Are there ways as a conversation designer to work on social and structural issues?
I’m a linguistics student who is very interested in the social aspects of language and structural issues like racism and accentism.
I want to pursue a career in conversation design but I wonder if there are career paths in CxD concerned with those problems. If so, what are some recommended resources?
Back in the late 1990s, when I first started working on VUIs (voice user interfaces) I rarely heard people talking about issues with speech recognition and race or gender. I do recall some folks in the marketing department being excited because we could “detect a caller’s gender” on the phone with something like 95% accuracy, leading me to ask, ok…but what are we doing to do with it? (The answer: nothing much.)
Fast forward to 2022. I am glad to see that there are people thinking about this more, and taking a more careful look at the data. There are several aspects to this:
ASR (Automatic Speech Recognition) When someone speaks, what is the WER (Word Error Rate)? That is, how many words are recognized correctly? Is the WER higher for certain groups of people?
NLU (Natural Language Understanding) Even if the words are correctly recognized, does the system understand what the person is asking? Can people speak with the vocabulary and syntax that is natural for them, rather than having to “code switch”?
TTS (Text to Speech) What do the synthetic voices we hear in voice assistants, automated phone systems, virtual agents, robots, etc sound like? Are they representative?
It is clear we have a problem. In the 2020 paper “Racial disparities in automated speech recognition,” Stanford researchers looked as the most popular ARS engines and found that
all five ASR systems exhibited substantial racial disparities, with an average word error rate (WER) of 0.35 for black speakers compared with 0.19 for white speakers.
And in her 2019 HBR article “Voice Recognition Still Has Significant Race and Gender Biases,” Dr. Joan Palmiter Bajorek notes
Disparities exist because of the way we’ve structured our data analysis, databases, and machine learning.
We need to think about this in all the ways we create conversational experiences, from identifying bias in LLMs (Large Language Models) to building TTS voices that represent a variety of backgrounds, ethnicities, and genders, to designing conversation prompts and flows that are inclusive. This includes moving away from the default model of female voice assistants.
Some developments I’m happy to see include:
Being able to talk to the Google Assistant in two different languages (many households have people who are bilingual). Amazon’s Alexa has this capability as well!
Siri no longer defaulting to its white, female-sounding voice (now, users must choose a voice when they set up Siri, which now includes more diverse options) I also like the fact that when setting up the Google Assistant, it chooses a male- or female-sounding voice randomly, exposing more people to the different ways a voice assistant can sound.
Now, back to the original question. What are the career paths in conversation design to invest in, if you’re passionate about these issues? My advice is to bring this passion into whichever aspect of conversation design interests you. If you are invested in machine learning, work on LLMs, NLU, or creating TTS. Or, as a user researcher, you can work with people in different communities to best understand their needs and pain points, and communicate these to the product team. And as a designer, you can work to ensure your features, flows and prompts are inclusive.
Thank you for bringing up this important topic.
Additional resources: