Researchers at Carnegie Mellon University have developed an automated method to make communication more polite. In a busy time, when a pandemic is raging, politicians are fighting for votes, and protesters are demanding racial justice, how people communicate with each other is becoming more important. Scientists are sure that a little politeness will not hurt. Researchers at Carnegie Mellon University have now developed an automated method to make communication more polite.
The method accepts non-political directives or requests – those that use impolite or neutral language – and restructures them or adds words to make them more educated. “Send me the data”, for example, may turn into “Could you send me the data?”
Researchers will present their research on computer politeness at the annual meeting of the Association of Computer Linguistics, which will be held almost July 5.
The idea of conveying a style or mood from one message to another — for example, turning negative statements into positive ones — is something that language technologists have been doing for a long time. Shrimai Prabhumoye, Ph.D. student at the Institute of Language Technologies (LTI), said that fulfilling the task of transmitting polite tones has long been a technology goal.
“This is extremely important for some applications, for example, if you want your emails or chatbots to sound more polite, or if you are writing a blog. But we could never find the right data to complete this task”.
Shrimai Prabhumoye, a Ph.D. student at the Institute of Language Technology (LTI)
A group of scientists solved this problem by generating a data set of 1.39 million sentences, labeled “for courtesy,” and they used it in their experiments.
The source of these suggestions may seem surprising. They were obtained from emails exchanged between employees of Enron, a Texas-based energy company that was known for corporate fraud and corruption before closing in 2001. But half a million corporate emails became publicly available as a result of lawsuits involving the Enron fraudulent scandal and were subsequently used as a data set for various research projects.
But even with the data set, the researchers simply had to determine politeness. It’s not just about using words like “please” and “thank you”. Sometimes this means making the language a little less direct. Therefore, instead of saying “you have to make X”, the sentence becomes something like “let’s make X”.
The courtesy dataset was analyzed to determine the frequency and distribution of words in polite sentences. The team then developed a conveyor belt for politeness. At first, impolite words or phrases are marked, and then the text generator replaces each marked element. The system takes care not to change the meaning of the sentence.