AI translation and language-learning apps can help people acquire vocabulary and manage basic exchanges, but they cannot by themselves deliver proficiency in another language, according to language educator Todd Hughes.
Hughes, 64, can read a Danish novel with little difficulty. Yet conversations with native Danish speakers remain far more challenging, exposing the gap between speaking smoothly and being able to understand and respond when an exchange moves beyond familiar territory.
“For me, it’s a question of skill and a question of ability over a question of flow,” Hughes said.
He distinguishes fluency, which describes the speed and ease of speech, from proficiency, a broader ability covering reading, writing, listening and speaking. A proficient speaker must be able to find the right vocabulary, follow what another person is saying and communicate accurately when the discussion changes direction.
Hughes trains language tutors at Rosetta Stone and speaks Spanish at home with his Costa Rican husband. He also teaches German, grew up hearing French from his grandmother and studies Nordic languages, while recently adding Japanese before a trip to Japan.
His own experience shows why a single label can be misleading. He considers his Spanish close to native and is comfortable speaking it, while he understands and reads French extremely well but makes mistakes when speaking. Danish presents the reverse challenge: reading is relatively easy, but listening and pronunciation are harder.
Why AI translation cannot replace human conversation
Social media has helped blur the distinction between fluency and proficiency. Videos of polyglots switching between languages, ordering food or surprising strangers can demonstrate quick recall and confident delivery, but may reveal little about how the speaker would cope with a longer, unscripted conversation.
Hughes says native speakers themselves pause, stumble and lose their train of thought, making smooth performance an unreliable test of language ability. Rather than asking how many languages someone speaks, he asks: “What can you do?”
Technology can support individual parts of the learning process. AI can create exercises, correct writing and produce reading material suited to a learner’s level. Apple’s Live Translation feature for AirPods has also been presented as a way to help people follow conversations in real time.
Hughes, who has worked with technology in language education since 1995, compares AI with a “huge library” that is always available. But he remains unconvinced that it can yet provide effective speaking practice.
“I’m not there yet,” he said.
Duolingo’s chief product officer, Cem Kansu, said real-time translation could help people understand one another in the moment, but could not replace the connection, cultural understanding and personal satisfaction of learning and speaking a language themselves.
That distinction is particularly important for people learning English for work or higher education. An accurate translation may convey the words used in one exchange, but proficiency requires being able to ask follow-up questions, continue the discussion and respond as its wording and tone change.
Language also carries social meaning that cannot always be captured by replacing one expression with another. Hughes gives the example of “don’t pull my leg”, whose Spanish counterpart translates literally as “don’t pull my hair”. An app may identify the equivalent idiom, but that does not necessarily show whether it is being used playfully or rudely, or whether it is appropriate in a particular relationship.
In a live conversation, learners must listen, choose a response, observe how it is received and adjust as the exchange develops. Hughes believes those social cues, along with cultural understanding and the ability to react in real time, remain central to proficiency.
“When was the last time that you had a really, really, really interesting conversation with a computer?” he asked.
