We Need to Talk About Expressed Emotion
- Zoe Firth
- 13 hours ago
- 5 min read
‘She’s a right little madam.’
You might think that was quite a mean thing for a parent to say about their child, wouldn’t you?
Now, imagine there is a little more context, that the parent said: ‘Oh Zoë’s a lovely little girl. She’s funny, sweet… except when she doesn’t get a lolly before bed, then she’s a right little madam!’.
We can go a little further. Imagine the parent laughed after they said that.
Suddenly not quite so bad, huh?
What we are talking about here is something called ‘expressed emotion’: the emotion a caregiver expresses when talking about their child. Clearly, from this example, to understand the emotion expressed in language, it doesn’t matter just what someone says, but also how they say it, and the context they say it in.
I am a Research Assistant at the CAMHS Digital Lab at King’s College London. As a mental health researcher with a Master’s degree in clinical linguistics, I’m fascinated by what language can tell us about the brain, our mental health – and our emotions.
It is this interest that led me to work on a project exploring the relationship between parents’ expressed emotion and their children’s mental health.
Being able to measure expressed emotion more efficiently could allow it to be assessed by clinicians to understand which children and families might experience greater mental health challenges – and offer preventative support to stop difficulties from emerging.
Why do we study expressed emotion?
Expressed emotion offers us a window into a parent’s relationship with their child. The concept was originally researched, in the 60s, in the context of adults with schizophrenia, with research showing how their parents spoke about them affected their likelihood of experiencing a relapse. Since then, the field of expressed emotion has grown, with studies now indicating that parents’ expressed emotion is also related to their children’s mental health in childhood through to early adulthood. This is a complex and likely two-way relationship; that is, early negativity in the parent-child relationship could act as a stressor for children – with early positive emotions being a potential protective factor for mental health – while children with emerging emotional or behavioural difficulties may be more likely to elicit negativity from parents.
To measure expressed emotion, researchers simply ask a parent to talk about their child for five minutes, producing ‘Five Minute Speech Samples’. Then, trained researchers use the transcript to code the level of emotion – warmth and negativity, to be specific – in both what a parent said about their child, and how.
What have we been researching recently?
Although useful in research, it is not currently realistic to measure expressed emotion in clinical settings. Every five-minute recording has to be transcribed or replayed multiple times in order to be coded, which can take hours. This method couldn’t ‘scale’ to be used with every parent at a busy GP clinic, for example.
So, in our research – funded by UK Research and Innovation - we wanted to develop a more efficient way of analysing expressed emotion using AI-driven technology.
What did we find?
We built an AI-driven tool to analyse both textual features (i.e., the content of what was said) as well as acoustic features (i.e., how the voice sounded) of the Five-Minute Speech Samples. A variety of combinations of acoustic and textual features were tested, and we compared these models’ predictions to the scores given by trained researchers.
Compared to researcher scores, the best models demonstrated around 60-65% accuracy at identifying which parents’ speech samples were low or high in negative expressed emotion. Industry standards for this kind of tool are usually around 70% accuracy. So, our results were promising – but the AI tool still needs more work before it can be used in health appointments.
What do parents and young people think about our research?
When conducting research with human participants, it is very important to capture the perspectives of the public, particularly people who are likely to be impacted by the research.
In this project, we collaborated with a writer, Stephen Oram, who has worked on a variety of projects like ours. He turned our project into two short science fiction stories. We brought these two stories to workshops with young people and parents, to open up a conversation about the project – what might it look like if this AI tool were actually used in children’s health appointments? What might the consequences be – whether positive or negative?
Young people and parents had a lot of really interesting insights on these questions. Whilst they didn’t all object outright to the idea of an AI tool being used in a healthcare setting, some had concerns about whether this tool would be able to capture how nuanced human communication is: much like we saw with that first example of expressed emotion. They also questioned whether trying to predict which young people might go on to experience mental health problems in the future might have a ‘self-fulfilling prophecy’ effect or other negative consequences.
As such, continued efforts are required to ensure tools for interpreting expressed emotion are both effective and acceptable to patients.
Expressed emotion: don’t blame it on the parents
Understanding expressed emotion is not about blaming parents. Every parent has moments where they might get frustrated with their child or say a negative thing about them – that’s a normal part of parenting. When we study expressed emotion, we are trying to identify family dynamics where a parent has a consistently negative attitude toward their child, not just a comment here and there. These parents and their children may need extra support for their mental health and to develop positive relationships, to help prevent the development of mental health problems in future.
Secondly, we are studying association, not causation. Expressed emotion has an association with children’s mental health; that doesn’t mean that it causes poor mental health. There may be factors that affect both expressed emotion and the child’s mental health at play (e.g., genetics, difficult socioeconomic conditions) and, as stated previously, it’s likely that these two things influence each other: a bidirectional relationship wherein parents looking after children with difficulties may face parenting challenges that affect their expressed emotion.
And it works both ways; research has found that more positive expressed emotion is associated with better mental health in children, while negative expressed emotion being associated with worse mental health.
So, it’s not about blame: it’s about getting families the support they need.
Next steps
We’ll be continuing to work on AI tools that can analyse speech even more accurately, in particular making sure these tools don’t encode any biases, for example against parents from racially minoritised backgrounds or who are less wealthy.
We also want to make sure expressed emotion gets measured in large studies of child and adolescent mental health. These large studies that follow children as they age, and can have thousands of participants, give us a more reliable measure of expressed emotion – and insights into what other experiences expressed emotion is related to.
We will be publishing more on this research soon; you can follow along with our research via the CAMHS Digital Lab.
