Localization and Conversation Design
By: Mary Tomasso
What do conversation design and localization have in common, if anything at all?
According to Racheli Kremer, “localization is the godmother of all digital content professions, including UX writing, microcopy and conversational AI.”
It’s an interesting point of view that makes a lot of sense. Localization is definitely older, and they all share the same ultimate goal: Making content more accessible to people. Another, more interesting question, might be: Can a localization expert become a conversation designer?
Before answering this question, let’s start with what conversation design is, and what is not.
What is Conversation Design?
Human conversations are complex and often unpredictable: Philosophers, psychologists and sociologists have tried to explain how conversations work.
Conversation has been theorized and studied, and it’s probably never been so relevant as it is now, due to the great amount of conversational interfaces available.
Knowing how conversations unfold between humans is paramount. One of the principles that we use the most in conversation design is Grice’s Cooperative Principle. This principle and the relative four maxims of conversations (i.e. the maxim of Quality, of Quantity, of Relevance and of Manner) help us create better experiences and minimize the errors. They are relevant, but not enough to create a flawless conversational flow.
Some might argue that conversations cannot be designed. This is definitely true for human conversations that happen naturally, but the conversation with devices can and should be designed.
We can then say that Conversation Design is the intentional creation of seamless, natural interactions between humans and machines.
Designer, Writer, Linguist, Developer or Scientist?
You might sometimes stumble upon a job description for a Conversation Designer who is actually a creative writer; some other times, the position is for a developer. As the field will evolve, things will be clearer for both companies and candidates; the requirements will hopefully standardize, and there will be less confusion.
However, it seems possible that the Conversation Designer and the other roles involved in conversational AI will be more and more multifaceted and less specialized. The trends seem to suggest that a combination of both humanistic and technological background will be favored.
What about Localization?
Localization experts are so used to making interactive elements more easily accessible to humans, that they can easily understand how conversation design works. They also have a plus: They already think in terms of "multilanguage", which not all conversation designers do.
Sometimes, a conversational project is meant to be local, and it doesn’t make sense to have it in another language. On the other hand, one may argue that there’s nothing really inherently local, because if you have a chatbot for pizza delivery in Italy, written in Italian:
It doesn’t harm to have it also in another language, to include the expats community. The text can be localized, and the chatbot can be for a pizza delivery in the US.
Both arguments are acceptable. In any case, the localization aspect is something that must be considered and thought through in the very first phase of the conversation design, i.e. when pondering the requirements.
What Are the Main Steps in Conversation Design?
In Conversation Design we ask ourselves a lot of questions. It’s not an exact science, so we always need feedback from the users, and we often need to pivot. AI learns from the interactions, and so do we: We include new ways of saying things to train AI.
We ask ourselves what channel is being used to begin the conversation; whether the first message is clear enough; if the user is informed since the beginning on the capabilities and constraints of the chatbot/voicebot.
We wonder whether curiosity or interest are stimulated, and if not, how we can improve the experience so that the user is interested in having a conversation with the bot. In other words, we try to put ourselves in the user’s shoes.
However, it’s not all questions, and it’s not all guessing: We start with gathering data to know where we stand and where we can go.
There’s a process to follow, and over time designers come to personalize that process and adapt it to their experiences, their clients and their requirements.
Google has shared an overview of the conversation design process, up to today probably the most renown.
To keep it really short:
- In the first phase you get to know your audience through data (data is crucial); you identify use cases for your project, and you design a bot persona.
- In the second phase you create sample dialogs and “high level flows”, also called “the happy path”, basically the conversational flow where nothing goes wrong. This is more of a script, where you create the conversation imagining what the users could say.
- In the third phase you start testing your flow and designing for “the long tail”, which means all the possible scenarios that might go wrong.
The process continues, you never really stop: You iterate, reiterate, and the more interactions the bot has, the more input it gets, and consequently the need to adjust or reiterate continues.
Localization of Conversational UIs
The localization of Conversational User Interfaces (CUI) is quite challenging, as factors other than language skills might determine its success. For instance, the natural language processing (NLP) support that the chatbot or voicebot uses, what languages it supports, how “trained” the AI is.
There’s a lack of relevant training data outside of English: The most common frameworks are essentially monolingual and built around English or Chinese, with some other major languages.
The same applies to NLP tools: There’s a lack of availability in many languages. The worst part is that to work around this limitation, some developers use machine translation, resulting in disastrous outputs.
Fortunately it’s not always the case and language skills are still at the heart of the conversational design, when building both monolingual and multilingual chatbots. I’m seeing more and more multilingual projects, and some companies are actively looking for “Localization UX or CUI Writers”.
These are some of the required skills:
- Being able to communicate fluently in two languages
- Experience in content, technical, or product writing
- Experience in writing and localizing content
- Ability to translate complex information into user-friendly text.
- Experience localizing conversational content for chatbots and/or voice flows
- Understanding of product design process, especially as it relates to conversational experiences
- Familiarity with NLP technology
And some of the tasks to be performed:
- Localize conversational text from one language to another.
- Manage the localization of conversational flows in the second language informed by market-specific research.
- Inform design decisions for conversational flows based on language requirements.
- Analyze content suitability and user satisfaction.
So we can say that localizing a conversational interface requires a mix of localization, technical writing, UX design and marketing research, with some bits of psychology, or at least, understanding of how humans communicate.
As a former translator and now conversation designer, I cannot stress enough the importance of involving localization from the beginning of the design phase.
If it is clear from the beginning that the project is going to be multilingual, having a localization expert even before choosing a name and creating a personality for the bot, may prevent unpleasant accidents from happening.
For example, some time ago I took part in a project where we were going to call a bot Zora, which is a beautiful female name of Slavic origin meaning dawn. I knew something sounded strange, and then I realized that by just adding an R, the word becomes a terrible offense for women in Spanish.
This can be easily avoided if localization is part of the conversation design.
As a localization expert, you have great opportunities to enter the Conversational AI space. If I could give one suggestion only, I would encourage you to learn more about NLU and/or NLP. These disciplines are probably the missing piece for language professionals, (be it translation, localization, or linguistics) to land interesting jobs or projects in conversational AI.