Face-to-face conversations are the first and most natural form of communication for humans – and they can also be extraordinarily complex, ranging from rapidly changing topics to references or phrases that can only be properly understood by talking. referring to shared cultures or contexts.
The complexity of spoken and written conversations presents major challenges for artificial intelligence (AI) models that mimic and participate in conversations, from digital assistants like Amazon’s Alexa and Apple’s Siri to chatbots used on retail websites. retail and trade, and much more.
“Dialogue and conversation is the most basic method of communication,” said PhD candidate Davan Harrison. in Natural Language Processing (NLP) at the Baskin School of Engineering. “Being able to achieve this level of comfort and ease of communication in areas where it currently does not exist can potentially have significant effects wherever people communicate, especially where there is some sort of educational or skill barrier to communication with an entity or organization.”
Harrison and Wen Cui, both with Ph.D. students in the Natural Language and Dialogue Systems Lab, led by computer engineering professor Marilyn Walker, are working to address some of these AI challenges in their research. Their work is supported by grants from LivePersonwhich creates conversational AI experiences for a variety of industries and uses and has been recognized as one of the best innovators in AI.
The two students are working to improve the AI for open domain dialogues, which are conversations not specific to a particular industry or domain. These types of dialogues are common with devices like Alexa and Siri, and underpin other technologies like chatbots that can be used in a wide variety of different industries, from retail to mobile banking.
Students conducting NLP research as well as students enrolled in UCSC’s professional NLP master’s program benefit from collaborating with industry partners and mentors who can help them better understand the wants and needs of companies likely to hire them. For example, last year, Harrison and Cui collaborated directly with master’s NLP students on a UCSC team competing in the Alexa Prize Contest by Amazon focused on open domain dialogue.
“The study of open-domain dialogue presents unique challenges and opportunities for our NLP students to create models that speak like humans,” said Marilyn Walker, NLP Program Director and Professor of Computer Science and Engineering. . “This is an area that is becoming increasingly relevant in today’s industry. LivePerson builds conversational systems that perform nearly one billion conversations per month. Our collaboration with LivePerson will help us better understand the research needed to make these systems more natural and useful. My students’ work in these areas will shape how people interact with products that can make the world more accessible and easier to navigate. »
The two doctoral students. student research focuses on different aspects of AI algorithms designed for open-domain dialogue; Cui focuses on named entity binding, while Harrison studies dialog management.
Cui’s work aims to develop a better entity linking system, connecting entities like “Lebron James” or “the Earth” to their various meanings in an existing knowledge database – in this case, Wikidata with its more than 97 million open source data items. This can improve the performance of dialog systems in how they understand human speech and generate better responses in open-domain dialogs, where conversations can change topics quickly and often revolve around popular entities such as movies. recent or new songs. So far, his work has proven that feeding the AI learning model with semantic information representing the context of the conversation improves the performance of this model.
Cui works with a team at LivePerson that annotates conversational data to learn how people use entities in their spoken or written conversations. By incorporating annotated data from LivePerson, it aims to better understand the role of discourse patterns in entity linking for dialogue and to test various algorithms.
Cui said she hopes to bring the model she trained to life and use it for interactions with real people, to learn from how they talk to her. Testing the system with real data and users can help researchers assess its strengths and weaknesses.
Harrison’s work focuses on dialogue managers, an aspect of AI dialogue models that control the flow of a conversation, in more casual chat-style settings.
Currently, he is focused on integrating formal linguistics principles of conversational flow, called discourse relations, into AI models to better guide open-domain dialogue. Annotating logical relationships in a conversation could help machines better navigate conversations that bounce from topic to topic.
Harrison says collaborating with industry researchers allows for a productive exchange of ideas. Companies often work on similar problems, but may have a different perspective because they have access to different information and understand the unique contours of the particular problem they are working on.
“There could be a situation where the conversation can be really helpful and beneficial to someone that I’m not aware of, but a company is, because it’s their bread and butter,” Harrison said. . “There are so many open issues on the pitch.”
As the two students continue to advance their research, future collaboration with LivePerson and other industry partners could embed their work into the products that many people interact with in their daily lives.
“I think it’s really difficult for cutting-edge research to turn into commercial products,” said Beth Ann Hockey, senior data scientist at LivePerson and a member of the NLP program’s industry advisory board. “LivePerson is very keen to move the field forward, to be more innovative, and to pursue these ideas to make the conversation more natural and responsive. I think the collaboration with UC Santa Cruz is really great because maybe we can transform some of that advanced research into real products.