingly affirm stuttering as a valid mode of communication [32], defined not by speech disfluencies but by the speaker's subjective experiences, such as spontaneity [43] and embodied sensations [254].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [32]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [43]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [254]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 4.3 Stuttering and Speech AI · p. 11

    Speech AI systems often cut off or misunderstand stuttered speech, further marginalizing PWS in every aspect of daily life including employment and social interactions.

    “They often cut off or misunderstood stuttered speech, further marginalizing PWS in every aspects of daily life including employment and social interactions [135, 144, 238, 277].”
    Quote in the paperSubstantial claim0 of 4 sources checked
    • cites [135]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [144]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [238]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [277]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • Rests on an assumption: Speech AI systems are optimized for fluency and speed.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 4.2 Blindness and Visual Assistive AI · p. 11

    These frictions highlight accessibility as a collective work which needs a unified policy framework that accommodates accessibility with other values, such as privacy and the interests of bystanders.

    “These frictions highlight accessibility as a collective work which needs a unified policy framework that accommodates accessibility with other values, such as privacy [243] and the interests of bystanders [3].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [243]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [3] [3] Taslima Akter, Tousif Ahmed, Apu Kapadia, and Manohar Swaminathan. 2022. Shared Privacy Concerns of the Visually Impaired and Sighted Bystanders with Camera-Based Assistive Technologies. ACM Trans. Access. Comput. 15, 2, Article 11 (May 2022), 33 pages. doi:10.1145/3506857 - source

  • May combine two claims.

    Premisecitation
    § 4.2 Blindness and Visual Assistive AI · p. 11

    Much research has focused on improving BLV people's photo-taking skills, as they often take 'low-quality' images considered hard for AI models to process, i.e., blurry, poorly framed, or missing key objects.

    “For instance, much research has focused on improving BLV people's photo-taking skills [116, 173, 222, 252], as they often take 'low-quality' images that are considered hard for AI models to process, i.e., blurry, poorly framed, or missing key objects [80, 81].”
    Quote in the paperSubstantial claim0 of 6 sources checked
    • cites [116]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [173]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [222]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [252]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [80]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [81]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning
    § 4.3 Stuttering and Speech AI · p. 12

    These concealment behaviors can blur the boundaries of stuttered and fluent speech, as a speaker can sound fluent by changing every word they block on and a speaker can appear stutter as they pause intentionally.

    “These behaviors can blur the boundaries of stuttered and fluent speech, as a speaker can sound fluent by changing every word they block on and a speaker can appear stutter as they pause intentionally.”
    Quote in the paperSubstantial claim

  • Classified as probably not a substantial claim.

    Premisecitation
    § 4.3 Stuttering and Speech AI · p. 12

    The Stuttering Severity Instrument (SSI) requires speech-language pathologists to annotate the occurrence and a clear boundary of each stuttering event in stuttered speech samples.

    “The Stuttering Severity Instrument (SSI), one of the most popular clinical tools for measuring stuttering severity, requires speech-language pathologists (SLPs) to annotate the occurrence and a clear boundary of each stuttering event in stuttered speech samples [49].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [49]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 4.3 Stuttering and Speech AI · p. 11

    Stuttering is a neuro-developmental condition characterized by involuntary disruptions in speech, such as sound repetitions, word repetitions, sound prolongations, and blocks.

    “Stuttering is a neuro-developmental condition characterized by involuntary disruptions in speech, such as sound repetitions, word repetitions, sound prolongations, and blocks [182].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [182]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation
    § 4.3 Stuttering and Speech AI · p. 12

    Christopher Constantino observed that pauses, hesitations, and silences carry semantic weight and are meaningful and purposeful.

    “As Christopher Constantino (PWS, licensed SLP) observes, ' Our pauses, hesitations, and silences carry semantic weight; they are meaningful and purposeful ' [42].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [42]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § Cripping AI · p. 13

    For many PWS, confronting their own stuttering can trigger intense emotional responses, imposing a significant emotional burden when they are asked to annotate or evaluate their own speech.

    “For many PWS, confronting their own stuttering can trigger intense emotional responses [274], imposing a significant emotional burden when they are asked to annotate or evaluate their own speech.”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [274]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationcitation
    § Cripping AI · p. 13

    Collecting speech data in the form of unscripted conversations between PWS grants the speakers full agency to engage in topics that are meaningful to them, fostering narrative sovereignty for a population whose voices are often silenced in society.

    “Collecting speech data in the form of unscripted conversations between PWS also grants the speakers full agency to engage in topics that are meaningful to them, fostering narrative sovereignty for a population whose voices are often silenced in society [140].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [140]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 4.4 Summary of Insights · p. 14

    These forms of embodied knowledge are crucial for moving beyond medical and able-bodied perspectives, building systems that truly support disabled people.

    “These forms of embodied knowledge are crucial for moving beyond medical and able-bodied perspectives, building systems that truly support disabled people.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation
    § Cripping AI · p. 13

    The authors' prior dataset work invited PWS community members or allies to take on interviewer roles and guide the data collection process.

    “As an exploratory effort, our prior dataset work invited PWS community members or allies to take on interviewer roles and guide the data collection process [143].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [143]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given
    § Cripping AI · p. 13

    The underperformance of ASR models could create additional labor for PWS such as fixing inaccurate transcriptions.

    “For example, the underperformance of ASR models could create additional labor for PWS such as fixing inaccurate transcriptions.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation
    § 5.1 Toward a More Expansive and Plural Understanding of Human Bodyminds · p. 14

    Mental health communities have cultivated Mad culture to promote patient-centered care and to frame fluctuations in emotions as part of their identity and natural human experiences.

    “In contrast, mental health communities have cultivated Mad culture to promote patient-centered care and to frame fluctuations in emotions as part of their identity and natural human experiences [37, 151, 205].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [37]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [151]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [205]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § Cripping AI · p. 13

    High variability within stuttering could lead to different types of model failures for different speakers: those who speak with more repetitions may see more hallucination in auto captioning, while those with more blocks may experience frequent cut-offs by voice assistants.

    “However, the high variability within stuttering [254] could lead to different types of model failures for different speakers: those who speak with more repetitions may see more hallucination in auto captioning [238], while those with more blocks may experience frequent cut-offs by voice assistants [135].”
    Quote in the paperSubstantial claim0 of 3 sources checked
    • cites [254]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [238]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [135]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable).

    Interpretationreasoning
    § 4.4 Summary of Insights · p. 14

    Lived disability experiences are critical expertise for AI development.

    “Lived disability experiences are critical expertise for AI development.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S4 (based).

    Interpretationnone given
    § 5.1 Toward a More Expansive and Plural Understanding of Human Bodyminds · p. 14

    These perspectives have yet to be incorporated into AI development.

    “These perspectives, however, have yet to be incorporated into AI development.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductcitation
    § 4.3 Stuttering and Speech AI · p. 12

    When the authors annotated stuttered speech datasets for ASR, they co-designed the annotation guidelines with PWS and invited PWS speakers to evaluate the quality of annotations on their own speech in concrete downstream applications.

    “For example, when we annotated stuttered speech datasets for ASR, we co-designed the annotation guidelines with PWS and invited PWS speakers to evaluate the quality of annotations on their own speech in concrete downstream applications [246].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [246]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Premisecitation
    § Cripping AI · p. 13

    The iceberg theory of stuttering suggests that stuttering is like an iceberg, with only a small part above the waterline and a much bigger part below.

    “The iceberg theory of stuttering suggests that [223], 'stuttering is like an iceberg, with only a small part above the waterline and a much bigger part below.'”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [223]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning
    § Cripping AI · p. 13

    These nuanced perspectives from PWS challenge the accuracy of existing stuttered speech annotations produced by non-PWS, and call into question common practices in speech model training and evaluation, such as trimming segments classified as 'non-speech' or 'silence', or using synthetic speech data to represent stuttering.

    “These nuanced perspectives from PWS challenge the accuracy of existing stuttered speech annotations produced by non-PWS, and also call into question some common practices in speech model training and evaluation, such as trimming segments classified as 'non-speech' or 'silence' (e.g., [184, 287]), or using synthetic speech data to represent stuttering [130, 284].”
    Quote in the paperSubstantial claim0 of 4 sources checked
    • cites [184]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [287]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [130]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [284]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. May combine two claims.

    Premisecitation
    § 6 Concluding Thoughts · p. 15

    Claiming a crip identity and access to disability justice can itself reflect a form of privilege and simplifies the histories and powers behind the word.

    “Claiming a crip identity and access to disability justice can itself reflect a form of privilege [123, 160, 226] and simplifies the histories and powers behind the word [225, 226].”
    Quote in the paperSubstantial claim0 of 4 sources checked
    • cites [123]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [160]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [226]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [225]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 5.3 Bridging Justice-Oriented AI Efforts · p. 15

    ASL was further influenced by racial segregation and gendered norms, leading to variations like Black ASL and gendered expressions.

    “ASL was further influenced by racial segregation and gendered norms, leading to variations like Black ASL [95] and gendered expressions (e.g., 'father' is typically signed higher on the face than 'mother') [257].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [95]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

    • cites [257]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing).

    Interpretationreasoning
    § 5.2 Cripping the Entire AI Pipeline and Ecosystem · p. 15

    These tensions are likely to persist in the design of AI systems.

    “These tensions are likely to persist in the design of AI systems.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation
    § 6 Concluding Thoughts · p. 15

    Ableism remains a pervasive reality for disabled people.

    “Additionally, we acknowledge that ableism remains a pervasive reality for disabled people [250], and that not all disabled individuals want to be called 'crip' [226].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [250]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning
    § 5.3 Bridging Justice-Oriented AI Efforts · p. 15

    Ableism not only affects disabled people but also sets narrow standards of 'ability' that reinforces ideologies that disadvantage women, queer people, racial minorities, people in the Global South, etc.

    “Second, a cross-community approach is also essential because ableism not only affects disabled people but also sets narrow standards of 'ability' that reinforces ideologies that disadvantage women, queer people, racial minorities, people in the Global South, etc.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Premisecitation
    § 5.3 Bridging Justice-Oriented AI Efforts · p. 15

    ASL is colonialized, racialized, and gendered.

    “For example, ASL is colonialized, racialized, and gendered [91].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [91]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Premisecitation
    § 5.3 Bridging Justice-Oriented AI Efforts · p. 15

    Deaf schools in the U.S. emerged from European settler colonial projects, which explains French Sign Language's influence over ASL.

    “Deaf schools in the U.S. emerged from European settler colonial projects [91], which explains French Sign Language's influence over ASL.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [91]

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

89 of 89

  • The paper relies on this without support.

    Centering disabled people's lived experiences is necessary for AI to be truly supportive of disability.

    “To truly center disability in AI development, this paper takes one step further from making AI accessible toward cripping AI as a guiding framework.”

  • The paper relies on this without support.

    The three proposed tenets are a sufficient starting framework for systematically cripping AI.

    “We propose three interconnected tenets as a starting point for systematically cripping AI”

  • The paper relies on this without support.

    The authors' three case studies adequately demonstrate how to apply the cripping AI framework.

    “We present three cases from our own research to exemplify how to put our framework in practice.”

  • The paper relies on this without support.

    Sign language technologies aim to fully convert sign languages into other languages.

    “The richness of sign language practices also questions the goal of sign language technologies to fully convert them into other languages.”

  • The paper relies on this without support.

    Speech AI systems are optimized for fluency and speed.

    “Optimized for fluency and speed, speech AI systems including voice commands, speech-to-text tools, and automatic captioning often have great difficulties when interacting with PWS.”

  • The paper relies on this without support.

    Visual assistive systems typically use images provided by BLV individuals or sourced from the Internet for training, while relying on sighted crowdworkers to generate corresponding visual descriptions.

    “These systems typically use images provided by BLV individuals or sourced from the Internet for training, while relying on sighted crowdworkers to generate the corresponding visual descriptions.”
6 of 6

Cites

Flagged first, then by citations

72 of the 288 references couldn’t be matched to a known paper.

Cited by

Sorted by citations

Authors
Tang, X., Lin, T.-A., Li, J., & Wu, S.
Affiliations
  • University of California, Irvine (US)
  • University of Connecticut (US)
  • AImpower.org, Mountain View, California, USA
ORCID
Venue
Proceedings of the 2026 ACM Conference on Fairness, Accountability, and Transparency
Publisher
ACM
Published
2026-06-25
Pages
3508-3536
DOI
10.1145/3805689.3806744
OpenAlex
W7160416972
Semantic Scholar
c897aa298cf511f77dce71dd8a594810c6efdaa2
Cited by
2
References
288
Citations by year
2026: 2
FWCI
19.68
Type
conference paper
Open access
Licence
Locations
  • ACM Conference on Fairness, Accountability, and Transparency (FAccT) (publishedVersion, cc-by)
  • arXiv (Cornell University) (submittedVersion, cc-by)
  • arXiv (Cornell University) (cc-by)
Funders
NSF (National Science Foundation) (2427710)
Topics
  • Social Sciences › Safety Research › Ethics and Social Impacts of AI
  • Social Sciences › Safety Research › Disability Rights and Representation
  • Social Sciences › Human Factors and Ergonomics › Digital Accessibility for Disabilities
Keywords
crip theory, disability studies, ableism, cripistemology

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-25.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-25.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review