How can a computer learn to understand human language? Natural language processing (NLP) expert Maria Niessen from the Volkswagen Data:Lab can explain.
“Hello. If you have a question about apples, say ‘apples’”, the taped voice on the telephone hotline explains. You dutifully answer “apples” but the machine replies, cheerfully: “You said ‘pears’”, and redirects you to the nightmare of being on perma hold. This kind of thing used to happen, but is now history. That’s because nowadays machines are getting better and better at “understanding” what people want from them. Intelligent systems that understand human language are increasingly important within the Volkswagen Group, both for designing innovative products and for optimising company processes.
â€œExplainingâ€? human language to machines
Maria is a natural language processing (NLP) expert at the Volkswagen Data:Lab in Munich, the Group’s centre of competence for artificial intelligence (AI). “My job? I translate between humans and computers”, she says with a smile. Originally from Groningen, in the Netherlands, she has been working at the Data:Lab for two years, in a team whose focus is text comprehension in programming.
Easy for humans, difficult for machines
Usually, things that people can do very easily and intuitively – at least in their native language – are difficult for machines. And vice versa: computers solve differential equations in an instant, while humans find them much harder. But why is natural language a problem for machines, apart from grammar and vocabulary? “We, the human users, want the machine to solve a problem or fulfil a wish. The magic word is context. Within it, the program tries to extrapolate the meaning of the spoken or written word”, explains Maria Niessen.
Texts and words
Maria works predominantly with texts – they are easier to analyse than the spoken word, and are not complicated by inflexion, dialect or speed of speech. Nonetheless, the programs have a great deal to learn. To this end, they are fed with data, which are the foundations from which they learn, and the team correct them where necessary. This method is known as “machine learning”. The team researches systems which are not only able to recognise and classify language but also to identify and carry out the writer or speaker’s wishes. “This is of interest for many areas: from automated telephone support services to voice control systems in the vehicle”, Niessen adds.
A precious resource
Requests for this service arrive from the entire Volkswagen Group, worldwide. “The exciting thing about our job is that we work with actual problems, and solve them too”. Whether it be a chatbot for internal orders, text analysis tools for IT, telephone services or user interfaces, the Volkswagen Data:Lab team help their colleagues by training machines to understand human language better. Here, too, it is the context which makes the difference: “We make state-of-the-art solutions here”, says Niessen, proudly. “All the colleagues are extremely competent and pleasant”. A scientific approach combined with the possibilities of a large corporation – it’s a researcher’s dream.
Source: TOGETHER.net – Volkswagen AG