Regular uses of Gmail and other such services will have noticed the growing role AI is playing in the construction of our emails, with the platform even providing reminders if we may have neglected to reply to something urgent.
New research from Carnegie Mellon highlights how such automated methods could also help to make our communication more polite. The method developed by the team takes any nonpolite corporate communications, such as a directive from a manager, and restructures them so they’re more well-mannered.
It’s a process that language technologists have been mulling over for a little while now, with a desire to turn negative statements into something more positive something with clear applications.
“It is extremely relevant for some applications, such as if you want to make your emails or chatbot sound more polite or if you’re writing a blog,” the researchers explain. “But we could never find the right data to perform this task.”
Polite communications
The researchers trawled through a dataset of around 1.4 million sentences that had been labeled according to their politeness. The data was gathered from the email database at Enron, which entered the public domain after lawsuits surrounding the fraud scandal underwent by the firm. Even with such a large dataset, however, it was not easy to define politeness (and not just because of the particular nature of life at Enron!)
“It’s not just about using words such as ‘please’ and ‘thank you,’” the researchers say. “Sometimes, it means making language a bit less direct, so that instead of saying “you should do X,” the sentence becomes something like “let us do X.””
What’s more, politeness can vary from one culture to another. It’s a situation the team tried to overcome by restricting their work to speakers using North American English in a formal setting.
Structure of politeness
The data was analyzed to try and understand the differences between polite and non-polite sentences, particularly in terms of frequency and distribution of words. Work was then done to help automate the transformation of more polite sentences. This began by tagging impolite words or phrases, with a text generator then replacing each tagged item, with the system striving not to change the meaning of each sentence.
“It’s not just about cleaning up swear words,” the researchers say. “Initially, the system had a tendency to simply add words to sentences, such as “please” or “sorry.” If “Please help me” was considered polite, the system considered “Please please please help me” even more polite.”
Over time, however, it became more realistic, with the changes made becoming subtler. For instance, first person singular pronouns were replaced with first person plural pronouns. Similarly, the word “please” was inserted in the middle of a sentence rather than at the beginning, such as with “Could you please send me the document?”.
Prabhumoye says the researchers have released their labeled dataset for use by other researchers, hoping to encourage them to further study politeness.
