Why computer learning human language so difficult to do the devil inside detail Bonn/Berlin – I did not understand you, please repeat your input”so the response of a robotic gesturing greengrocer on the request of a customer after three apples in a TV commercial. This satire about the automatic speech recognition and its pitfalls raises the viewer engineer and opened a complicated and double-edged box closer. To speech communication actually provide greater comfort for the people at the man-machine, it suffers in practice often shortcomings. Some companies fear therefore still a loss of image through the use of voice applications. A technology that was not perfect, could hurt customer confidence. However, this correlation is not far from unique. A study of the Fraunhofer Institute shows so that companies use the voice application, as innovative and are perceived professionally. Individual applications, like about a timetable, work also now reliable and stable.
Because here the computer on a limited vocabulary of the operator can be trained. But worldwide, computer scientists, engineers, Phoneticians and linguists to greater research. The computer should can detect not only the language of the people, but also its content to understand. Modern systems consider the user as a partner and allow natural language dialogs in the highest quality. That means businesses a high customer satisfaction”, so the experience of Lupo Pape, Managing Director of the Berlin company SemanticEdge. In 1993, the Federal Ministry for education and research began the project Verbmobil.
The system works in dialog situations on the phone as an automatic interpreter. It recognizes spoken spontaneous speech, translated into a foreign language and sends them. The so-called homophones are a stumbling block. These are words with different meaning, the However sound exactly the same sound as sea”and more”. What is self-evident and distinguishable from the respective purchase out for us, is a great challenge for speech recognition systems. Nothing seems as familiar as the voice. But if we look closer at their use, shows that it is full of problems and paradoxes. What distinguishes the voice from the sea of sounds and noises, what makes them something special in the infinite series of acoustic phenomena, their inner relation to the importance of “the author Mladen Dolar explains in his book his master BBs voice. A theory of the voice”. “The system must therefore take into account the substantive context and understand that the rate I want on the more” makes no sense. “Otherwise, translations such as I would want to go to the more” come about. Equally important is the correct detection of accents. I resulting in the record would like to bypass the tree”by the mere shifting of accent a significance difference. To want We dodge the tree or take it over? Intuitive distinguishable for us, for the machine during the Verstehensprozesses but the question of the distinction of a tiny physical nuance. Natural language recognition, a system can also deal with a variety of pronunciation variants. For man as a producer of language is usually too lazy to articulate every syllable of a set correctly. “” So the sentence can what we have today for a date? “are articulated so reduced, that at the end of that because today hammer?” remains. The devil is in the details, so and lets cast out only with more research time. And in view of these problems and the technological advances it is language dealing with computers until then patient exercise forbearance. Patrick Schroeder
Comments