Over the past few years, more and more people have been adding voice assistants like Alexa, Siri and Google Assistant to their homes and using them on their smartphones.
The most popular way to use these assistants is to ask questions - and we know users often expect to find answers from government. So we decided to put together a small team to look at how to meet this emerging need on GOV.UK.
Why voice assistants matter for GOV.UK
Smart speaker ownership is growing rapidly in the UK - 8% of adults now own one, up 3% in 2018 so far. In 2016 Google reported 20% of searches on Android devices were voice searches.
The leading voice platforms do not share data on user’s specific queries, but conversations we’ve had with teams at Amazon and Google made it clear that many users are asking questions where government is the best source.
Voice interfaces are nothing new for many people with access needs, who might use software like Dragon Naturally Speaking. But there's much excitement in the accessibility community about voice assistants. Their dramatically simpler interface has the potential to help lots of people who find computers and phones hard to use right now.
For GOV.UK, working on voice is an opportunity to meet the rising expectations of users and make government more accessible.
Tackling voice at scale
As government we need to approach voice services in a consistent way.
The recent GDS Innovation Survey has shown many local authorities, agencies and government departments are already exploring how to they can use voice to deliver information and services.
So in keeping with government’s design principles GOV.UK’s approach to this needs to be:
- cross-platform
- cross-government
- consistent
- scalable
Our team had experience creating apps and skills for voice assistants, but we wanted to find out if it was possible to support all the major voice platforms without having to build apps for each one.
Starting with the answers
We started by getting to grips with the different ways voice services provide users with answers.
We found there are three sources of information they use:
- search engines – these scan the web to provide relevant links and speakable snippets of content
- knowledge engines – these use a combination of data and computation to provide answers to fact-based questions
- third-party applications – known as skills on Alexa and Cortana, every major voice platform now has an app store
Here’s how it works for some of the most popular voice assistants:
| Search engine | Third-party applications | Knowledge engine | |
|---|---|---|---|
| Siri Apple |
iOS & SiriKit |
Wolfram Alpha & Siri Knowledge |
|
| Assistant |
Google actions |
KnowledgeGraph & Wikidata |
|
| Alexa Amazon |
Bing | Alexa skills |
Evi & Alexa Knowledge |
| Cortana Microsoft |
Bing | Cortana skills |
Bing Satori |
This analysis helped us realise we could increase the number of answers we provide in voice assistants simply by making it easier for search and knowledge engines to use GOV.UK as a data source.
Making GOV.UK more understandable to search engines
Because we publish on the open web and prioritise good content design, GOV.UK guidance is already a source of speakable answers on some voice platforms.
This works because search engines crawl our content and use machine learning to extract speakable answers. You can see it in action in this Google Assistant demo we made:
But we realised there was more we could do. By using the schema.org structured data standard we can give search engines extra context to help make sense of our pages.
This quarter we implemented:
- Article schema for all guidance on GOV.UK
- NewsArticle schema for all news on GOV.UK
- HowTo schema for our new step-by-step journeys
There are also wider implications for our content strategy that we will need more time to think through. We already write content with all the clarity, natural language and brevity we can, but good spoken answers require even more. For example, Amazon recommend a voice answer be speakable in just one breath.
With this in mind we plan to:
- double down on open publishing, ensuring guidance isn’t hidden in services
- improve our use of structured data to make our content more understandable to search engines
- integrate (even more) concise answers into our guidance
Getting GOV.UK data into knowledge engines
Where search engines scan the web to find answers, knowledge engines use databases of known facts to work them out.
We think the most powerful way to help users of voice assistants is to make the canonical data that government publishes available in a format which knowledge engines can use to provide answers.
We’re currently talking to some of the major knowledge engines to understand the easiest standards-based way for them to use the data published on GOV.UK via an API.
We already have an API for Content on GOV.UK, but it does not provide the detailed structure that knowledge engines need. So a few weeks ago we began a small experiment and created a new API that provides a limited set of highly structured content.
We’ll let you know the results of the experiment after it’s concluded.
The current limitations of voice apps and skills
You might have noticed we’ve focused on being able to answer questions. We have not mentioned transactions with government - things like making a benefit claim or booking a driving test. This is because at the moment voice interfaces do not really have any way for us to handle:
- data privacy – many voice services hold records of conversations
- verifying identity – many government services require a high level of identity assurance
- inputting personal data – even inputting a full address at the moment can be awkward
These are significant blockers for many government transactions. But we’ll continue to keep a close eye on the voice landscape, and hopefully as the technology improves we'll start to see some new features and cross-platform standards in these areas.
What the future sounds like
If you want to get a sense of the potential of voice, here are few queries you can try out on Google Assistant on Android or iOS right now:
- How long does it take to get a new passport?
- How much does a driving test cost?
- When will I get child benefit paid?
18 comments
Comment by Rafal Drewnowski posted on
Is there a way to publish gov content/knowledge in a way that it's accessible for developers of voice assistants? Maybe something like RDF schema for an onthology?
Comment by Sam Dub posted on
Hi Rafal,
This is really what we'd like to work towards, and we'd love to hear suggestions on how to do this. As the piece mentions we've been looking at adding more structure to our data using schema.org but its early days.
Sam
Comment by Naman Modi posted on
Best work you have done, this online website is cool with great facts and looks. I have stopped at this blog after viewing the excellent content. I will be back for more qualitative work.
Comment by Sohail Ahmad posted on
Very Nice
Comment by Ben Rusholme posted on
Great work, sounds like structuring for voice will help focus on creating the kind of clear and concise content you are already striving for.
Comment by Alex posted on
I think that quote about 50% of searches being voice by 2020 is misleading. That number was specifically talking about adoption in China and might not reflect what it will be in the UK (or elsewhere) by then. The stat originated from a guy called Andrew Ng at Baidu (in China). He works on voice recognition for a corporation so there may also be some bias there.
Comment by Sam Dub posted on
Hi Alex,
Thanks for raising that, I was a bit wary of quoting that one, it’s been cited in quite a few places attributed to ComScore but I couldn’t find the original source. We've now amended the piece to take this stat out.
Whilst they reflect current trends, we're taking these kind of predictions with a pinch of salt. What we'd really like to see is the equivalent of web analytics but for voice search and assistants – but the tools aren't quite there yet.
Comment by Dave Howard posted on
The capabilities of conversational AI (not jargone-y I hope) will also influence this for those orgs that build skills to integrate with the 'voice services' of those listed at top of this article. Increasingly orgs need an iterative way of delivering change/value; this complemented by people inside govt with tech knowledge wanting to move faster, not go thru lengthy COTS/solution based procurements but use tools they know from cloud providers to build faster not buy. Look at Newcastle City Council and others as examples. Gartner also citing in 2018 report this being a critical area of consumer (citizen) engagement but just over 1/3rd orgs looking/actively experimenting. Can we do more? Yes, and that also means considering diversity, inclusion and accessibility as point made above. Great chance to make a big difference through small incremental change if it can be allowed.
Comment by Robb posted on
One thing for sure- Voice is leaving the deaf and hard of hearing population out in the wilderness.
Comment by Sam Dub posted on
Hi Robb,
We share your worry about the potential of these assistants to exclude people with hearing or speech impairments. There's a lot more that these services need to do become truly inclusive, but there are some encouraging signs of progress.
For example, both Google Assistant and Siri have recently introduced features that allow users to type their questions then read the answers back as text. Developers have also been working on ways for assistants to understand sign language using a webcam.
Even so, as things stand we wouldn't make any information available to these assistants that wasn't also published on GOV.UK in a text format.
Comment by Geoff Simpson posted on
A very interesting post although I did have to look up what 'canonical data' is.
Comment by Jane posted on
What do you mean by 'double down on open publishing'?
Comment by Sam Dub posted on
Hi Jane,
On reflection this does sound a bit jargon-y. What we mean is that we will continue to push for all government content to be published as HTML on GOV.UK, and make that information available for reuse via public APIs.
Hope that helps!
Comment by John Ploughman posted on
Hi Sam
This is great to hear, and we’re excited about the opportunities that voice will open up.
There are still some things blocking wider adoption of HTML publishing. My personal opinion is that if GDS could help get those moving, we’d be able to get much more content into HTML.
Some particular examples that would benefit us are:
- the ability to export HTML content as different formats to enable reuse - for example, PDF and Open Document Format
- better print style sheets for HTML publications
- ‘Previous’ and ‘Next’ navigation on the manuals format
- the ability to print a whole manual with one click, rather than having to do it section by section
- improved search within manuals to highlight keyword matches
I know these would make it much easier to ‘sell’ using HTML to subject matter experts who are nervous about moving away from tried and tested older formats.
Thanks
John
Comment by Sam Dub posted on
Hi John,
We're as keen as you are to enable more HTML publishing across government. Work on improvements to our HTML formats is planned, I'll capture your suggestions and pass them on to the team that picks this up.
Sam
Comment by Louise Brazier posted on
Brilliant to see GDS taking voice tech seriously. I can see it being of benefit to lots of people. Our customers are always looking for easier access to information to suit their busy life styles.
Comment by Rafal Drewnowski posted on
Great stuff! I appreciate that Gov is looking into voice interaction. Have you explored open source voice assistants like mycroft.ai ?
Comment by Costinel Lazar posted on
Seems to be easy but not very safety and I believe still the best way is online, email or by post. Anyway I think could be easy for people which don't know to use computers, tables or smartphone. Good thing to keep updating with technology.