How do you ensure your chatbot speaks in its native tongue?
A chatbot can be a powerful touchpoint through which a brand communicates to its customers and shapes its image. Getting its colloquial voice right is more important than you might first think, as Just AI's Svetlana Volskaya and Nastasya Savina explain.
There are at least 700+ words interpreted slightly (or completely) differently just between American and British versions of English.
If we consider other global versions and vast variety of word combinations, the task of adopting a chatbot to its appropriate language becomes quite challenging. Furthermore, a brand, catering to a specific subculture, should adopt an appropriate tone of voice. “Hey, dude!” as a greeting might work perfectly for surfers, but not for CFOs.
The goal of adopting a chatbot to the language of your customers can be split into two major tasks. The first and essential one is ensuring that the chatbot’s version of a specific language is the one used by your target audience. Chatbot developers might need to go extra mile to teach it that “I’m not so sure about that” means “definitely no” or “I suppose so” means “yes”. Or that “brew” most likely would mean beer or coffee in American, rather than “tea”.
Looking back into history, as chronicled by Winston Churchill, the opposite meanings of the verb “to table” created a misunderstanding during a meeting of the Allied forces. In British English “to table an item on an agenda” means to open it up for discussion whereas in American English, it means “to remove it from discussion”, or at times, “to suspend or delay discussion”.
Once your chatbot uses the appropriate language version to be understood properly, you can approach the second task, which is more delicate, demanding and exciting. Now you can fine-tune its terminology, style and tone to the habits and preferences of your audience.
Only then you can create a chatbot that inspires the wow effect for being “on the same wave” with your clients. We would suggest seven technological and organisations tips and tricks, that might help you achieve this challenging goal.
1. Study the language of your region / subculture.
It’s always a good idea to start with the research conducted by someone else. So, if you can find articles or books, outlining the specific language used by your target audience - use this data to train your bot.
2. Analyse the logs.
If your company is a lucky owner of a bot or a live chat, then you record histories of client conversations (i.e. “logs”). These logs are invaluable to train your chatbot. Using a robust NLU engine, like Just AI Conversational Platform (JAICP), you can identify default words and phrases that define key “intents” as well as common words, terms and idioms, popular with your audience. For example, a greeting can be conveyed by the variety of expressions ranging from “hi”, “hello”, “good day”, “good afternoon” or “good evening” to teens’ slang like “hiya” or “yo”.
Moreover, just automatic analysis without chatbot development can provide brand managers with valuable insights into customers' language, eloquence, the tempo of the conversation. Knowledge of the customers' language might help understand their Persona, traits and preferences.
3. Create a glossary.
Just like any other language tool, a bot can benefit greatly from creating a brand-specific glossary. This glossary would be beneficial to ensure consistency for a bot development team, as well as marketing and communications. “Candidates” for glossary entries could be discovered during the log analysis stage.
4. Include representatives of your target audience into the development team.
One should never underestimate the importance of local knowledge. If your developer or a copywriter represents your clientele – ask them to read and comment bot scenarios. If your bot is talking to mathematicians, Scotsmen, mommies, or gamers – such people in your team will make sure your bot will not spill out something weird or inappropriate. “Natives” in your project team could become a filter not just for the language, but for the “cultural” mishaps too.
It is particularly important because a good bot includes a “chatter”. It is a component responsible for all necessary polite rituals and “oiling” the conversation to make it run more smoothly and naturally for a user. This component represents the best opportunity to “nail it” or “miss it” in terms of tone.
5. Make the brand manager one of the project’s internal customers.
A good brand manager lives and breathes the values and the language of a brand. She understands her audience better than they know themselves. The presence of such a person in the project team can prove to be very helpful to define objectives that need to be achieved by language training.
6. Test the tone of a bot as well its functionality.
Testing for such a fuzzy feature as “an appropriate tone” can prove tricky. A brand manager could be the best person to define the criteria and desired KPIs. As for testing methodology, bot developers can employ the same approach as GUI testers – track users’ behaviour (timing events in a chat, mouse moves, facial expressions, etc.) and collect verbal feedback.
In addition to that it is suggested to roll out the bot gradually from the “inner circle” to the broader public:
- First test the bot by the project group. Presence of a “native” in the team proves to be particularly useful.
- Test by the focus-group or loyal customers. You will get necessary feedback without the risk of significant damage to the brand.
- Test by the broad public.
By the moment the bot reaches the public, the worst mistakes would be noticed and resolved.
7. Collect feedback and update the bot.
It is possible to program the bot to collect both quantitative and qualitative feedback. For example, after the end of a conversation, ask users to assess bot’s performance in marks or stars. Alternatively, you can ask users if the bot was friendly, polite, helpful, appropriate, informative, cool etc.
This feedback is invaluable for the bot’s update. Just like a website, a bot is a live tool constantly adapting to the changing business needs.
The task of fine-tuning language or a tone of a bot might seem excessive now. But it is easy to imagine that it will become a norm in the nearest future. Maybe then the final frontier in bot development would be the dream of Natalia Piaggio, a Global Head of Customer Success in ICIS: “I'm thinking of a bot that can better connect with people by understanding their culture, background, and personal preferences on the fly and adapting its language and approach to individual users.”
This article was co-authored by Just AI's Svetlana Volskaya and Nastasya Savina