CoChat
Sign Up
Cited by 59Social Robot Interaction and HRISpeech and dialogue systemsDementia and Cognitive Impairment Research

A Conversational Robot for Older Adults with Alzheimer's Disease

Open access
Pou-Prom, Chloé, Raimondo, Stefania, Rudzicz, Frank·2020·ACM Transactions on Human-Robot Interaction
DOI 10.1145/3380785
Open this paper in CoChat — free accountOpen PDF

Abstract

Amid the rising cost of Alzheimer’s disease (AD), assistive health technologies can reduce care-giving burden by aiding in assessment, monitoring, and therapy. This article presents a pilot study testing the feasibility and effect of a conversational robot in a cognitive assessment task with older adults with AD. We examine the robot interactions through dialogue and miscommunication analysis, linguistic feature analysis, and the use of a qualitative analysis, in which we report key themes that were prevalent throughout the study. While conversations were typically better with human conversation partners (being longer, with greater engagement and less misunderstanding), we found that the robot was generally well liked by participants and that it was able to capture their interest in dialogue. Miscommunication due to issues of understanding and intelligibility did not seem to deter participants from their experience. Furthermore, in automatically extracting linguistic features, we examine how non-acoustic aspects of language change across participants with varying degrees of cognitive impairment, highlighting the robot’s potential as a monitoring tool. This pilot study is an exploration of how conversational robots can be used to support individuals with AD.

AI snapshot

Written from the paper’s full text.

TL;DR

This pilot tested Ludwig, a conversational robot, with older adults with Alzheimer's disease in picture-description dialogues across human, Wizard-of-Oz robot, and autonomous robot sessions. Human partners produced richer conversations, but the robot was often liked and could hold attention, and transcript features varied with cognitive impairment.

Key findings

  • In 19 participants (3 male) with Alzheimer's disease and MMSE scores from 2 to 19, participants spoke less with the robot: average utterances per image dropped from 21 with a human partner to 13 with the Wizard-of-Oz robot to 4 with the autonomous robot, and vocabulary also declined in Session 3.
  • Human conversations were longer, more engaging, and less misunderstood; the robot was generally well liked and captured interest, and misunderstanding or intelligibility problems did not appear to deter participants from the experience.
  • The autonomous robot's speech recognition had a 49.72% word error rate in Session 3, with 76% of errors being deletions; robot repairs were often unrelated or unsuccessful, while the human partner had more successful repairs.
  • Fifteen linguistic features differed across MMSE groups, and none showed an Image Order effect. Word length and mean sentence length varied by MMSE but not by session, suggesting possible monitoring markers; severe-versus-mild and severe-versus-moderate differences were most prominent.

Methods

  • Pilot study with 19 nursing-home residents diagnosed with Alzheimer's disease. Three sessions over about a year: human conversation partner, Wizard-of-Oz robot, and autonomous robot. Each session involved semi-structured picture description of three previously unseen images, for nine pictures total. The robot was a Milo R25 named Ludwig; a Kinect tracked facial animation units, and questionnaires were administered.
  • Conversations were transcribed and annotated by a speech-language pathologist for dialogue acts and trouble-source-repair variables, with inter-rater reliability of Fleiss' kappa 0.72. N-way ANOVAs used Session, MMSE, Participant ID, and Image Order as covariates, with post hoc Bonferroni or Tukey tests.
  • Linguistic features were automatically extracted from transcripts, including lexical, syntactic, and semantic measures. Acoustic features were excluded because utterance-level audio segmentations were unavailable.

Limitations

  • Stated: Alzheimer's symptoms vary day to day, and MMSE scores cannot capture good days versus bad days; more conversations would help. The Wizard-of-Oz condition did not perfectly replicate human conversation because the operator filtered fragments and stalls and faced extra cognitive burden. The autonomous robot was limited by speech recognition quality, rule-based inflexible dialogue, no dialogue context, ignored out-of-vocabulary words, keyword matching, no facial or gestural or prosodic cues, and synthetic voice clarity. The Wizard-of-Oz condition always preceded the autonomous condition, which may have biased expectations. The linguistic analysis was restricted to transcript features; trouble-source-repair variables were manually annotated and acoustic features were excluded. The small sample limits generalization, and some participants with very low MMSE had trouble interacting with the device.
  • Evident from the text: session completion was uneven—Table 8 indicates no mild-MMSE participants completed Session 3—so session comparisons by impairment level are unbalanced. Also, the MMSE group ranges in the methods text differ from the labels in Table 7, which complicates interpretation of group-based results.

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

Integrity signals

Registry

Crossref · 10.1145/3380785

Cites

45 of 48 references verified0 citation errors · 3 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

7 of 48 cited sources checked · 3 supported · 0 flagged · accuracy not yet measured

  • Premisecitation

    Acoustic, syntactic, and semantic features, including grammatical complexity and information content, can correlate strongly with cognitive decline.

    “Acoustic, syntactic, and semantic features, including grammatical complexity and information content, can correlate strongly with cognitive decline [11,32].”
    Quote in the paperSubstantial claim1 of 2 sources checked
    • cites [11] 11. Kathleen C. Fraser, Jed A. Meltzer, and Frank Rudzicz. 2015. Linguistic features identify Alzheimer's disease in nar- rative speech. J. Alzheim. Dis. 49, 2 (2015), 407-422. DOI:https://doi.org/10.3233/JAD-150520 source
    • cites [32] 32. Brian Roark, Margaret Mitchell, John-Paul Hosom, Kristy Hollingshead, and Jeffrey Kaye. 2011. Spoken language derived measures for detecting mild cognitive impairment. IEEE Trans. Audio Speech Lang. Process. 19, 7 (Sep. 2011), 2081-2090. DOI:https://doi.org/10.1109/TASL.2011.2112351 source

      The Experimental Results (section IV.D; page not provided) report significant differences between healthy elderly subjects and those with mild cognitive impairment for some speech, syntactic, and content-density measures. But idea density did not differ significantly, and the study compares groups rather than measuring correlations with cognitive decline over time. The claim overgeneralizes both the measures and the outcome. (LLM, conf=0.84)

      Experimental — accuracy not yet measured
  • Premisecitation

    People tend to be forgiving of robots that cannot guarantee optimal performance, so long as the robot clearly acknowledges its own limitations.

    “Fortunately, people tend to be forgiving of robots that cannot guarantee optimal performance, so long as the robot clearly acknowledges its own limitations [40].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [40] 40. Rik van den Brule, Gijsbert Bijlstra, Ron Dotsch, Pim Haselager, and Daniel H. J. Wigboldus. 2016. Warning signals for poor performance improve human-robot interaction. J. Hum.-Robot Interact. 5, 2 (2016), 69-89. source
  • Findingown data

    The robot was able to capture participants' interest in dialogue.

    “While conversations were typically better with human conversation partners (being longer, with greater engagement and less misunderstanding), we found that the robot was generally well liked by participants and that it was able to capture their interest in dialogue.”
    Quote in the paperSubstantial claim
  • Findingown data

    The robot was generally well liked by participants.

    “While conversations were typically better with human conversation partners (being longer, with greater engagement and less misunderstanding), we found that the robot was generally well liked by participants and that it was able to capture their interest in dialogue.”
    Quote in the paperNot a substantial claim
  • Findingown data

    Miscommunication due to issues of understanding and intelligibility did not seem to deter participants from their experience.

    “Miscommunication due to issues of understanding and intelligibility did not seem to deter participants from their experience.”
    Quote in the paperSubstantial claim
  • Classified as probably not a substantial claim. 40
  • May combine two claims. 46

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

    Premisecitation

    Other language tasks, such as naming, verbal fluency, and story recall, have also been used for automatic monitoring of AD.

    “Other language tasks, such as the naming task [20], the verbal fluency task [18], and the story recall task [32], have also been used for automatic monitoring of AD.”
    Quote in the paperNot a substantial claim1 of 3 sources checked
    • cites [20] 20. H. S. Kirshner, W. G. Webb, and M. P. Kelly. 1984. The naming disorder of dementia.Neuropsychologia 22, 1 (1984), 23-30. source
    • cites [18] 18. Julie D. Henry, John R. Crawford, and Louise H. Phillips. 2004. Verbal fluency performance in dementia of the Alzheimer's type: A meta-analysis. Neuropsychologia 42, 9 (Jan. 2004), 1212-1222. DOI:https://doi.org/10.1016/j. neuropsychologia.2004.02.001 source
    • cites [32] 32. Brian Roark, Margaret Mitchell, John-Paul Hosom, Kristy Hollingshead, and Jeffrey Kaye. 2011. Spoken language derived measures for detecting mild cognitive impairment. IEEE Trans. Audio Speech Lang. Process. 19, 7 (Sep. 2011), 2081-2090. DOI:https://doi.org/10.1109/TASL.2011.2112351 source

      The Abstract and §II.B describe automatically derived measures from a story-recall task, supporting the portion of the sentence attached to this citation. But the study tests detection of mild cognitive impairment, not automatic monitoring of diagnosed Alzheimer’s disease; naming and verbal-fluency tasks have separate citations in the citing sentence. The indexed passages provide section labels but no page numbers. (LLM, conf=0.88)

      Experimental — accuracy not yet measured

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

    Premisecitation

    Robotic companions introduced to communities of people with dementia imitated many benefits of animal therapy without the burden or cost of maintaining live animals.

    “In early work, robotic companions were introduced to communities of people with dementia and imitated many of the benefits of animal therapy but without the burden or cost of maintaining live animals [28,36].”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites [28] 28. Wendy Moyle, Marie Cooke, Elizabeth Beattie, Cindy Jones, Barbara Klein, Glenda Cook, and Chrystal Gray. 2013. Exploring the effect of companion robots on emotional expression in older adults with dementia: A pilot randomized controlled trial. J. Gerontol. Nurs. 39, 5 (May 2013), 46-53. DOI:https://doi.org/10.3928/00989134-20130313-03 source
    • cites [36] 36. Takanori Shibata and Kazuyoshi Wada. 2011. Robot therapy: A new approach for mental healthcare of the elderly - a mini-review. Gerontology 57, 4 (2011), 378-86. DOI:https://doi.org/10.1159/000319015 source

      The “Interactive Autonomous Robots” section says robots were used as substitutes for animals in animal-assisted therapy, reports similar effects, and includes dementia patients among those served. The Introduction explains that institutions may avoid live animals because of allergies, infections, bites, and scratches. Neither passage supports the specific claim that robots avoided the burden or cost of maintaining live animals. (LLM, conf=0.86)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Premisecitation

    Analysis of spontaneous connected speech offers a glimpse into semantic processing and syntactic complexity, and can be used to identify the deterioration of communication skills.

    “Analysis of spontaneous connected speech offers a glimpse into semantic processing and syntactic complexity, and can be used to identify the deterioration of communication skills [29].”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites [29] 29. Kimberly D. Mueller, Bruce Hermann, Jonilda Mecollari, and Lyn S. Turkstra. 2018. Connected speech and language in mild cognitive impairment and alzheimer's disease: A review of picture description tasks. J. Clin. Exp. Neuropsychol. 40, 9 (2018), 1-23. DOI:https://doi.org/10.1080/13803395.2018.1446513 source

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

    Study design and conductcitation

    The linguistic analysis for cognitive monitoring followed Fraser et al. [11] and Roark et al. [32] and was restricted to features automatically extractable from transcriptions.

    “Our linguistic analysis for cognitive monitoring has followed the work of Fraser et al. [11] and Roark et al. [32] and is restricted to examining features that can be automatically extracted from transcriptions of conversations.”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Fraser et al. [11] 11. Kathleen C. Fraser, Jed A. Meltzer, and Frank Rudzicz. 2015. Linguistic features identify Alzheimer's disease in nar- rative speech. J. Alzheim. Dis. 49, 2 (2015), 407-422. DOI:https://doi.org/10.3233/JAD-150520 source
    • cites Roark et al. [32] 32. Brian Roark, Margaret Mitchell, John-Paul Hosom, Kristy Hollingshead, and Jeffrey Kaye. 2011. Spoken language derived measures for detecting mild cognitive impairment. IEEE Trans. Audio Speech Lang. Process. 19, 7 (Sep. 2011), 2081-2090. DOI:https://doi.org/10.1109/TASL.2011.2112351 source

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

    Premisecitation

    Some recent work has used reinforcement learning to learn policies that can at least limit instances of misunderstanding.

    “Some recent work has used reinforcement learning to learn policies that can at least limit those instances [5].”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites [5] 5. Hamidreza Chinaei, Leila Chan Currie, Andrew Danks, Hubert Lin, Tejas Mehta, and Frank Rudzicz. 2017. Identifying and avoiding confusion in dialogue with people with alzheimer's disease. Comput. Ling. 43, 2 (Mar. 2017), 1-42. DOI:https://doi.org/10.1162/COLI_a_00290 source

      Section 6 describes a POMDP dialogue policy designed to avoid trouble-indicating behaviors associated with confusion, and §6.3 reports better simulated performance than baseline policies. But §6.1 and §6.3 say the policy was optimized by point-based value iteration with a manually set reward function, not learned through reinforcement learning; inverse reinforcement learning is described as future work. The copy provides simulated results, not evidence that the policy reduced misunderstandings in live dialogue. (LLM, conf=0.87)

      Experimental — accuracy not yet measured

  • May combine two claims.

    Findingown data

    Conversations with human conversation partners were typically better than conversations with the robot, being longer, with greater engagement and less misunderstanding.

    “While conversations were typically better with human conversation partners (being longer, with greater engagement and less misunderstanding), we found that the robot was generally well liked by participants and that it was able to capture their interest in dialogue.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Participants produced more Trouble Sources overall than their healthy human partners, with a larger proportion of turns having Discourse or Semantic Trouble Sources.

    “Participants produced more Trouble Sources overall than their healthy human partners, with a larger proportion of turns having Discourse or Semantic Trouble Sources.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In post hoc analysis, word error rates were 49.72% for the third session, 76% of which was due to deletion errors, across all participants.

    “In post hoc analysis, we find word error rates at 49.72% for the third session, 76% of which was due to deletion errors, across all participants.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Overall, participants were less talkative with the robot, with participant utterances declining from Sessions 1 to 3, from an average of 21 to 13 to 4 per image.

    “Overall, participants were less talkative with the robot, with participant utterances declining from Sessions 1 to 3, from an average of 21 to 13 to 4 per image and their vocabulary also declining in Session 3.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    There are significant differences across all categories of annotated variables and across conversations with different participants.

    “There are significant differences across all categories of annotated variables and across conversations with different participants, suggesting that it is important to adapt to individual participants rather than generalize.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Though participants did not use more Repair Initiators overall than their partners, a higher proportion of the initiators they did use were of type L2-Non-specific terms.

    “Though the participants do not use more Repair Initiators overall than their partners, a higher proportion of the initiators they did use were of type L2-Non-specific terms.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    P11 did not want to participate in the picture description task in Session 1 and did not answer questions in either of the robot sessions.

    “For example, P11 did not want to participate in the picture description task in Session 1 (e.g., H:”
    • Also stated in: § 4.3.4 Eliciting Reactions. · plain
      “Can you look at the photo down here?,”
    • Also stated in: § 4.3.4 Eliciting Reactions. · plain
      “P11:”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Participants like P2 (medium-mild) and P16 (mild) were talkative and correctly answered questions.

    “Participants like P2 (medium-mild) and P16 (mild) were talkative and correctly answered questions.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    The human conversation partner and the remotely controlled robot were able to respond to out-of-domain vocabulary and flawed syntax, could sense whether a conversation should continue or end, and could interpret affect from facial expressions.

    “The human conversation partner and the remotely controlled robot were able to respond to outof-domain vocabulary and flawed syntax, could sense whether a conversation should continue or end, and could interpret affect from facial expressions.”
    Quote in the paperSubstantial claim

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

    Findingown data

    During the third session, P13 mentioned that she would have enjoyed the robot more if he moved, while P15 remarked that it 'didn't look at me' when asked if she thought he understood what she was saying.

    “During the third session, P13 mentioned that she would have enjoyed the robot more if he moved, while P15 remarked that it”
    • Also stated in: § 4.3.3 Intelligence of the Robot. · plain
      “didn't look at me”
    • Also stated in: § 4.3.3 Intelligence of the Robot. · plain
      “when asked if she thought he understood what she was saying.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The greatest differences in features are between the Severe and Mild groups and the Severe and Moderate group.

    “We find that the greatest differences in features are between the Severe and Mild groups and the Severe and Moderate group.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingnone given

    Overall, we observed mixed reactions to the robot, although many participants reflected positively on their interactions.

    “Overall, we observe mixed reactions to the robot, although many participants reflected positively on their interactions.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    P16 and P14 struggled to hear the robot, with P14's conversations suffering in Session 3 due to the autonomous robot's ability to reword utterances that needed to be repeated.

    “P16 and P14 also struggled to hear the robot, with P14's conversations suffering in Session 3 due to the autonomous robot's ability to reword utterances that needed to be repeated”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Conversations were smoother and richer with the human conversational partner: they were longer with a larger vocabulary, and participants were more 'open' and engaged more with the human participant and engaged more in the repair process.

    “In general, conversations were smoother and richer with the human conversational partner: Conversations were longer with a larger vocabulary, and participants were more”
    • Also stated in: § 4.3.1 Likeability of the Robot. · plain
      “open”
    • Also stated in: § 5.2 Human and Robot · plain
      “and engaged more with the human participant and engaged more in the repair process.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Both human and the Wizard-of-Oz robot were able to perform repairs.

    “Both human and the Wizard-of-Oz robot were able to perform repairs, but the human (i.e., in Session 1) had more instances of successful repairs.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    P9 and P13 struggled to understand the robot and would often turn to the researchers in the room whenever they did not understand the robot.

    “P9 and P13 struggled to understand the robot and would often turn to the researchers in the room whenever they did not understand the robot”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    P4, P8, and P15 fell asleep during most of their sessions and were barely responsive.

    “For example, P4, P8, and P15 fell asleep during most of their sessions and were barely responsive.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Reactions to the robot's likeability and participants' eagerness to interact with it were mixed.

    “We observed mixed reactions with regards to robot likeability and eagerness to interact with the robot.”
    • Also stated in: § 4.3.6 Summary. · plain
      “(1) Likeability of the robot was variable.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    Despite some instances of misunderstanding, P12 claimed to enjoy her interactions with the robot and was looking forward to speaking with him again.

    “However, despite some instances of misunderstanding, P12 claimed to enjoy her interactions with the robot and was looking forward to speaking with him again.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    P9 did not want to interact with her human partner and ended Session 1 early; however, the robot was able to capture her attention and she stayed for the entire duration with Ludwig.

    “P9 (moderate MMSE) did not want to interact with her human partner and ended Session 1 early; however, the robot was able to capture her attention and she stayed for the entire duration with Ludwig.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Some participants were skeptical about using a robot and were noticeably less talkative in Sessions 2 and 3.

    “Some participants were skeptical about using a robot and were noticeably less talkative in Sessions 2 and 3.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    P10's restless banging on the table significantly deteriorated the quality of speech recognition, yet she said talking to Ludwig was 'fun.'

    “P10's restless banging on the table significantly deteriorated the quality of speech recognition, yet she said talking to Ludwig was”
    • Also stated in: § 4.3.3 Intelligence of the Robot. · plain
      “fun.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    P16 was skeptical about the robot, calling him 'just another toy' and did not wish to participate in Session 3, despite being receptive to the picture description task earlier.

    “P16 (mild MMSE) was skeptical about the robot, calling him”
    • Also stated in: § 4.3.1 Likeability of the Robot. · plain
      “just another toy”
    • Also stated in: § 4.3.1 Likeability of the Robot. · plain
      “and did not wish to participate in Session 3, despite being receptive to the picture description task earlier.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    P17 and P13 noted that the robot was 'not fast enough,' which was especially a problem when long participant utterances slowed ASR.

    “P17 and P13 noted that the robot was was”
    • Also stated in: § 4.3.3 Intelligence of the Robot. · plain
      “not fast enough,”
    • Also stated in: § 4.3.3 Intelligence of the Robot. · plain
      “which was especially a problem when long participant utterances slowed ASR.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Some features vary across MMSE, with differences between the Mild and Severe group being the most prominent.

    “We found that some features vary across MMSE, with differences between the Mild and Severe group being the most prominent.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    P2 indicated eagerness to meet and interact with the robot in the pre-session questionnaire but was quiet and unresponsive during robot sessions, only occasionally looking up at Ludwig.

    “P2 (severe MMSE) indicated that she was eager to meet and interact with the robot during her presesion questionnaire.However, during sessions with the robot, she was quiet and unresponsive, only occasionally looking up at Ludwig.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    P3 was receptive to open-ended and personal questions in the human-human session but did not elaborate on answers to personal questions in Sessions 2 and 3, despite displaying no difficulties interacting with the robot.

    “P3 (moderate-mild MMSE) was receptive to both open-ended and personal questions in the human-human session”
    • Also stated in: § 4.3.1 Likeability of the Robot. · plain
      “but did not elaborate on answers to personal questions in Sessions 2 and 3, despite displaying no difficulties interacting with the robot.”
    Quote in the paperNot a substantial claim

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

    Findingnone given

    P13 spoke in English with the robot, while she only responded in her native tongue to the human partner.

    “Interestingly, P13 spoke in English with the robot, while she only responded in her native tongue to the human partner.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    P13 noted that the robot asked 'reasonable questions' and P14 enjoyed her interactions with the robot.

    “P13 noted that the robot asked”
    • Also stated in: § 4.3.2 Understanding and Intelligibility. · plain
      “reasonable questions”
    • Also stated in: § 4.3.2 Understanding and Intelligibility. · plain
      “and P14 enjoyed her interactions with the robot”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The autonomous version of the robot was not as flexible as its remotely controlled counterpart; it did not remember previous utterances or repeat or re-word misunderstood utterances.

    “Evidently, the autonomous version of the robot was not as flexible as its remotely controlled counterpart.It did not remember previous utterances or repeat or re-word misunderstood utterances.”
    Quote in the paperSubstantial claim
    • Rests on an assumption: The WoZ and autonomous robot sessions are comparable enough to attribute differences in flexibility to the robot's autonomy.

  • May combine two claims.

    Findingown data

    Despite difficulties, P13 and P15 acknowledged that the robot seemed friendly and intelligent, and P17 remarked that the robot was 'good' and 'soothing.'

    “Despite these difficulties, P13 and P15 acknowledged that the robot seemed friendly and intelligent, and P17 remarked that the robot was”
    • Also stated in: § 4.3.3 Intelligence of the Robot. · plain
      “good”
    • Also stated in: § 4.3.1 Likeability of the Robot. · plain
      “and”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    The proportion of PP phrases with respect to the total number of grammatical constituents is significant (differentiates between MMSE groups).

    “We find that the proportion of PP phrases with respect to the total number of grammatical constituents is also significant.”
    Quote in the paperSubstantial claim

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

    Findingown data

    P13 reported in the post-session questionnaire that she 'sometimes didn't understand' Ludwig.

    “P13 later noted in the post-session questionnaire that she”
    • Also stated in: § 4.3.2 Understanding and Intelligibility. · plain
      “sometimes didn't understand”
    • Also stated in: § 4.3.1 Likeability of the Robot. · plain
      “Ludwig.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    Some features did not differ across Session but were influenced by MMSE (e.g., mean length of sentence).

    “Some features did not differ across Session but were influenced by MMSE (e.g., mean length of sentence).”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    P12 and P15 were initially nervous about meeting the robot but enjoyed their sessions with Ludwig, expressing gratitude at the end of the conversations.

    “In contrast, P12 (moderate MMSE) and P15 (severe MMSE) were initially nervous about meeting the robot but enjoyed their sessions with Ludwig, expressing gratitude at the end of the conversations”
    Quote in the paperNot a substantial claim

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

    Findingown data

    P8 spoke a few words in Session 3.

    “Surprisingly, P8 in fact spoke a few words in Session 3.”
    Quote in the paperNot a substantial claim

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

    Findingown data

    MMSE has a significant effect on age-of-acquisition nouns (F=38.9, p<0.001).

    “Age-of-acquisition - | 8.64 (3.19e-4) *** 38.9 (1.24e-13) *** | 5.30 (9.60e-9) *** | 0.561 (0.572)”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Diagnosing dementia is typically done through clinician-administered assessments such as the MMSE, which is primarily used as an initial screening tool.

    “Diagnosing dementia is typically done through assessments administered by a clinician, such as the Mini-Mental State Examination (MMSE) [9], which is nevertheless primarily used as an initial screening tool.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] 9. Marshal F. Folstein, Susan E. Folstein, and Paul R. McHugh. 1975. "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician. J. Psychiatr. Res. 12, 3 (1975), 189-198. source

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

    Interpretationreasoning

    Participants with severely low MMSE (i.e., ≤10) are less aware of their lack of understanding.

    “Coupled with a decrease in repair initiators, specifically with Listener L2 (Non-specific terms), this might suggest that participants with severely low MMSE (i.e., ≤10) are less aware of their lack of understanding.”
    Quote in the paperNot a substantial claim

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

    Study design and conductown data

    Acoustic features were not included in the current analysis because they would require segmentations of the audio at a per utterance level, which are unavailable.

    “Note that we do not include acoustic features in the current analysis, since they would require segmentations of the audio at a per utterance level, which are unavailable.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    ASR systems have performed poorly for individuals with dementia, and older adults generally, since they are typically designed for younger, healthy adults.

    “So far, ASR systems have performed poorly for individuals with dementia, and older adults generally, since they are typically designed for younger, healthy adults [47].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [47] 47. Luke Zhou, Kathleen C. Fraser, and Frank Rudzicz. 2016. Speech recognition in alzheimer's disease and in its assess- ment. In Proceedings of the 17th Annual Meeting of the International Speech Communication Association (Interspeech'16). 1948-1952. DOI:https://doi.org/10.21437/Interspeech.2016-1228 source

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

    Interpretationnone given

    Live-in robots may provide more temporally fine-grained longitudinal analysis of speech and behaviour than is currently feasible for human health-care professionals.

    “Live-in robots may also provide more temporally fine-grained longitudinal analysis of speech and behaviour than is currently feasible for human health-care professionals.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    Previous work has successfully assessed dementia based on automatic analysis of speech and text and has correlated the level of cognitive decline with speech impairment.

    “Previous work has successfully assessed dementia based on automatic analysis of speech and text and has correlated the level of cognitive decline with speech impairment [45].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [45] 45. Maria Yancheva, Kathleen C. Fraser, and Frank Rudzicz. 2015. Using linguistic features longitudinally to predict clin- ical scores for alzheimer's disease and related dementias. In Proceedings of the 6th Workshop on Speech and Language Processing for Assistive Technologies (SLPAT'15). 134. DOI:https://doi.org/10.1016/j.bandl.2008.08.002 source

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

    Premisecitation

    According to Lai et al., increased 'Assess' may indicate increased conversation satisfaction and cohesiveness.

    “According to Lai et al. [22], this may indicate increased conversation satisfaction and cohesiveness.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [22] 22. Catherine Lai, Jean Carletta, and Steve Renals. 2013. Modelling participant affect in meetings with turn-taking fea- tures. In Proceedings of the Workshop on Affective Social Speech Signals (WASSS'13).

      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

    Participants were more interested in interacting with artificial conversational agents that produced social signals, such as smiles and nods.

    “For example, Sakai et al. found that participants were more interested in interacting with artificial conversational agents that produced social signals, such as smiles and nods [35].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [35] 35. Yoichi Sakai, Yuuko Nonaka, Kiyoshi Yasuda, and Yukiko I. Nakano. 2012. Listener agent for elderly people with dementia. In Proceedings of the 7th Annual ACM/IEEE International Conference on Human-Robot Interaction (HRI'12). ACM Press, New York, NY, 199. DOI:https://doi.org/10.1145/2157689.2157754 source

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

    Premisecitation

    The picture description task is used to elicit connected speech that allows for the detection of subtle changes in cognitive decline, even in early stages of AD.

    “Often, the picture description task is used to elicit connected speech that then allows for the detection of subtle changes in cognitive decline, even in early stages of AD [10,13].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [10] 10. Katrina E. Forbes-McKay and Annalena Venneri. 2005. Detecting subtle spontaneous language decline in early alzheimer's disease with a picture description task. Neurol. Sci. 26, 4 (Oct. 2005), 243-254. DOI:https://doi.org/10. 1007/s10072-005-0467-9 source
    • cites [13] 13. Elaine Giles, Karalyn Patterson, and John R. Hodges. 1996. Performance on the boston cookie theft picture description task in patients with early dementia of the alzheimer's type: Missing information. Aphasiology 10, 4 (May 1996), 395- 408. DOI:https://doi.org/10.1080/02687039608248419 source

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

    Study design and conductcitation

    To mitigate the effect of mood and speaker variability, the researchers verified with nurses and personnel that participants were in an amenable condition before each session.

    “To mitigate the effect of mood and speaker variability [44], we verified with nurses and personnel that participants were in an amenable condition before each session.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [44] 44. Bengt Winblad, Philippe Amouyel, Sandrine Andrieu, Clive Ballard, Carol Brayne, Henry Brodaty, Angel Cedazo- Minguez, Bruno Dubois, David Edvardsson, Howard Feldman, Laura Fratiglioni, Giovanni B. Frisoni, Serge Gauthier, Jean Georges, Caroline Graff, Khalid Iqbal, Frank Jessen, Gunilla Johansson, Linus Jönsson, Miia Kivipelto, Martin Knapp, Francesca Mangialasche, René Melis, Agneta Nordberg, Marcel Olde Rikkert, Chengxuan Qiu, Thomas P. Sakmar, Philip Scheltens, Lon S. Schneider, Reisa Sperling, Lars O. Tjernberg, Gunhild Waldemar, Anders Wimo, and Henrik Zetterberg. 2016. Defeating alzh source

  • May combine two claims.

    Premisecitation

    Conversations with people with AD are far more likely to involve communication breakdown, which can be quantified through the trouble-source repair (TSR) framework.

    “Conversations with people with AD are far more likely to involve communication breakdown, which can be quantified through the trouble-source repair (TSR) framework [30].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [30] 30. Joseph B. Orange, Rosemary B. Lubinski, and D. Jeffery Higginbotham. 1996. Conversational repair by individuals with dementia of the alzheimer's type. J. Speech Hear. Res. 39, 4 (Aug. 1996), 881-95. source

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

    Interpretationreasoning

    The increased use of Stalls by the healthy human partner may relate to the relative low verbal rate of the participants.

    “Stalls were used significantly more by the healthy human partner, which is unexpected but may relate to the relative low verbal rate of the participants.”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    Social robots have been used to provide personalized and affordable speech and cognitive therapy.

    “For example, social robots have been used to provide personalized and affordable speech and cognitive therapy [39] and to engage individuals in social interactions, critical in the maintenance of good cognitive health [24,27].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [39] 39. Adriana Tapus, Juan Fasola, and Maja J. Matarić. 2008. Cognitive assistance and physical therapy for dementia pa- tients. In ICRA Workshop on Social Interaction with Intelligent Indoor Robots.

      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

    As the population ages, the prevalence of AD will increase drastically.

    “However, as the population ages, the prevalence of AD will increase drastically [15], prohibitively increasing the burden of assessment and monitoring.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [15] 15. Liesi E. Hebert, Jennifer Weuve, Paul A. Scherr, and Denis A. Evans. 2013. Alzheimer disease in the united states (2010-2050) estimated using the 2010 census. Neurology 80, 19 (May 2013), 1778-83. DOI:https://doi.org/10.1212/WNL. 0b013e31828726f5 source

  • May combine two claims.

    Premisecitation

    Social robots have been used to engage individuals in social interactions, which is critical in the maintenance of good cognitive health.

    “For example, social robots have been used to provide personalized and affordable speech and cognitive therapy [39] and to engage individuals in social interactions, critical in the maintenance of good cognitive health [24,27].”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites [24] 24. Francisco Martín, Carlos E. Agüero, José M. Cañas, Meritxell Valenti, and Pablo Martínez-Martín. 2013. Roboth- erapy with dementia patients. Int. J. Adv. Robot. Syst. 10, 1 (2013), 10. DOI:https://doi.org/10.5772/54765 arXiv:http://dx.doi.org/10.5772/54765 source
    • cites [27] 27. Elaine Mordoch, Angela Osterreicher, Lorna Guse, Kerstin Roger, and Genevieve Thompson. 2013. Use of social commitment robots in the care of elderly people with dementia: A literature review. Maturitas 74, 1 (2013), 14-20. DOI:https://doi.org/10.1016/j.maturitas.2012.10.015 source

  • Classified as probably not a substantial claim.

    Study design and conductcitation

    Audio of the conversations were transcribed at the utterance level and annotated by a trained speech-language pathologist with speaker information, dialogue acts (DAs), and trouble-source-repair (TSR) variables.

    “Audio of the conversations were transcribed at the utterance level and annotated by a trained speech-language pathologist with speaker information, dialogue acts (DAs), and trouble-source-repair (TSR) variables [30].”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites [30] 30. Joseph B. Orange, Rosemary B. Lubinski, and D. Jeffery Higginbotham. 1996. Conversational repair by individuals with dementia of the alzheimer's type. J. Speech Hear. Res. 39, 4 (Aug. 1996), 881-95. source

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

    Premisecitation

    The cost of care in AD tends to be expensive.

    “The cost of care in AD tends to be expensive, and the burden on caregivers is rapidly increasing, along with the number of people with AD [6,12].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [6] 6. Cathleen M. Connell, Mary R. Janevic, and Mary P. Gallant. 2001. The costs of caring: Impact of dementia on family caregivers. J. Geriatr. Psychiat. Neurol. 14, 4 (Dec. 2001), 179-187. DOI:https://doi.org/10.1177/089198870101400403 source
    • cites [12] 12. Dennis Getsios, Steve Blume, Khajak J. Ishak, Grant Maclaine, and Luis Hernández. 2012. An economic evaluation of early assessment for alzheimer's disease in the united kingdom. Alzheim. Dement. 8, 1 (Jan. 2012), 22-30. DOI: https://doi.org/10.1016/j.jalz.2010.07.001 source

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

    Study design and conductnone given

    Gold standard transcripts were used for subsequent analysis.

    “We note that gold standard transcripts are used for subsequent analysis.”
    Quote in the paperNot a substantial claim

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

    Study design and conductnone given

    Assent was provided by family members where informed consent was not possible, according to an approved ethics protocol approved by the University of Toronto (protocol 32947).

    “Assent was provided by family members where informed consent was not possible, according to an approved ethics protocol approved by the University of Toronto (protocol 32947).”
    Quote in the paperNot a substantial claim

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

    Premisecitation

    The burden on caregivers is rapidly increasing along with the number of people with AD.

    “The cost of care in AD tends to be expensive, and the burden on caregivers is rapidly increasing, along with the number of people with AD [6,12].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [6] 6. Cathleen M. Connell, Mary R. Janevic, and Mary P. Gallant. 2001. The costs of caring: Impact of dementia on family caregivers. J. Geriatr. Psychiat. Neurol. 14, 4 (Dec. 2001), 179-187. DOI:https://doi.org/10.1177/089198870101400403 source
    • cites [12] 12. Dennis Getsios, Steve Blume, Khajak J. Ishak, Grant Maclaine, and Luis Hernández. 2012. An economic evaluation of early assessment for alzheimer's disease in the united kingdom. Alzheim. Dement. 8, 1 (Jan. 2012), 22-30. DOI: https://doi.org/10.1016/j.jalz.2010.07.001 source

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

    Premisecitation

    Previous research has utilized machine learning approaches to assess dementia based on speech and transcripts of a picture description task.

    “Previous research has utilized machine learning approaches to assess dementia based on speech and transcripts of a picture description task [8,19].”
    Quote in the paperNot a substantial claim0 of 2 sources checked
    • cites [8] 8. Juliana Onofre De Lira, Karin Zazo Ortiz, Aline Carvalho Campanha, Paulo Henrique Ferreira Bertolucci, and Thaís Soares Cianciarullo Minett. 2011. Microlinguistic aspects of the oral narrative in patients with alzheimer's disease. Int. Psychogeriatr. 23, 03 (Apr. 2011), 404-412. DOI:https://doi.org/10.1017/S1041610210001092 source
    • cites [19] 19. William Jarrold, Bart Peintner, David Wilkins, Dimitra Vergryi, Colleen Richey, Maria Luisa Gorno-Tempini, and Jennifer Ogar. 2014. Aided diagnosis of dementia type through computer-based analysis of spontaneous speech. In Proceedings of the Workshop on Computational Linguistics and Clinical Psychology: From Linguistic Signal to Clinical Reality. 27-37. DOI:https://doi.org/10.3115/v1/W14-3204 source

  • May combine two claims.

    Interpretationown data

    Misunderstanding is unavoidable, and participants will have problems with intelligibility.

    “Our experiments highlight the need for variety and flexibility in dialogue; misunderstanding is unavoidable, and participants will have problems with intelligibility.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    The autonomous robot system is robust but inflexible due to rule-based approaches and simple error handling.

    “Through rule-based approaches and simple errorhandling, we produced a robust but inflexible system.”
    Quote in the paperNot a substantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationnone given

    Acoustic features would likely be useful in cognitive monitoring.

    “Similarly, acoustic features would also likely be useful in cognitive monitoring, but were not included in our analysis.”
    Quote in the paperNot a substantial claim

  • Classified as probably not a substantial claim.

    Interpretationnone given

    The automatic linguistic analysis demonstrated with Ludwig can be used in longitudinal monitoring and assessment of dementia.

    “With Ludwig, we demonstrate the type of automatic linguistic analysis that we envision being used in longitudinal monitoring and assessment of dementia.”
    Quote in the paperSubstantial claim

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

    Premisecitation

    Past research has shown that robots are useful as a companion to older adults.

    “While past research has shown that robots are useful as a companion to older adults [28,41,42]”
    Quote in the paperNot a substantial claim0 of 3 sources checked
    • cites [28] 28. Wendy Moyle, Marie Cooke, Elizabeth Beattie, Cindy Jones, Barbara Klein, Glenda Cook, and Chrystal Gray. 2013. Exploring the effect of companion robots on emotional expression in older adults with dementia: A pilot randomized controlled trial. J. Gerontol. Nurs. 39, 5 (May 2013), 46-53. DOI:https://doi.org/10.3928/00989134-20130313-03 source
    • cites [41] 41. Kazuyosh Wada and Takanori Shibata. 2007. Living with seal robots-Its sociopsychological and physiological influ- ences on the elderly at a care house. IEEE Trans. Robot. 23, 5 (Oct. 2007), 972-980. DOI:https://doi.org/10.1109/TRO. 2007.906261 source
    • cites [42] 42. Kazuyosh Wada, Takanori Shibata, Toshimitsu Musha, and Shin Kimura. 2008. Robot therapy for elders affected by dementia. IEEE Eng. Med. Biol. Mag. 27, 4 (Jul. 2008), 53-60. DOI:https://doi.org/10.1109/MEMB.2008.919496 source

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

    Premisecitation

    MMSE scores of 18 to (presumably 18-20) are associated with mild cognitive impairment, 11 to 17 with moderate impairment, and 10 or less with severe cognitive impairment.

    “A score of 18 to”
    • Also stated in: § B.3 Semantic Features · plain
      “is associated with mild cognitive impairment, 11 to 17 with moderate impairment, and 10 or less with severe cognitive impairment[9].”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites [9] 9. Marshal F. Folstein, Susan E. Folstein, and Paul R. McHugh. 1975. "Mini-mental state". A practical method for grading the cognitive state of patients for the clinician. J. Psychiatr. Res. 12, 3 (1975), 189-198. source

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Controlling the robot places extra cognitive burden on the operator.

    “Controlling the robot places extra cognitive burden on the operator.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationnone given

    Decreased intelligibility of the robot's synthetic voice and the lack of flexible prosodic, gestural, and facial cues may have contributed to participants being less responsive to a WoZ robot.

    “Decreased intelligibility of the robot's synthetic voice (and the lack of flexible prosodic, gestural, and facial cues) may have contributed to the participants being less responsive to a”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S2 (checkable), S3 (load-bearing), S4 (based).

    Interpretationnone given

    Miscommunication features would be useful in assessing dementia.

    “However, these features would be useful in assessing dementia.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Study design and conductnone given

    In the WoZ sessions, the operator largely filtered out fragment and stall utterances and used more measured or thought-out utterances.

    “In the process of passing commands to the robot, the operator largely filtered out fragment and stall utterances (stuttering, rephrasing, etc.) and used more measured or thought-out utterances.”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning

    MMSE scores cannot account for day-to-day symptomatic variation in Alzheimer's disease.

    “However, the MMSE scores cannot account for day-to-day symptomatic variation (e.g., good days and bad days).”
    Quote in the paperNot a substantial claim
    • Rests on an assumption: Alzheimer's disease symptom expression can vary drastically from day to day.

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

    Study design and conductreasoning

    Miscommunication was not included in the linguistic analysis because TSR variables were manually annotated.

    “Although miscommunication could be a useful feature in cognitive monitoring, it has not been included in this analysis, because TSR variables were manually annotated.”
    Quote in the paperNot a substantial claim

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

    Study design and conductreasoning

    Acoustic features were not included in the analysis.

    “Similarly, acoustic features would also likely be useful in cognitive monitoring, but were not included in our analysis.”
    Quote in the paperNot a substantial claim
76 of 76

  • The paper relies on this without support.

    Language can be a very sensitive measure for various cognitive and neuro-psychiatric conditions.

    “Indeed, language can be a very sensitive measure for various cognitive and neuro-psychiatric conditions.”

  • The authors state this as an assumption.

    The Wizard-of-Oz condition maintains the conversational abilities of the human interlocutor while replacing the physical form of a human with that of a robot.

    “In contrast, the Session 2 Wizard-of-Oz condition demonstrated the effect of replacing the physical form of a human with that of a robot, while maintaining the conversational abilities of the human interlocutor.”

  • The authors state this as an assumption.

    The first session with a human interlocutor provides a baseline for conversations surrounding a picture description.

    “The first session provided a baseline for conversations surrounding a picture description.”

  • The paper relies on this without support.

    MMSE scores validly classify participants into cognitive impairment groups (mild, moderate, severe).

    “We report 15 linguistic features that differ across MMSE groups (Mild vs. Moderate vs. Severe)”

  • The paper relies on this without support.

    The linguistic features extracted from conversations are valid measures of language ability relevant to cognitive impairment.

    “These features are good candidates for linguistic markers that may be used for monitoring cognitive impairment in people with AD”

  • The paper relies on this without support.

    The WoZ and autonomous robot sessions are comparable enough to attribute differences in flexibility to the robot's autonomy.

    “Evidently, the autonomous version of the robot was not as flexible as its remotely controlled counterpart.”

  • The paper relies on this without support.

    Alzheimer's disease symptom expression can vary drastically from day to day.

    “Alzheimer's disease is a degenerative disorder for which symptom expression can vary drastically from day to day.”

  • The paper relies on this without support.

    Facial features, gestures, and prosody are important conversational cues.

    “Furthermore, the autonomous robot does not have access to facial features, gestures, and prosody, which are important conversational cues, and is limited by the clarity of its synthetic voice, which is particularly important for interacting with older adults.”

  • The paper relies on this without support.

    The clarity of synthetic voice is particularly important for interacting with older adults.

    “Furthermore, the autonomous robot does not have access to facial features, gestures, and prosody, which are important conversational cues, and is limited by the clarity of its synthetic voice, which is particularly important for interacting with older adults.”

  • The paper relies on this without support.

    As the population ages, technological interventions will be needed to offset the increased burden placed on caregivers.

    “As the population ages, technological interventions will be needed to offset the increased burden placed on caregivers.”
10 of 10

Claims other papers make about this one

Each is checked against this paper’s own text: does it say what the citing paper says it does?

  • The autonomous robot in the feasibility study was limited by inflexible dialogs, unable to personalize conversations, or recover from breakdowns.

    “the robot was limited by inflexible dialogs, unable to personalize conversations, or recover from breakdowns.”

    From Conversational Affective Social Robots for Ageing and Dementia Support (2022) · cites this paper as [57]

    Section 5.4 (Limitations; page unavailable) describes the autonomous robot as inflexible and unable to adapt to new situations. Section 6 (Conclusions; page unavailable) says it could not identify and recover from communication breakdowns. The copy does not establish the more specific claim that it could not personalize conversations for individual users. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • Very few studies in Table I attempted to implement conversational strategies to handle dialogs with cognitive impaired individuals in an effective manner.

    “Very few studies in Table I attempted to implement conversational strategies to handle dialogs with cognitive impaired individuals in an effective manner (see [55], [57], [58]).”

    From Conversational Affective Social Robots for Ageing and Dementia Support (2022) · cites this paper as [57]

    Section 3.2.3 describes an implemented rule-based dialogue engine that used participants’ responses to select follow-up questions, supporting this paper as an example of an attempt to manage dialogue with adults with Alzheimer’s disease. But Section 5.3 says the autonomous robot could not repeat or reword misunderstood utterances, and Section 4.1.2 reports that most of its repairs were unsuccessful. The paper therefore does not establish that its strategies handled those dialogues effectively; the claim about how few studies appear in the citing paper’s Table I is the citing authors’ own comparison. (LLM, conf=0.84)

    Experimental — accuracy not yet measured

  • In the feasibility study, the human and the remotely controlled robot were able to respond to open-ended or flawed answers, sense when to finish conversation, and interpret affect from facial expressions; this was not verified for the autonomous robot.

    “The human and the remotely controlled robot [scenarios 1) and 2), respectively] were able to respond to open-ended or flawed answers, sense when to finish conversation, and interpret affect from facial expressions, which was not verified for the autonomous robot [scenario 3)]”

    From Conversational Affective Social Robots for Ageing and Dementia Support (2022) · cites this paper as [57]

  • In the feasibility study, conversations were more engaging with a human interlocutor, but there was an overall likeness toward verbal interactions with the robot.

    “Whilst conversations were more engaging with a human interlocutor, there was an overall likeness toward verbal interactions with the robot.”

    From Conversational Affective Social Robots for Ageing and Dementia Support (2022) · cites this paper as [57]

  • In a feasibility study with 19 seniors with dementia, linguistic analysis could be used in longitudinal monitoring and assessment of dementia, with lexical features automatically extracted.

    “As a proof of concept, the authors demonstrated how linguistic analysis could be used in longitudinal monitoring and assessment of dementia; lexical features were automatically extracted.”

    From Conversational Affective Social Robots for Ageing and Dementia Support (2022) · cites this paper as [57]

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • MMSE group boundaries change between methods and results

    Major

    The methods classify scores of 18-23 as mild and 11-17 as moderate. Tables 7 and 9 instead use 21-25 and 11-20, changing which group includes participants scoring 18-20.

  • Word length is called session-independent despite a significant session result

    Major

    The discussion names word length as unaffected by session, whereas Table 10 reports a significant session effect for it at p = 0.0331.

  • Partner assessment claim reverses the table's group means

    Major

    The text says partners used more Assess utterances with the severe group. Table 9 gives that group the lowest Assess percentage: 12.8%, versus 15.2% for moderate and 33.1% for mild.

  • Reported participant utterance means disagree with Table 5

    Major

    The text gives approximately 21, 13, and 4 participant utterances per image across the sessions. Table 5 instead reports means of 22.5, 15.5, and 8.41; the third-session difference is particularly large.

  • Reported MMSE range excludes a listed participant

    Minor

    The stated upper MMSE score is 19, but Table 1 lists P19 with scores of 21 in two sessions.

  • PP type proportion has incompatible definitions

    Minor

    The results describe this feature as a proportion of grammatical constituents that are PP phrases. Appendix B instead defines it using PP length divided by the transcript's word count, which measures something different.

  • Autonomous robot's rewording ability is described oppositely

    Minor

    One passage attributes P14's difficulty to the autonomous robot's ability to reword utterances; the discussion says that robot could not repeat or reword misunderstood utterances.

  • P2 receives incompatible impairment labels

    Minor

    The qualitative analysis calls P2 both severe and medium-mild, without indicating a change in classification between those descriptions.

8 of 8

Cites

Flagged first, then by citations

Research works printed but not linked: 3

  1. Lai Catherine, 2013, Proceedings of the Workshop on Affective Social Speech Signals (WASSS’13)
  2. Sakai Yoichi, Proceedings of the 7th Annual ACM/IEEE International Conference on Human-Robot Interaction (HRI’12)
  3. Tapus Adriana, ICRA Workshop on Social Interaction with Intelligent Indoor Robots.

Cited by

Sorted by citations

Related

Authors
Pou-Prom, C., Raimondo, S., & Rudzicz, F.
Affiliations
University of Toronto (CA)
ORCID
Venue
ACM Transactions on Human-Robot Interaction
Publisher
Association for Computing Machinery (ACM)
Published
2020-05-31
Volume
9
Issue
3
Pages
1-25
ISSN
2573-9522 (ISSN-L)
DOI
10.1145/3380785
OpenAlex
W3033476063
Cited by
59
References
45
Citations by year
2026: 6 · 2025: 13 · 2024: 12 · 2023: 9 · 2022: 9 · 2021: 8 · 2020: 2
FWCI
7.65
Type
journal article
Language
en
Open access
Licence
www.acm.org/publications/policies/copyright_policy#Background (vor)
Locations
ACM Transactions on Human-Robot Interaction (publishedVersion)
Funders
  • AGE-WELL (AGE-WELL Catalyst - National - 2017)
  • JP Bickell Foundation (JP Bickell Foundation Research grant - National -)
  • Ontario Ministry of Research, Innovation and Science
Topics
  • Social Sciences › Psychology › Social Psychology › Social Robot Interaction and HRI
  • Physical Sciences › Computer Science › Artificial Intelligence › Speech and dialogue systems
  • Health Sciences › Medicine › Psychiatry and Mental health › Dementia and Cognitive Impairment Research
Keywords
conversational robots, Alzheimer's disease, older adults, cognitive assessment, caregiver burden, human-robot interaction, dialogue analysis, miscommunication, linguistic features, cognitive impairment, robot monitoring, user acceptance

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 2026-10-11.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-11.

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