Chatbots with Personality NGUYEN (H.), MORALES (D.), CHIN (T.), A Neural Chatbot with Personality, 2017

How can a historical figure be accurately represented by a voicebot?

In order for the voicebot to work successfully, it needs to accurately represent a historic person.  

According to the research paper “Evaluating Quality of Chatbots and Intelligent Conversational Agents”, there are a number of attributes that determine the quality of a chatbot. This list of properties was created by summarizing all relevant previous works about chatbots, making this an extensive list. The full list of quality attributes for a chatbot can be found in the appendix. (Radziwill, Benton, 2017)

Two of the features mentioned on this list, are detrimental to creating a voicebot that depicts a historical figure.

First and foremost, the voicebot needs to be knowledgeable. It needs to be able to answer specific questions about the historic person and maintain a conversation about a relevant topic. (Radziwill, Benton, 2017)

In 2013, two researchers attempted to do just this. By purely using data gathered about a certain historic figure from websites and other plain text sources, they designed a chatbot that could accurately answer questions and hold a themed discussion. They accomplished this by using fact extraction software to turn the written text into facts, which were in turn added to the chatbot.

Whilst this chatbot was very knowledgeable, it only presented facts in a very monotone and matter-of-fact way. It had all the functionalities, but no charm or personality. (Haller, Rebedea, 2013)

Coincidentally, the second crucial property of the voicebot is that it must have its own personality and be able to have convincing, natural interactions with users. (Radziwill, Benton, 2017)

Generally speaking, there are two ways a bot can be given a personality. Either it can be created automatically by mimicking an existing person, or a personality can be created manually, using a framework to guide the process.

In 2017, it was attempted to generate four chatbots with a personality using neural networks. They were made to represent characters from popular TV shows: Barney from How I Met Your Mother, Sheldon from The Big Bang Theory, Michael from The Office, and Joey from Friends.
To do this, scripts from these four shows were used to train a machine learning model. The data from the TV shows, which was about 50.000 pairs of dialog and response per show, turned out not to be enough to create fully functional chatbots. This meant that the chatbots couldn’t take the context or the previous parts of the conversation into account when responding.

However, they were able to create four bots that respond to the last user input, each in their very own way, very clearly in the style of the TV character they mirror. (Nguyen, Morales, Chin, 2017)

Another way of giving a chatbot a personality is by manually creating one. Smestad designed and tested a framework to create chatbot personalities based on four components: the brand the chatbot represents, the needs of the users, the role of the chatbot and an appropriate personality model.
Whilst manually crafting a personality for a chatbot requires a lot of work and writing skills, the efforts do seem to pay off. According to Smestads research, a chatbot that has a personality significantly improves both the user experience and the overall perception of the brand, compared to a chatbot that provides the same functionalities but has no personality. (Smestad, 2018)

To conclude, in order for a historical figure to be accurately represented by a voicebot, the bot needs to have enough data about the person to answer specific questions and maintain a conversation, whilst also conveying the personality of the person it represents.

03/12/2018

Sources:

HALLER (E.), REBEDEA (T.), Designing a Chat-bot that Simulates an Historical Figure, 2013, https://www.researchgate.net/publication/251895907_Designing_a_Chat-bot_that_Simulates_an_Historical_Figure Date of reference: 12th of October 2018.

NGUYEN (H.), MORALES (D.), CHIN (T.), A Neural Chatbot with Personality, 2017, https://web.stanford.edu/class/cs224n/reports/2761115.pdf Date of reference: 26th of November 2018.

RADZIWILL (N.), BENTON (M.), Evaluating Quality of Chatbots and Intelligent Conversational Agents, 2017, https://arxiv.org/ftp/arxiv/papers/1704/1704.04579.pdf Date of reference: 3rd of November 2018.

SMESTAD (T. L.), Personality Matters! Improving The User Experience of Chatbot Interfaces, 2018, https://brage.bibsys.no/xmlui/handle/11250/2502575 Date of reference: 26th of November 2018.

Image via:
NGUYEN (H.), MORALES (D.), CHIN (T.), A Neural Chatbot with Personality, 2017, https://web.stanford.edu/class/cs224n/reports/2761115.pdf Date of reference: 26th of November 2018.