Welcome to a new column at 3PO-Labs, which we're (for now) calling VUXcellence. The idea behind this series is to shine a light on interesting aspects of voice UX that different developers are utilizing that make their voice interfaces more intuitive, efficient, or natural. First up, we're gonna take a look at TsaTsaTzu's massive RPG "Six Swords".
The folks at TsaTsaTzu have an extensive catalog of interesting skills, and we'll certainly delve into their history for more examples, but for our first article in this series we wanted to look at this one simple trick they implemented in their most recent game skill, Six Swords.
Sometimes Less is More
One of the main challenges of voice user interface is that it is far more linear in nature than a graphical user interface. Audio will be relayed to the user in a single thread, and the amount of time it takes to relate the entirety of the content depends primarily on how much content there is. Contrast this to a textual or graphical user interface, where the user has the option to skim ahead and look for the exact keywords that interest them.
To that end, the general solution for skill developers has been to condense or distill information as much as possible. All sorts of shortcuts are used to provide the most content in the fewest possible words, and nested or branching topics are represented by multi-turn conversations. For most skills, this is a fairly reasonable approach, and does the job efficiently, if not elegantly. In some cases, though, this approach is more stifling than it is useful.
Sometimes More is More...
Blaise Pascal is famously quoted as saying:
I would have written a shorter letter, but I did not have the time.
Pascal would not have been well suited as an author to today's high fantasy. This is a world dominated by effusive authors like George R.R. Martin and Brandon Sanderson, whose flowery language result in multi-volume storylines where every entry is a giant tome.
And this affliction does not seem to be confined just to the written word. The Peter Jackson Lord of the Rings films all run 3-4 hours (or 4+, if you're watching the superior Extended Editions), and fantasy video games like The Elder Scrolls are an 80+ hour commitment.
So when bringing content of this sort to a voice-only platform, developers are faced with a dilemma - do they abandon the hallmarks of the the genre and condense their storytelling, or do they swim against the current of VUX best practices and make the user sit through the verbal flood?
In the case of Six Swords, the answer is "neither"...
And this affliction does not seem to be confined just to the written word. The Peter Jackson Lord of the Rings films all run 3-4 hours (or 4+, if you're watching the superior Extended Editions), and fantasy video games like The Elder Scrolls are an 80+ hour commitment.
So when bringing content of this sort to a voice-only platform, developers are faced with a dilemma - do they abandon the hallmarks of the the genre and condense their storytelling, or do they swim against the current of VUX best practices and make the user sit through the verbal flood?
In the case of Six Swords, the answer is "neither"...
One giant beep for mankind...
One easy approach to this problem is to provide the users with the most important information - that which lets them continue the interaction - and then let them expressly ask for all of the context. It's essentially an "opt-in" model to information deluge. This is a fairly simple thing to do with a voice assistant - you read out the most important text, and then prompt the user with a yes or no question, asking if they want to hear more, or you give them the applicable options for asking clarifying questions. Essentially, you are encapsulating each iteration step of your skill into its own multi-turn dialog.
And technically, this works. The problem, though, is that users absolutely loathe being overprompted. Spitting out nearly identical prompts after every request is one of the quickest ways to trigger a user to disengage from your application. On the Alexa side, this has been a major point of contention between many developers (including 3PO-Labs) and the certification team, as their guidelines actually require a level of prompting that our research has shown users cannot stand.
The mental leap that Six Swords makes is "what if we could optimize that prompt down to almost nothing?", and the way they do that is rather ingenious. Six Swords does a little bit of training up front to teach the users that the sound of a beep is an audio equivalent of the prompt for "there is more information available". Then, as the user is navigating the conversation, if they hear a beep appended to the end of the output, they know that they have the option to ask for more. If they do not hear a beep, then they have heard everything there is to hear.
The result is that almost instantly after hearing the most important information, the user can make the choice to drop into the multi-turn dialog and get more context, or they can send a command to take a different action (like moving to a new area, continuing combat, etc). They don't have to sit and wait for a repeated phrase that they are already familiar with, and therefore they are much less likely to abandon the interaction, and much more capable of progressing further in the game.
And technically, this works. The problem, though, is that users absolutely loathe being overprompted. Spitting out nearly identical prompts after every request is one of the quickest ways to trigger a user to disengage from your application. On the Alexa side, this has been a major point of contention between many developers (including 3PO-Labs) and the certification team, as their guidelines actually require a level of prompting that our research has shown users cannot stand.
The mental leap that Six Swords makes is "what if we could optimize that prompt down to almost nothing?", and the way they do that is rather ingenious. Six Swords does a little bit of training up front to teach the users that the sound of a beep is an audio equivalent of the prompt for "there is more information available". Then, as the user is navigating the conversation, if they hear a beep appended to the end of the output, they know that they have the option to ask for more. If they do not hear a beep, then they have heard everything there is to hear.
The result is that almost instantly after hearing the most important information, the user can make the choice to drop into the multi-turn dialog and get more context, or they can send a command to take a different action (like moving to a new area, continuing combat, etc). They don't have to sit and wait for a repeated phrase that they are already familiar with, and therefore they are much less likely to abandon the interaction, and much more capable of progressing further in the game.
Taking things further
Now, the generic idea of a non-verbal audio cue is obviously not entirely novel - as I write this post there are several applications open on my machine that have their own distinct alert sounds that have specific meanings associated with them. And in that same vein, it stands to reason that we may start seeing voice assistant applications that use multiple different cues with unique meanings.
There's obviously a balancing act here, as users can only be expected to recognize so many distinct sounds and memorize their meanings, but it isn't a stretch to imagine a future where there's a sort of mimetic growth of this pattern. As users become more comfortable with voice assistants, the amount of training any individual skill would need to do for a given cue would decrease, so long as that sounds followed an established pattern of other applications that the majority of users are familiar with.
There's obviously a balancing act here, as users can only be expected to recognize so many distinct sounds and memorize their meanings, but it isn't a stretch to imagine a future where there's a sort of mimetic growth of this pattern. As users become more comfortable with voice assistants, the amount of training any individual skill would need to do for a given cue would decrease, so long as that sounds followed an established pattern of other applications that the majority of users are familiar with.
advertisement
We at 3PO-Labs are super curious to see how far this pattern can be taken, and if we have dialog heavy skills going forward, you can be sure we'll try something along these lines. We'd love to hear what you guys think about this approach - hit us up in the comments or directly via our contact page.
Also, as this is the first iteration of VUXcellence, we're looking for more candidates to spotlight going forward. If you have skills that you think are particularly clever or solve an interesting user experience problem in the voice space, please let us know (even if it's your own skill)!
And we're not married to the column name yet, so if you have better ideas than "VUXcellence", shoot us an email and get in while the gettin's good.
Also, as this is the first iteration of VUXcellence, we're looking for more candidates to spotlight going forward. If you have skills that you think are particularly clever or solve an interesting user experience problem in the voice space, please let us know (even if it's your own skill)!
And we're not married to the column name yet, so if you have better ideas than "VUXcellence", shoot us an email and get in while the gettin's good.
RSS Feed
