For all their brainpower, machines are still surprisingly tangled and awkward in the art of conversation, especially when it comes to answering questions. Computers are certainly not dumb. However, if you’ve ever talked to a chatbot or a virtual assistant, you’ll be forgiven if you ever think that way.
To overcome this weakness, computer engineers at the University of Maryland are feeding machine learning algorithm questions designed to address their challenges, in the hope that they will become more trained in communicating with people using language. The study compiled a collection of more than 1,200 questions that, despite being relatively easy for people to answer, confuse the best computer answering systems, Published in the Journal of the Association of Computational Linguistics. Here are some selected examples of the Trivia question:
- “Name this river that determines the boundary between Texas and the four Mexican states.”
- “Name this European nation after World War II that was divided into East and West.”
- “Decorate this metal that is used for decorative covers and many musical instruments.”
- “Name this kind of symbiotic relationship by a tea pirate that sucks blood from its host.”
- “Name this scene painted by Leonardo Da Vinci where Christ indicates that Judas will capture him.”
- “Identify these pieces of protective gear that, contrary to popular belief, were worn by samurai more than Vikings.”
- “Name this South African leader who became president in 1994 after 29 years in prison.”
- “Name this photographer who set many black-and-white landscapes in the American Southwest.”
- “Name this current Israeli prime minister, who observed Operation Protective Edge in early 2014.”
Computers break down and answer questions using very different methods to humans. There are more reasons for language than knowledge. It is noteworthy that some of the questions are strangely described. This is because they are
involved in six different language phenomena that consistently stump computers, but do not incline people periodically.
Alternatively, the question may require some reasoning skills, such as for the clues in the question that require a mental triangulation of elements in a question, or multiple steps must be put together to form a conclusion. These strategies include unpredictable contexts, such as the mention of a political
figure, such as the display of formula about something related to politics. Computers may be somehow confused or “confused” in this additional context, but it can probably give birth to a valuable thought in the human brain.
“People are able to generalize more and see deeper connections. They don’t have unlimited computer memory, but they still have the advantage of being able to see the forest for the trees, “explained Jordan Boyd-Graber, an associate professor of computer science at UMD and senior author of the paper.” Listing the problems that computers need to solve helps us understand the problems we need so that we can actually start seeing computers through trees and how people can answer questions.”