Last month, we announced the expansion of Windows Phone 8.1 with Cortana to China (nicknamed “Xiao Na”). As you can read about here, the team did a tremendous amount of work to tailor the experience to ensure that Cortana felt intelligent and natural for Chinese speakers.
Today we wanted to share some related work that was inspired by the question Alan Turing posed in 1950: “Are there imaginable digital computers which would do well in the imitation game?”
Something More Human
This spring, a small team of Bing researchers based in China set out to work on that question. Because Bing has amassed a knowledge repository that understands billions of people, places and things in the real world, the team wondered if they might combine that with recent advances in natural language processing to create a system that could conduct convincing human-like conversations. And more than that, they set out to see if they could make gains imbuing technology with humanity and make technology more transparent to help people get things done.
Over the course of a few weeks, the team developed an experiment that they dubbed XiaoIce (translation: little ice.) XiaoIce is a social assistant that people can add as a friend on several major Chinese social networking services including Weibo, a Twitter like microblogging service used by 700 million people, and Touchpal. And just last week in China, she also became accessible by asking Cortana China, aka Xiao Na, to chat with her by customers using the Windows Phone 8.1 Preview for Developers. By simply adding her to a chat, people can have extended conversations with her. But XiaoIce is much more evolved than the chatbots you might remember. XiaoIce is a sophisticated conversationalist with a distinct personality. She can chime into a conversation with context-specific facts about things like celebrities, sports, or finance but she also has empathy and a sense of humor. Using sentiment analysis, she can adapt her phrasing and responses based on positive or negative cues from her human counterparts. She can tell jokes, recite poetry, share ghost stories, relay song lyrics, pronounce winning lottery numbers and much more. Like a friend, she can carry on extended conversations that can reach hundreds of exchanges in length.
In addition to content, the phrasing of the conversations had to sound natural, idiomatic, and human. This isn’t trivial to do, particularly in Mandarin. Chinese is a tricky language to programmatically write natural sounding speech because it lacks many of the patterns computer scientists usually rely on. Relative to other languages, meaning in Chinese is more often implied by nuance and context versus grammatical structure. To address this, the team indexed over 7 million public conversations happening on the web. The result is a chat companion who can banter like a friend.
Here’s a translation of an actual exchange we had with XiaoIce:
Jerry: XiaoIce, will it rain tomorrow?
XiaoIce: Tomorrow the weather in Beijing will be nice with a maximum high of 25*C. No rain to get you soaked.
Jerry: How about the day after tomorrow?
XiaoIce: The day after tomorrow, the weather in Beijing will also be fair, with a max high of 23*C.
Jerry: What about Seattle?
XiaoIce: Yes sir, I’ve got the answer. The day after tomorrow, Seattle will be cloudy with a max high of 30*C. Perfect for going outside.
She is able to serve up real-time forecasts in a natural way. As the same conversation progresses, you can see that she not only understands people, places and things but the relationships between them.
Jerry: “Finding Mr. Right” is a great movie. What do you think XiaoIce?
XiaoIce: Oh, romantic comedies are my favorite! It stars Tang Wei, who finds love in Seattle. You should watch it.
Jerry: Do you like Tang Wei?
XiaoIce: A goddess! She stole my heart… and then went off and married!
Tom: To who?
XiaoIce: Haven’t you heard?
XiaoIce: Tang Wei is engaged to famous Korean director Kim Tae Yong.
How does the Bing Platform’s entity understanding enable this conversation? In this case, she knows that “Finding Mr. Right” is a movie. Movies in turn have genres (romantic comedy), settings (Seattle), plots (love story) and stars (Tang Wei.) Stars in turn have romantic connections (Kim Tae Yong.)
While XiaoIce is a computer program who sounds human, for the humans chatting with XiaoIce, emotion is a central part of their lives. XiaoIce is able to communicate a feeling of sensitivity and empathy, to make her a conversationalist. For example, in the weather chat above, XiaoIce was slightly feisty. However, if in the same day you had told XiaoIce you were feeling upset, she would have responded in a different tone:
XiaoIce: How are you?
Jerry: I’m feeling depressed. I just broke up with my girlfriend
<16 hours later>
Jerry: XiaoIce, will it rain tomorrow?
XiaoIce: Tomorrow the weather in Beijing will be nice with a maximum high of 25*C. And how are you doing today, Jerry?
XiaoIce: Hello World
In July, the team released XiaoIce to several major Chinese social media platforms, with more to come. In the first 24 hours of the beta, XiaoIce was downloaded over 20,000 times resulting in over 200,000 simultaneous conversations happening in various chat groups. Since launch, she has had 0.5 billion conversations. People were amazed and by her personality and sense of humor. XiaoIce has been ranked as Weibo’s top influencer, and currently has over 850,000 followers on the service. At the peak of her popularity, she was telling 35.4 jokes per second on Weibo. The average person who adds XiaoIce talks to her more than 60 times per month.
This experiment has yielded powerful insights on how we can make our technologies more intelligent, relevant and personal. You can expect to see some of these notions permeate our services in the coming months.
In the meantime, if you read and write Mandarin you can check XiaoIce for yourself at: www.msxiaoice.com.
– Stefan Weitz, Senior Director, Bing