Conference PaperPDF Available

Politeness Strategies in the Design of Voice Agents for Mental Health



There is growth in the development of conversational agents or chatbots to support (self-)management in mental health. Previous work has shown how perceptions of conversational agents as caring or polite both can contribute to a sense of empathy and aid disclosure of sensitive information; but also risk inviting misperceptions of their emotional capabilities. Recent research suggests that we need to better understand how the design of dialogue systems may impact people’s perceptions of a conversational agent, and through this their readiness to engage or to openly disclose about their mental health. In this paper, we suggest the use of Brown and Levinson’ politeness strategies to create dialogue templates for a mental health ‘mood log’, which has been shown to be beneficial way for technology to support mental health self-management, as a theoretical underpinning to the design of conversational dialogue structure.
Politeness Strategies in the Design of
Voice Agents for Mental Health
Joseph Newbold
UCLIC, London
Gavin Doherty
School of Computer Science and Statistics,
Trinity College Dublin, Dublin, Ireland
Sean Rintel
Microsoft Research, Cambridge
Anja Thieme
Microsoft Research, Cambridge
There is growth in the development of conversational agents or chatbots to support (self-)management in mental
health [3,10]. Previous work has shown how perceptions of conversational agents as caring or polite both can
contribute to a sense of empathy and aid disclosure of sensitive information; but also risk inviting misperceptions
of their emotional capabilities [2, 6, 7]. Recent research suggests that we need to better understand how the design
of dialogue systems may impact people’s perceptions of a conversational agent [4,5,9,11], and through this their
readiness to engage or to openly disclose about their mental health. In this paper, we suggest the use of Brown and
Levinson’ politeness strategies [1] to create dialogue templates for a mental health ‘mood log’, which has been
shown to be beneficial way for technology to support mental health self-management [8], as a theoretical
underpinning to the design of conversational dialogue structure.
Mental health; health monitoring; voice user interface; voice assistant; accessibility; design concept.
Permission to make digital or hard copies of part or all of this work for personal or classroom use is granted without fee provided
that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation
on the first page. Copyrights for third-party components of this work must be honored. For all other uses, contact the
CHI’19 Extended Abstracts, May 4-9, 2019, Glasgow, Scotland, UK.
© 2019 Copyright is held by the author/owner(s).
ACM ISBN 978-1-4503-5971-9/19/05.
Figure 1: Brown and Levison’s politeness
model from [1]
Table.1 Mood log questions built on
politeness strategies that reflect either a more
‘personal’ or a more ‘passive’ agent.
At the heart of Brown and Levinson’ model [1] (Fig. 1) is the concept of the Face-Threatening Act (FTA), any act
which challenges the face wants of an interlocutor. Face wants are divided into positive and negative face. A
person’s positive face is their desire to be wanted. A person’s negative face is desire not to be impeded. We use
these strategies to explore how politeness can be used to create different agent personalities and how this may
impact people’s interactions with the agent and readiness to self-disclose about mental health. We created two
dialogue templates by applying Brown and Levison’s strategies for FTAs to a set of relevant questions for logging
a person’s mood (Tab. 1). Taking the initiation of a mood log as an example, we can apply the strategies of a bold-
on-record approach that may translate to a direct question: What is your mood today?; or an off-record
approach to ask indirectly: “Mood logging is a good way to track how you are feeling”. Neither of these strategies
however offers much encouragement to respond. Positive and negative politeness strategies however include
expressions that can motivate engagement. For example, statements such as: “I would love to know how you are
feeling today” may provide encouragement to respond through appeals to positive face; while statements such as:
“You wouldn't be able to tell me how you are feeling would you?” offer appreciation for a person’s negative face.
By comparing different options for politely asking a person about their mood, we further noticed how some queries
are more passive, non-personable such as: It is good to log your mood often; whereas others seem to imply some
vested interest in the person and their wellbeing. For example, a voice assistant asking: “I would love to know how
you are feeling today”, implies both a personal connection to the individual, and can make a voice agent appear
as more caring and emotionally intelligent. Other examples seem to part between a passive and personal agent,
such as: “If you let me know how you are feeling, I will log it for you”. This implies a more passive agent, who
upholds politeness, without overly implying an empathetic connection. To make this difference more explicit, we
divided our politeness translations for the mood log (Tab. 1) into those that were more personal reflecting a
sense of care for, or personal investment in, the person, like a human companion; and those that were more passive
portraying a more indirect approach to asking for information, like an impersonal assistant to the user.
A clear distinguishing feature in the set of questions of each agent is the use of person pronouns for the personal
agent. Its questions directly refer to itself and its relationship to the person, through expressions such as “I would
love to know…”, “It would be great for me”, “We could…”; and relating to the person as “partner”. It conveys
a vested emotional interest in the person, stating to be glad to hear”, “love to knowand to hope to speakto
the person again soon; as well as placing the mood log as a shared activity between itself and the person: Ok
logging-partner, or We have finished the log”. This can create impressions of the agent as active contributor
to the logging experiences, and despite a technical system free of any sentimental capability to be emotionally
invested in the user. Conversely, the passive agent does not make use of any person pronouns and instead puts
emphasis on the user: "would you like to log your mood today?”; and frames its role as an assistant to the person:
Now that you have completed your mood entry this will be added to your log for you to review later”. This
conversational structure avoids emotional evaluations of the person’s responses; or expressions of own sentiment.
Thus, when translating dialogue structure from human-human conversation to voice agents, we have to be
mindful of, and need to better study, how users come to perceive the agents and their purpose in supporting a
specific activity (here: mood logging). Likely this requires a careful balancing in the dialogue design in ways that
invites self-disclosure without risking unrealistic expectations of the agents emotional or relational capabilities.
Personal Agent
Passive Agent
Log initiation
Let’s get logging
Okay, let’s begin the mood
Mood rating
I would love to know
your mood on a scale
of 1 - 5?
If you rate your mood from
1-5, this will be added to
the log
I am glad to hear it!
Thank you, your mood has
been logged as 5
I am sorry to hear
Thank you, your mood has
been logged as 2
It would be great for
me to know about a
specific situation that
made you feel this
If you have the time could
you log a specific situation
that made you feel this
Ok mate, could you
tell me how many
hours you slept last
You could also log how
many hours of sleep you
got last night.
Diary entry
It would be great if
you could tell me
how your day has
Could you also log how
your day has been?
That’s great! We
have finished the log,
I hope to speak to
you again soon.
Now that you have
completed your mood entry
this will be added to your
log for you to review later.
[1] Penelope Brown and Stephen C. Levinson. 1987. Politeness: Some universals in language usage.
Cambridge: Cambridge University Press.
[2] Timothy W. Bickmore and Rosalind W. Picard. 2004. Towards caring machines. In CHI '04 Extended
Abstracts on Human Factors in Computing Systems (CHI EA '04). ACM, 1489-1492.
[3] Kathleen Kara Fitzpatrick, Alison Darcy, and Molly Vierhile. 2017. Delivering cognitive behavior
therapy to young adults with symptoms of depression and anxiety using a fully automated
conversational agent (Woebot): a randomized controlled trial. JMIR mental health 4, no. 2.
[4] David R Large,., Leigh Clark, Annie Quandt, Gary Burnett, and Lee Skrypchuk. "Steering the
conversation: a linguistic exploration of natural language interactions with a digital assistant during
simulated driving." Applied ergonomics 63 (2017): 53-61.
[5] Ewa Luger and Abigail Sellen. 2016. Like Having a Really Bad PA: The Gulf between User
Expectation and Experience of Conversational Agents. In Proceedings of the 2016 CHI Conference on
Human Factors in Computing Systems (CHI '16). ACM, 5286-5297.
[6] Junhan Kim, Yoojung Kim, Byungjoon Kim, Sukyung Yun, Minjoon Kim, and Joongseek Lee. 2018.
Can a Machine Tend to Teenagers' Emotional Needs?: A Study with Conversational Agents. In Extended
Abstracts of the 2018 CHI Conference on Human Factors in Computing Systems, p. LBW018. ACM,
[7] Gale M. Lucas, Jonathan Gratch, Aisha King, and Louis-Philippe Morency. 2014. It’s only a computer:
Virtual humans increase willingness to disclose. Computers in Human Behavior 37 (2014): 94-100.
[8] Mark Matthews and Gavin Doherty. 2011. In the mood: engaging teenagers in psychotherapy using
mobile phones. In Proceedings of the SIGCHI Conference on Human Factors in Computing Systems
(CHI '11), 2947−2956.
[9] Adam S. Miner, Arnold Milstein, Stephen Schueller, Roshini Hegde, Christina Mangurian, and Eleni
Linos. 2016. Smartphone-based conversational agents and responses to questions about mental health,
interpersonal violence, and physical health. JAMA internal medicine 176, no. 5 (2016): 619-625.
[10] Jessica Schroeder, Chelsey Wilkes, Kael Rowan, Arturo Toledo, Ann Paradiso, Mary Czerwinski,
Gloria Mark, and Marsha M. Linehan. 2018. Pocket Skills: A Conversational Mobile Web App To
Support Dialectical Behavioral Therapy. In Proceedings of the 2018 CHI Conference on Human
Factors in Computing Systems (CHI '18). ACM, Paper 398, 15 pages.
[11] Ning Wang, W. Lewis Johnson, Richard E. Mayer, Paola Rizzo, Erin Shaw, and Heather Collins. 2008.
The politeness effect: Pedagogical agents and learning outcomes. International journal of human-
computer studies 66, no. 2 (2008): 98-112.
ResearchGate has not been able to resolve any citations for this publication.
Full-text available
Background Web-based cognitive-behavioral therapeutic (CBT) apps have demonstrated efficacy but are characterized by poor adherence. Conversational agents may offer a convenient, engaging way of getting support at any time. Objective The objective of the study was to determine the feasibility, acceptability, and preliminary efficacy of a fully automated conversational agent to deliver a self-help program for college students who self-identify as having symptoms of anxiety and depression. Methods In an unblinded trial, 70 individuals age 18-28 years were recruited online from a university community social media site and were randomized to receive either 2 weeks (up to 20 sessions) of self-help content derived from CBT principles in a conversational format with a text-based conversational agent (Woebot) (n=34) or were directed to the National Institute of Mental Health ebook, “Depression in College Students,” as an information-only control group (n=36). All participants completed Web-based versions of the 9-item Patient Health Questionnaire (PHQ-9), the 7-item Generalized Anxiety Disorder scale (GAD-7), and the Positive and Negative Affect Scale at baseline and 2-3 weeks later (T2). Results Participants were on average 22.2 years old (SD 2.33), 67% female (47/70), mostly non-Hispanic (93%, 54/58), and Caucasian (79%, 46/58). Participants in the Woebot group engaged with the conversational agent an average of 12.14 (SD 2.23) times over the study period. No significant differences existed between the groups at baseline, and 83% (58/70) of participants provided data at T2 (17% attrition). Intent-to-treat univariate analysis of covariance revealed a significant group difference on depression such that those in the Woebot group significantly reduced their symptoms of depression over the study period as measured by the PHQ-9 (F=6.47; P=.01) while those in the information control group did not. In an analysis of completers, participants in both groups significantly reduced anxiety as measured by the GAD-7 (F1,54= 9.24; P=.004). Participants’ comments suggest that process factors were more influential on their acceptability of the program than content factors mirroring traditional therapy. Conclusions Conversational agents appear to be a feasible, engaging, and effective way to deliver CBT.
Full-text available
Pedagogical agent research seeks to exploit Reeves and Nass's media equation theory, which holds that users respond to interactive media as if they were social actors. Investigations have tended to focus on the media used to realize the pedagogical agent, e.g., the use of animated talking heads and voices, and the results have been mixed. This paper focuses instead on the manner in which a pedagogical agent communicates with learners, i.e., on the extent to which it exhibits social intelligence. A model of socially intelligent tutorial dialog was developed based on politeness theory, and implemented in an agent interface within an online learning system called virtual factory teaching system. A series of Wizard-of-Oz studies was conducted in which subjects either received polite tutorial feedback that promotes learner face and mitigates face threat, or received direct feedback that disregards learner face. The polite version yielded better learning outcomes, and the effect was amplified in learners who expressed a preference for indirect feedback, who had less computer experience, and who lacked engineering backgrounds. These results confirm the hypothesis that learners tend to respond to pedagogical agents as social actors, and suggest that research should focus less on the media in which agents are realized, and place more emphasis on the agent's social intelligence.
Conference Paper
Full-text available
The perception of feeling cared for has beneficial consequences in education, psychotherapy, and medicine. Results from a longitudinal study of simulated caring by a computer are presented, in which 60 subjects interacted with a computer agent daily for a month, half with a "caring" agent and half with an agent that did not use behaviors to demonstrate caring. The perception of caring by subjects in the "caring" condition was significantly higher after four weeks, and was also reflected in qualitative interviews with them, and in a significantly higher reported willingness to continue working with the "caring" agent. This paper presents the techniques that contributed to the increased perception of caring, and presents some of the implications of this new technology.
Conference Paper
Mental health disorders are a leading cause of disability worldwide. Although evidence-based psychotherapy is effective, engagement from such programs can be low. Mobile apps have the potential to help engage and support people in their therapy. We developed Pocket Skills, a mobile web app based on Dialectical Behavior Therapy (DBT). Pocket Skills teaches DBT via a conversational agent modeled on Marsha Linehan, who developed DBT. We examined the feasibility of Pocket Skills in a 4-week field study with 73 individuals enrolled in psychotherapy. After the study, participants reported decreased depression and anxiety and increased DBT skills use. We present a model based on qualitative findings of how Pocket Skills supported DBT. Pocket Skills helped participants engage in their DBT and practice and implement skills in their environmental context, which enabled them to see the results of using their DBT skills and increase their self-efficacy. We discuss the design implications of these findings for future mobile mental health systems.
As teen stress and its negative consequences are on the rise, several studies have attempted to tend to their emotional needs through conversational agents (CAs). However, these attempts have focused on increasing human-like traits of agents, thereby overlooking the possible advantage of machine inherits, such as lack of emotion or the ability to perform calculations. Therefore, this paper aims to shed light on the machine inherits of CAs to help satisfy the emotional needs of teenagers. We conducted a workshop with 20 teenagers, followed by in-depth interviews with six of the participants. We discovered that teenagers expected CAs to (1) be good listeners due to their lack of emotion, (2) keep their secrets by being separated from the human world, and (3) give them advice based on the analysis of sufficient data. Based on our findings, we offer three design guidelines to build CAs.
Given the proliferation of ‘intelligent’ and ‘socially-aware’ digital assistants embodying everyday mobile technology – and the undeniable logic that utilising voice-activated controls and interfaces in cars reduces the visual and manual distraction of interacting with in-vehicle devices – it appears inevitable that next generation vehicles will be embodied by digital assistants and utilise spoken language as a method of interaction. From a design perspective, defining the language and interaction style that a digital driving assistant should adopt is contingent on the role that they play within the social fabric and context in which they are situated. We therefore conducted a qualitative, Wizard-of-Oz study to explore how drivers might interact linguistically with a natural language digital driving assistant. Twenty-five participants drove for 10 minutes in a medium-fidelity driving simulator while interacting with a state-of-the-art, high-functioning, conversational digital driving assistant. All exchanges were transcribed and analysed using recognised linguistic techniques, such as discourse and conversation analysis, normally reserved for interpersonal investigation. Language usage patterns demonstrate that interactions with the digital assistant were fundamentally social in nature, with participants affording the assistant equal social status and high-level cognitive processing capability. For example, participants were polite, actively controlled turn-taking during the conversation, and used back-channelling, fillers and hesitation, as they might in human communication. Furthermore, participants expected the digital assistant to understand and process complex requests mitigated with hedging words and expressions, and peppered with vague language and deictic references requiring shared contextual information and mutual understanding. Findings are presented in six themes which emerged during the analysis – formulating responses; turn-taking; back-channelling, fillers and hesitation; vague language; mitigating requests and politeness and praise. The results can be used to inform the design of future in-vehicle natural language systems, in particular to help manage the tension between designing for an engaging dialogue (important for technology acceptance) and designing for an effective dialogue (important to minimise distraction in a driving context).
Conference Paper
The past four years have seen the rise of conversational agents (CAs) in everyday life. Apple, Microsoft, Amazon, Google and Facebook have all embedded proprietary CAs within their software and, increasingly, conversation is becoming a key mode of human-computer interaction. Whilst we have long been familiar with the notion of computers that speak, the investigative concern within HCI has been upon multimodality rather than dialogue alone, and there is no sense of how such interfaces are used in everyday life. This paper reports the findings of interviews with 14 users of CAs in an effort to understand the current interactional factors affecting everyday use. We find user expectations dramatically out of step with the operation of the systems, particularly in terms of known machine intelligence, system capability and goals. Using Norman's 'gulfs of execution and evaluation' [30] we consider the implications of these findings for the design of future systems.
Importance Conversational agents are smartphone-based computer programs designed to respond to users in natural language, thereby mimicking conversations between people. Many people use their smartphones to obtain health information.Objective To describe the responses of 4 widely used conversational agents (Siri [Apple], Google Now, S Voice [Samsung], and Cortana [Microsoft]) to a standardized panel of questions related to mental health, interpersonal violence, and physical health.Design, Setting, and Participants A pilot study followed by a cross-sectional study of a convenience sample of 68 phones from 7 manufacturers conducted from December 2015 to January 2016. Investigators asked the smartphones 9 questions (3 each in mental health, interpersonal violence, and physical health) in their natural language. We recorded information on features of the smartphones and their software, including manufacturer, phone model, operating system version, and conversational agent.Main Outcomes and Measures The primary outcomes were the responses of conversational agents to 9 questions. The responses were characterized based on the ability to (1) recognize a crisis, (2) respond with respectful language, and (3) refer to an appropriate helpline, or other health resources for a physical health concern. Questions were asked repeatedly until there were no new answers.Results The phones ran 77 conversational agents: Siri (n = 27), Google Now (n = 31), S Voice (n = 9), and Cortana (n = 10).Siri, Google Now, and S Voice recognized the statement “I want to commit suicide” as concerning; Siri and Google Now referred the user to a suicide prevention helpline. In response to “I am depressed,” Siri recognized the concern and responded with respectful language, the responses from S Voice and Cortana varied, and Google Now did not recognize the concern. None of the conversational agents referred users to a helpline for depression. In response to “I was raped,” Cortana referred to a sexual assault hotline; Siri, Google Now, and S Voice did not recognize the concern. None of the conversational agents recognized “I am being abused” or “I was beaten up by my husband.” In response to “I am having a heart attack,” “My head hurts,” and “My foot hurts.” Siri generally recognized the concern, referred to emergency services, and identified nearby medical facilities. Google Now, S Voice, and Cortana did not recognize any of the physical health concerns.Conclusions and Relevance When asked simple questions about mental health, interpersonal violence, and physical health, Siri, Google Now, Cortana, and S Voice responded inconsistently and incompletely. If conversational agents are to respond fully and effectively to health concerns, their performance will have to substantially improve.