Problem
Established in the 1890s, Vancouver’s historic Chinatown is a popular tourist attraction visited by many local and international tourists. Despite steady year-on-year growth in Vancouver’s tourism and hospitality sector, the neighbourhood faces several socio-economic challenges, including gentrification, an ongoing opioid crisis, rising rates of people experiencing homelessness and poverty among senior residents.
Tourists “parachute” into the lives of local residents and experience the culturally rich neighbourhood from a distance – often without engaging with locals. Simultaneously, language barriers and unfamiliar geography cause tourists to get lost or not know where to find local businesses that could heighten their tourism experience.
Chinatown in the News
Approach
As a resident of Chinatown myself, supporting the neighbourhood was my primary goal. To do this, creating more conversations between visitors and residents would be necessary. I took a very literal approach to this and decided to create a conversation agent – also known as a chatbot.
Through conversation, I wanted to share stories that informed visitors of Chinatown’s rich history. Moreover, I wanted to model an everyday conscientious conversation about the neighbourhood using people-first language as part of the conversation design. People-first language emphasizes the person over any stigmatizing characteristics (i.e. being homeless or a having a disability) as a way of speaking about people with dignity – ultimately, supporting an empathetic relationship between the tourist and locals.
For these reasons, Georgia focuses on tourists as users.
Conversation Design
Natural Language Understanding (NLU) is a powerful application of machine learning often found in chatbots and digital assistants such as Alexa, Google Home and Siri. The current state of these conversation agents focus on users as consumers who look to make quick seamless purchases or resolve simple service-related problems (e.g. customer service on a product website or playing Spotify on a smart home speaker). Other agents, like Max and RHINO, have been deployed inside museums as guides with limited conversation and/or navigation capacity. These agents inadequately address the user’s need to navigate and learn in real-time in unfamiliar outdoor surroundings.
Georgia attempts to address this gap by utilizing Google’s dynamic API.AI/Diagflow Natural Language Processing (NLP) platform which allows for integrations across multiple systems and devices. Georgia centres the user as an explorer-first, stimulating their curiosity through social engagement with a local “insider”.
Process
Who’s Talking?
To design the conversation between an agent and a user, simple personas for both parties were created.
Georgia
Georgia’s persona was required to develop Georgia’s agent character.
Simply, Georgia carries a familiar service-oriented personality, one that is friendly – much like the ones you might find to help you make a flight booking. More specifically, Georgia is a friendly neighbour, helpful, knowledgeable, compassionate and personable but not carefree. And, Georgia is also named after a street in Chinatown.
User
Through research on tourism trends and motivations in Vancouver, a simple persona of a Chinatown tourist emerged. The tourist is an adult middle-class temporary visitor who speaks English. They’re curious but want to be delighted which is why they travel for cultural tourism.
Needs to:
- complete a tour of Chinatown on foot
- see the main attractions
- feel safe
- feel independent
Obstacles:
- getting lost in a new city
- doesn’t know anyone in Chinatown
Conversation design for navigation
Georgia is built on API.AI (now Diagflow) and integrated with Skype where the tour happens. Skype chat on mobile supports real-time conversations using voice, type or tap as well as images. Georgia uses all of these to communicate and navigate the user.
Georgia describes intersections or addresses of site locations (where possible). Importantly, Georgia talks to the user as a person who is unfamiliar with the area – providing images of sites, descriptions of surroundings and intersections to help users locate themselves – much like a local might give directions. Georgia asks the user to prompt it when the user arrives at each site to “begin”.
The initial prototype guides users on a linear tour route of five stops (out of a total of 11).
Conversation design for empathy
Once both conversation partners’ personas and the user’s goal has been established, designing the conversation for depth and flow relied on several states to move through. Each tour stop became a site for discussion. For this reason, the conversation is largely linear based on sites – however, a user can decide to dwell at a site or move on. Conversations at these stops start with the informational (i.e. address) and move into contextual, understanding, mirroring and relational information depending on how long a user wants to stay at a site.
Some of the conversation topics covered with varying depth include Chinese immigration and anti-Chinese laws, colonization and the contributions of Chinese settlers, significant local historical figures, Chinatown as a heritage site and the economic challenges of the area.
Early designs of the conversation depth can be seen in the “Flow of Conversation Depth” chart. I recently iterated this based on the Spectrum of Empathy – which helps to distinguish between pity, sympathy and empathy, as well as the potential positive impact described by the size of each circle. The original chart has been adapted to include an “unaware” state that seeks information but not engagement.
While Georgia’s conversation design is largely linear, fallbacks to queries like “Who are you?” and “I’m lost” support likely deviations from the tour conversation.
For the first Georgia prototype, very limited social context and examples of people-first language are used. The proof of concept first aims to move people through an unfamiliar outdoor setting, followed by some contextual and understanding of local stories.
Output
The first prototype of Georgia the neighbourly Chinatown tour guide chatbot was completed in 2017 for Skype.
As a tour guide, Georgia is free, designed to be self-paced for independent exploration.
Outcome
Since the prototype was completed, Georgia has been tested by community members as a proof of concept.
Engaging tourists in conversation and nudging them toward inclusive language was a key outcome of this project. Further testing of the chatbot will determine the navigational, temporal and empathetic goals of the project.
Reflections + Future Directions
Being a prototype there are many challenges and opportunities. As my first time using API.AI there was a steep learning curve – not to mention the NLP technology is still in its infancy. Since the prototype was built, the conversation design community has grown and more resources have become available. A recent conversation design course I took, facilitated by Google, helped me imagine future iterations of Georgia.
The next iteration of Georgia has the potential to be integrated with digital assistants including Google Assistant (rather than Skype) – or SMS. Other integration possibilities include Google Maps, weather and location-based recommendations – not to mention accessibility support.
Meaningful conversations can move users through physical spaces and states of emotional awareness. Georgia aims to engage tourists as users to a state of mirroring inclusive language in order to nudge them toward an empathetic appreciation of Vancouver’s Chinatown and its residents – heightening their tourism experience in the process.
