— by Merijn Beeksma
Computers beat us in games such as Go, glasses and watches are designed to be smart, Siri cracks us up with her jokes every time, and if that is not enough – we even put our lives in the hands of machines when we make use of self-driving cars or robot surgery. It may not be surprising that this technological revolution increases the prominence of machines in our lives – computers are faster, more efficient, and in a sense smarter than us. Additionally, they don’t need coffee breaks, don’t mind 24 hour work days, are extremely patient, and don’t roll their eyes when you ask them for help (- although that might just be due to not having eyes). They might seem to be perfect colleagues. Heck – as the movie Her, in which a writer falls hopelessly in love with his operating system, shows: they may even be the perfect life partners.
Human-computer interaction is one of the key factors driving technological advances. We do not just want computers to follow instructions that are specifically tailored to be interpreted by a computer. We want to interact with them and be understood in the way that is most natural to us: with natural, human language. Language technology aims to do all kinds of useful things with human language, regardless of whether it’s been written or spoken.
Voice assistants such as Siri, Alexa, Cortana and Google Assistant have become increasingly popular and their popularity is predicted to grow even further throughout this decade. Voice assistants are typically activated when you speak their name: ‘Alexa, can you read me the news?’ will alert Alexa to listen to your message and to determine what it is you want.
Although language technology tends to work well within certain domains, for many of us it seems to be easier to recall the mistakes than the successes – not in the least because they tend to amuse us. Websites such as Damn you autocorrect! and Why Siri why? are dedicated to collecting such funny instances of ‘fails’, and they are popular.
Suzan Verberne, information retrieval en text mining expert at Leiden University, explains that language technology is useful nonetheless. “It is important to remember that although computers certainly are not perfect, neither are humans.” She explains that language technology applications are trained with human data as examples to learn from. “In our field, we call such data the ‘ground truth’, but this is a questionable term, because when you give people a certain task, people rarely react exactly in the same way. Nevertheless, we expect computers to learn from such behavior and imitate a human perfectly.”
People rarely react exactly in the same way. Nevertheless, we expect computers to learn from such behavior and imitate a human perfectly.
Even in human-to-human interaction we misinterpret each other sometimes. This can have many causes: maybe the other person didn’t hear us correctly, or thought we meant something else. However, no matter how sloppy our utterances are, most of the time humans understand each other perfectly, and if not, they are able to correct miscommunications incredibly fast. “It are irony, subtleties and non-literal use of language that cause most of the problems in automatic language processing. These things are apparently very important to humans”, Suzan explains. The main reason that humans understand each other with more ease than computers understand us, is that we are able to use the contexts of our interactions to our advantage. Because we possess all kinds of knowledge about the world and about each other, we know which interpretation of a sentence is most likely to be correct. Computers however do not have such knowledge of the world, they have knowledge of small domains such as ‘music’ or ‘operating a car’ at best. Therefore, they have a hard time estimating which reactions would be considered logical by humans, and which ones would be regarded as highly unusual or totally weird.
Computers have other qualities, however. The value of language technology is most apparent when we try to find specific information within overwhelming heaps of data, Suzan explains. “We spend a lot of time Googling stuff for our own interest or gathering information for our job. Without the help of smart algorithms, we would not be efficient in doing so. Specialized databases like libraries, legal archives or medical records for example, are easily accessible and searchable thanks to technologies such as machine translation, text classification, and summarization”.
The importance of interaction between man and machine should not be overlooked, especially when life-altering decisions need to be made. Computers can help in deciding who to hire for a job, or which medical decisions to make for example, but instead of simply accepting the computer’s results, they should be judged by the user. “Automatic processing of data can help to rank candidates or predict the right moment for a medical intervention, but ultimately, a human makes the decision”, Suzan explains. “Computers support people, rather than replace them. We should focus on finding the optimal balance between utilizing the qualities of computers and humans.”
Computers are able to make extremely fast calculations. We are creative. Computers are great at making objective decisions. We empathize with and understand emotions of others in the snap of a second. Computers are excellent at following instructions . We have purpose and passion. Rather than trying to compete with humans when developing smart software, and determining whether technology is successful based on how well humans perform in the same task, we should recognize the differences and fully exploit them. Only then can technology truly advance us.