Politeness Strategies in the Design of Voice Agents for Mental Health

Authors: Joseph Newbold, Gavin Doherty, Sean Rintel, Anja Thieme
Venue: CHI 2019 Workshop
Author's copy (PDF)

AThe paper

A workshop paper that applies a theory of politeness to the questions a mental health voice agent asks when helping a person log their mood. It proposes two dialogue styles and has no user study.

  1. Politeness theoryBrown and Levinson's model of face, meaning a person's wish to be wanted and wish not to be imposed on.
  2. A voice agent mood logDialogue for a voice agent that helps a person log their mood, a mood situation, sleep and a diary entry.
  3. Compare the wordingThe paper compares the wording each politeness strategy produces, including pronouns and expressions of sentiment.

BThe method

  1. Start from the face threatEach mood log question is treated as a possible threat to the person's face.
  2. Choose a politeness strategyA question can be asked directly, hinted at, or worded to honour one of the two face wants.ask directlyhintappeal to being wantedavoid imposing
  3. Personal templateAppeals to the wish to be wanted. The agent speaks as I and we, calls the user partner and voices care.
  4. Passive templateRespects the wish not to be imposed on. The agent uses no first-person pronouns, acts as an assistant and makes no emotional evaluations.

Both templates word the same mood log questions, from starting the log to completing it. Neither template has yet been tested with users.

CWhat the paper argues

Wording a mood questionargued from politeness theory

  • The paper argues blunt or vague mood questions give little encouragement to respond.A direct question or an indirect hint offers little reason to answer. Wording that honours either face want can motivate engagement.
  • The paper argues some polite wording makes a voice agent seem caring and emotionally intelligent.Other wording stays polite without implying empathy, such as 'If you let me know how you are feeling, I will log it for you'.

The two templateswhat each dialogue style conveys

  • The personal template speaks as I and we, and frames the mood log as joint.The personal template calls the user partner and says it is glad to hear, would love to know and hopes to speak again soon.
  • The passive template keeps the focus on the user.The passive template uses no first-person pronouns, presents the agent as an assistant and makes no emotional evaluations.
  • The paper argues caring wording can imply feelings a voice agent does not have.The personal style can make a system with no emotional capability seem invested in the user. That risks misleading users about what the agent can do.

DKey takeaway

Politeness strategies can give mental health voice agents a principled way to word face-threatening questions, as caring companions or neutral assistants. Caring wording may invite disclosure yet imply empathy the agent lacks.

Made for this site with AI assistance from the verified research brief. It is not the paper's own figure.

Abstract

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.

Research brief

By Joseph Newbold, Gavin Doherty, Sean Rintel, and Anja Thieme.

Takeaway: Mental health voice agents can word mood questions to appeal to people's wish to feel wanted, as a caring companion ('I would love to know…'). Or they can appeal to the wish not to be imposed on, as a neutral assistant. Caring wording risks implying empathy the agent lacks.

Why this matters

What was known
Conversational agents for mental health self-management are growing, and recent research suggests we need to better understand how dialogue design shapes people's perceptions of them. 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 (§Introduction, p. 1)
What this paper adds
Rather than writing agent dialogue by intuition, we derive it from a theory of human conversation, applying Brown and Levinson's politeness strategies to the same mood-log questions to produce two contrasting dialogue templates. (§Designing Polite Conversational Voice Agents for Mood Logging, p. 2)
Who it matters to
Designers of dialogue for voice agents and chatbots that support mental health self-management, who must balance inviting disclosure against unrealistic expectations.

Research questions

Method and data

Design
Design concept (CHI 2019 workshop paper): applying Brown and Levinson's politeness strategies for face-threatening acts to the questions of a mental health mood log.
Setting
Dialogue for a conversational voice agent that helps a person log their mood, a mood situation, sleep and a diary entry.
Participants / data
A set of mood log queries, from log initiation to log completion, each written in a 'personal' and a 'passive' version (Table 1). No user study.
Analysis
Comparison of the wording each politeness strategy produces, and of the features (such as person pronouns and expressions of sentiment) that distinguish the two templates.

Findings

Questions this paper answers

Why does wording matter when a chatbot asks how you feel?
Asking about mood can threaten 'face' (a person's wish to be wanted and their wish not to be imposed on). Blunt questions or vague hints give little reason to answer, while wording that honours either wish can encourage people to engage. (See §Designing Polite Conversational Voice Agents for Mood Logging, p. 2.)
Should a mood-tracking voice assistant sound caring?
There is a trade-off. Earlier work shows caring or polite agents can foster empathy and the disclosure of sensitive information. But caring wording also makes a system with no feelings sound emotionally invested, which invites misperceptions of what it can do. (See §Designing Polite Conversational Voice Agents for Mood Logging, p. 2; §Designing Polite Conversational Voice Agents for Mood Logging, p. 2.)
Can theories of human conversation guide chatbot design?
They can supply structure. Brown and Levinson's politeness strategies turned one set of mood-log questions into two systematically different dialogue styles, from a caring companion to an impersonal assistant. Whether users respond as the theory predicts has not yet been tested. (See §Designing Polite Conversational Voice Agents for Mood Logging, p. 2; §Designing Polite Conversational Voice Agents for Mood Logging, p. 2.)

Contributions and implications

Limitations

›

Concepts

Face-Threatening Act (FTA)
The core of Brown and Levinson's politeness model: the concept of the Face-Threatening Act (FTA), any act which challenges the face wants of an interlocutor (§Designing Polite Conversational Voice Agents for Mood Logging, p. 2)
Positive and negative face
Brown and Levinson divide face wants in two: A person's positive face is their desire to be wanted. A person's negative face is desire not to be impeded. (§Designing Polite Conversational Voice Agents for Mood Logging, p. 2)
Mood log
The activity our agents support: a mental health 'mood log', which has been shown to be beneficial way for technology to support mental health self-management (§Introduction, p. 1)
Personal versus passive agent
The two styles we distinguished (the passive style asks indirectly, like an impersonal assistant): 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 (§Designing Polite Conversational Voice Agents for Mood Logging, p. 2)

Figures

Tree diagram of politeness strategies for a face-threatening act: do it on record baldly, on record with positive or negative politeness, off record, or don't do it.
Figure 1: Brown and Levison's politeness model from [1] From the paper, Figure 1, p. 2.

How to cite

Joseph Newbold, Gavin Doherty, Sean Rintel, and Anja Thieme. 2019. Politeness Strategies in the Design of Voice Agents for Mental Health. CHI 2019 Workshop.

This brief was prepared with AI assistance and published by Sean Rintel. It is not the authors' published text. Page numbers refer to this PDF.

BibTeX

@inproceedings{newbold2019politeness,
author = {Newbold, Joseph and Doherty, Gavin and Rintel, Sean and Thieme, Anja},
title = {Politeness Strategies in the Design of Voice Agents for Mental Health},
booktitle = {CHI 2019 Workshop},
year = {2019},
month = {May},
}