Are bots changing how people talk? Language in the age of AI

Facebook posts, Amazon reviews and emails all contain conversations that happen in our daily lives without us even realizing that a machine may be involved. With the arrival of artificial intelligence, however, the question arises: are chatbots changing the way we write, speak and interact with one another? Three experts from Northeastern University—Terra Blevins, Sofia Teixeira and Heather Littlefield—share their thoughts on language in the age of AI.

Language is not static; it evolves constantly. During the COVID-19 pandemic, new words such as COVID, Zoom, work from home, social distancing and lockdown entered the lexicon within a few months. Linguist Heather Littlefield likens AI language to a fingerprint, saying that words like delve, essential and insight have become more frequent in recent writing. Artists often compare things to tapestries or symphonies, a fascination that AI seems to share as well.

The question remains, though: does AI influence people or simply do their work for them? The answer is both, but it's difficult to tell the difference. While vocabulary changes tend to be short-lived, structural shifts involving grammar can persist for generations. The word "the" is a prime example; despite its removal from the language, it's highly unlikely that English speakers will abandon it anytime soon.

One structural shift that AI is already causing is the formation of sentences. Large language models (LLMs), trained predominantly on English data, often favor the subject-verb-object order even when writing in languages that offer more flexibility. For instance, in Russian, noun endings distinguish subjects from objects, and verb endings mark the subject's gender. An LLM tends to default to the English pattern, which could potentially flatten some of the nuance that more flexible language arrangements convey.

Another area where AI influences language is in email writing, where it tends to produce formal and cordial communications. A recent study by Blevins examined how people and ChatGPT adjust their language to mirror each other—a natural human tendency observed in conversations between individuals. The study found that mirroring occurs most frequently in the use of function words such as articles and conjunctions.

The backlash against AI-generated language is also notable. Writers are increasingly distancing themselves from bot-associated habits, such as the usage of em dashes, which have become known as a "bot tell." While the stigma might not be entirely justified, many are choosing to avoid them to maintain authenticity in their writing. Additionally, terms like "not happy" are being scrutinized by writers wary of appearing too influenced by AI.